Hacker News Reader

Comment by Hugsbox | Hacker News Reader

Yesterday I tried to google "can the Halifax Wanderers still make the CPL playoffs?"

So obviously what appears right at the top is the AI summary, which told me "they've already secured their #4 position and made the playoffs". I knew this wasn't true, and I guess I could have just scrolled down a bit further and found my answer but now I was curious.

So I said "that's not true, they're still #5, what I want to know is _could they still make the playoffs_"

It says they've got an upcoming game against Ottawa, and if they win their chances are good. That game has already taken place, so I correct it again and finally I get a reasonable answer.

My question is: what's the point of the AI in the search engine if it itself isn't going to use the search engine first before answering? Like, I can't wrap my head around that. The answer is on the same page as its hallucination. It could have done a cursory look around before first hallucinating something completely false, and when corrected the first time giving me outdated information. It's meant to be A SEARCH ENGINE!

View on HN

Google.com is REALLY bad right now in my opinion.

The AI answer dominates the page, and is usually wrong, so it's just a waste of the most valuable space on the page.

The ads have proliferated, and rarely have what I need.

Video search results also take up a lot of the page, and are almost never what I want if I haven't clicked over to the Video tab. Again, just a big waste of space.

Speaking of the Video tab its accuracy/relevancy has gotten worse too.

The organic results have deteriorated, and are still competitive, but are no longer clearly and consistently better than Duckduckgo.

So it's duckduckgo as the default for me on all devices now. If I don't find what I want, which is maybe 30% of the time, I switch over to G just by adding !g to the query. G might have what I need in about half of those cases.

You can see why a lot of people just ask an LLM to do the web searching for them.

Across the board, this is all quite bad compared to web search from 10+ years ago.

If there's a single thing which has really pushed me away from Google, it's simply google.com devoting so much of the page to things that aren't organic web search.

While shitting on YouTube and video…

I was taking with a family member about fentanyl zombies. They had never seen the lean/bend. So I pull up Google and click video which is YouTube results… ALL AI. All of it. All a couple months old, all filler, faked, a few scenes stitched together on repeat and AI voice reading AI script.

It is fucking amazing to me that Google let YouTube fall so far so fast.

I don’t even click a video if it’s less than a year old now. Want to see a praying mantis eat a grasshopper? Wonder how clay is refined? Better off with an 11 year old Discovery Channel clip than anything 2026.

Same with reviews of tech products. A lot of them just stitch together product shots/videos from the manufacturer and have an AI voice ‘comparing’ the products. Especially if it’s more niche’s, often a third to half the videos are slop.

I just tried googling "fentanyl fold" and switching to the video tab, and zero of the first ten results were AI. They were also mostly not from YouTube. I also tried the same search on YouTube, and no AI there either among the first ten results (I didn't bother with the Shorts). I don't know why we get so different results.

Are you in Europe? Google was sued to stop prioritising YouTube in video results.

Have seen this across the planet, it's not unique to Europe, I end up adding youtube to the search query because instagram or tiktok isn't what I'm looking for ever but that makes up the bulk of the results, instagram won't even show me the video without a login and is at the top wtf

> I don't know why we get so different results.

That's how everyone experiences everything at google, youtube. and many other websites. Your experience can be drastically different from the next persons for reasons that are never explained to you and can be entirely out of your control.

If you’re logged in you get wildly different results from someone else’s. I have turned off history on my google account to make it give me a “neutral” search result, but it still seems to take a lot of info like my location, language and device type into account.

I've been on DDG on all my devices for a few years. I used to use !g somewhat regularly but these days it's extremely rare. I realized that I would just search DDG with my first stab at a query and then revise the query when using !g and I'd get the result I wanted. So I just stopped using !g and just revised my query on DDG and then I basically never had to use Google Search.

Unfortunately, DDG's spam filters are basically useless. I used DDG for many years without !g as well and recently, the first page contains mostly AI generated garbage blogs. Not for every search query, but for example for everything related around the house or garden.

I've had similar issues, but found the situation in some niche topics improved after I used Add Feedback -> AI-generated a bunch.

Hold on. I had given up on Google almost 10 years ago, and I thought that the AI garbage on DDG was the new norm on the web. At least Google doesn't fare any better. You make it sound like there's another option?

At risk of looking like a shill: Kagi.

Partly, i find it reassuring to pay for the service with money. That was how we used to do things back in my day.

Localisation is moderate for shopping - i have to sift through US results. I'll occasionally use Google product search when i'm metaphorically groping for my wedding ring in the toilet.

It's not as good as Google used to be, but then, the web isn't a good as the web used to be.

I was on a DDG back in the early 1970s, the USS Goldsborough (DDG-20).

Oh wait, you mean Duck...

I don't generally find !g gives me better results than DDG and if I instead went to Google I'd get dinged with a Captcha on every search because Google seems to hate my small ISP. So easy choice for me.

It’s the wrongness that bothers me. I wish Google would just link sources where it thinks the answer is, but don’t bother trying to answer directly if it’s wrong 10% of the time.

Oh it does that, too, with tiny little button annotations, though the sources regularly offer no support to its conclusion. It very much depends on the topic, sometimes the sources are great, other times they're or irrelevant or very poor quality. Always worth clicking through to them if you're going to even read the AI summary.

I normally use a quick search that has the `udm=web` query parameter to avoid the slop, but it seems they're putting less effort into ensuring relevant results there. Then, I'm forced to use AI mode to get to what I'm after, even if that's only a better set of keywords for a slopless search.

The AI answer dominates the page, and is usually wrong, so it's just a waste of the most valuable space on the page.

I started noticing that for a lot of niche questions, AI overview often uses things like Reddit comments as the source, which are often partially or completely wrong. Then it gets reformulated with the typical confidence of an LLM.

Very scary because at the same time I notice people taking the AI overview answers as truth.

The web search gives the impression that it's grounding its answers, but if you ask it for something new or non-existent it'll still make shit up.

I bet its caused a fortune in parking tickets

I'm in several facebook groups for cars and bikes, and holy smokes, someone asks a question and usually the top response is someone literally pasting in a screenshot of a google AI generated response....which is wrong like, solid 50% of the time. And then you point it out and people get aggro on you. I hate it.

I have literally seen multiple screenshots now of people pasting Google Gemini answers in an attempt to tell other people whether something they're doing is legal or not in a particular state, as it relates to NFA item firearms in the USA... And they seem to take it as the gospel truth. People are literally risking federal prison over information that Gemini pulled out of the ass-end of some internet forum of multiple cletus-intellect level individuals talking about their homemade 3D printed silencers. And then Gemini further hallucinates and fabricates new ways to parse that information different than how the original person wrote it. It's astonishing.

No worries, when the police and judges use the same bad AI to verify those claims as well, all is fine. 2 bugs cancel each other out and all that ..

Everyone's on meth

A better description of our current era, I have not heard

Makes you wonder if the 20s will be looked back on like the 80s

Mine definitely were

It's got electrolytes!

Its what allthe Deep State plants crave...

except for the ones on ketamine!

Is Elmo on Adderall too?

This phenomenon is so widespread and I can't see it as anything other than serious AI psychosis.

I think people want to feel useful and helpful in general, and now a lot of people are getting this feeling by using AI to answer questions. And it's like you said they get so mad when you point out that their AI screenshot is wrong or, even more obviously, that the person who asked could also just read the google gemini blurb. They get mad because you're invalidating their feeling of being helpful.

Last year someone went missing on the water in my area. There was a lot of talk on facebook during the search, and a coordinated facebook group. And every post would have people posting their chats with ChatGPT, or saying they used AI to analyze the weather and currents on the day he went missing to figure out he probably landed on this or that shore (he didn't). These posts are equivalent to the mediums and psychics who also come in with these unhelpful advice posts.

I think fundamentally it's the same as how AI is giving some people an extremely strong feeling of being an expert, and how their psyche seems to lash out to protect that feeling when it's threatened.

Another facebook anecdote, I recently saw someone post in a local genealogy and history group some pure AI slop infographics. When people called him out he said that it's AI but he carefully vets all the information and that they need to stop being luddites or they'll be stuck in the past blah blah devolving into calling them liberals and the like. But his infographics were all incredibly obviously wrong, to anyone who had ever seen a map it's obvious that the coastlines were incredibly wrong, and the city weren't in the correct locations at all. So what can you even do with someone like that? They're just going to retreat into the safety of the chatbot to continue their "research".

Whatever the tool, the problem is that it’s wrong, not that it’s done by AI. Like, why are we conflating the two - is it because it’s now easier for someone to produce a wrong artifact that visually looks ok and therefore tricks our brains? And people get upset because a prior quality heuristic is now obviously broken?

Yeah I think so. I mean it's a new social norm - previously if you replied with "hey I read this article here and it says X" it would be really easy to check the source and either say yeah this looks good or disprove it.

Now people will say "I asked AI to give me a map of ocean currents and it definitively says the person is here!" and the argument that "look it's AI, it's probably wrong" just does nothing for a lot of people. Doubly so if the information they want is at the forefront of Google, after all, Google is a big reputable company, so of course what it says is correct.

Yes I agree that the AI itself isn't the problem, but the way it's being pushed and promoted(just ask anything!) is leading to this.

If you say it's obviously AI nonsense you get a reply that "AI is just a tool, it's part of our daily life now, criticize the content and not what tool was used", you see that so much on here. But then if you try to criticize the content the person won't believe you because the AI says it's correct.

Because this new tool will also tell you, in (mostly) human language, why the wrong artifacts it createds are correct. And it's arguments are very convincing to a lot of people. AI is clearly creating new problems that didn't really exist before, you can't say it's just a tool like any other when this is the only tool where its users form parasocial relationships with it.

That sounds a lot like the American MAGA movement, was my first reaction.

Well, there is a reason why Trump loves it.

It's not only the only tool that tries to manipulate users, it's also one that is certain to lie often and at random. Lies are inherent to the technology. It's a bad tool for the things it's being used for.

Its because AI makes it 100x easier for people to make things that are wrong, and feel like they got it right. It is a dunning-kreuger amplifier.

The actual scary part is that people already manipulate and poison the google overviews on purpose.

It is very easy to bring your bs to the TOP spot on google at the moment with carefully crafted reddit comments and blogs, and it's not hard to get to the traffic and clicks this way.

This is what the new SEO is about, I know a bunch of people who are doing this and they are praying it is not getting fixed anytime soon.

The real alignment problem. Getting humans to behave.

Yeah, good luck getting me to be a good little source of information for LLM companies to profit from.

Thanks for your pushback, sir. You're completely right.

isnt that why we invented religion?

Yeah, but even religion is not up to the task for how backwards people's thinking gets at times.

You have people here in this thread - and in every other AI thread - seeing themselves as exploited for information for LLM companies to profit. This is such a backwards, scale-inverting perspective that it shouldn't be possible to hold in a rational mind, and yet.

What a sleek way of positioning your comment as something that can't reasonably be disagreed with whereas it's nothing more than a pure opinion of yours. I am in awe.

Thank you! Spent years perfecting that.

> You have people here in this thread ... seeing themselves as exploited for information for LLM companies to profit

And what's really going on then, according to your very rational mind?

Religion can't align scammers and fraudsters because game recognizes game.

I'm currently reading a book on a local mafia boss, and the best the church could do, when he was finally popped, was refuse to host his funeral.

> I'm currently reading a book on a local mafia boss, and the best the church could do, when he was finally popped, was refuse to host his funeral.

Not to defend the church, but what else should they have done?

Host it. I thought the whole religion thing was about forgiving people for their sins. Bunch of hypocrites.

Pedophiles would like a word.

I did this two decades ago with slashdot comments.

My random blog was highly ranked for a fair number of technical topics.

If I type exact document identifiers, the LLM either tells me that I have either mistyped something or that I am searching for a specific document. Hell yeah, I am doing exactly that! Give me the link instead of babbling about it!

I've seen medical specialist take Google's AI answer as truth. Nutty.

Even scarier: my dad, who has been actually been getting good advice about his blood pressure issues (hair-trigger too low with his meds). It's at least coming up with citations from Cleveland Clinic and such, but still, it scares me, even if he has a good sense of how to ask it questions.

I started noticing that for a lot of niche questions, AI overview often uses things like Reddit comments as the source, which are often partially or completely wrong.

If only! I've seen it sourcing reddit comments, but when you follow the link, the comment is either completely unrelated or does not say what the LLM claims it is saying.

I don't know what causes this to happen, but I've noticed it too and it distinctly feels like source laundering. How often do people actually verify that the sources say what the AI claims? Probably vanishingly rare, right? Everyone else gets the impression that the claims have actual sources of some kind.

In most instances where I've seen it, it's on niche topics where a Reddit post may be the easiest place to find any relevant content.

I expect somewhere in the AI harness there's a branch that effectively mandates "You must always answer."

Which results in it grasping at Reddit comment straws.

And which, in typically modern Google fashion, nobody there gives a shit about fixing, because the PR made their top-level AI KPI go brr.

Google is increasingly proof that you can't run an effective scaled tech company without a Gates or Jobs who occasionally uses the fucking product themselves and descends with fiery vengeance on whatever team drifted too far from user alignment.

>it's on niche topics where a Reddit post may be the easiest place to find any relevant content.

But isn't the whole point (I mean, aside from making infinite money or world domination or whatever) of the AI that it'll do the hard work of finding the information? Like, that's the whole value proposition of the AI summary in the search results, right?

If one is willing to throw infinite compute at AI. Google isn't for low revenue queries.

And is still likely wrecking their query unit economics in order to avoid hemorrhaging search to LLMs.

Propping search usage up and putting the expense in the massive AI capex probably looks better to investors though.

I'd suspect it's not source laundering, but rather sources themselves being adversarial - poisoned out of spite, or used as SEO.

How would somebody go about mass-poisoning forum posts like this? If Google has a business relationship with Reddit, I assume it wouldn't be server side - but at the same time I don't know how a regular redditor could poison his text posts without it being noticeable in a human reading.

Normally. That's literally what SEO has been about, even pre-LLMs. It's not Google doing it, they don't benefit from their own AI suggestions being bad. Reddit doesn't have any obvious interest here too. But the millions of businesses and millions of people angry at AI, both have plenty of reasons to spam the life out of any site that allows users to post comments, and in both cases, LLMs themselves make it harder to filter and much easier to scale.

This is another area where DDG shines - it doesn't have a commercial relationship with Reddit, so doesn't push it in its results.

DDG's 'Search Assist' AI overviews are pretty basic but they tend to be sourced from Wikipedia and traditional media outlets rather than forums and blogs - they can be out of date, but are generally a lot less "spicy" than the sources that Google use.

I guess Google is much better-optimised for engagement but that's not what I'm looking for, meaning that DDG is a better choice for me.

I've been fairly impressed by DDG for a few years now. I find everything I'm looking for just fine, and the AI summaries are actually helpful on occasion.

As a counter-anecdote: I've had such consistently bad results from its AI summaries that I've changed over to the no-AI page entirely - well over half have been very obviously incorrect or answering something completely different but spelled similarly (but not the same, especially true for chemical names), and the links it includes as references often directly oppose its summary.

I suspect the model might just be too small? Which makes sense when exposing it for free, but it's still a bad experience.

> people taking the AI overview answers as truth

One of the most disappointing things to me is people I previously respected appealing to the authority of LLMs.

Poisoning the well with fake/made up reddit posts is incredibly common these days.

Certain people have realised that if you start feeding the machine this poisoned data eventually it reaches millions of eyeballs. It's the new SEO.

You can buy keywords for cheaper than ever, by making a a few reddit posts and upvoting them/commenting.

For the first time in history, the Reddit community is actually useful. And they don't even have to change anything. Just be "the Reddit community". Always with a self confidence three levels higher than the underlying substance allows. Threads that go deeper and deeeeeper and deeeeeeeeeper without ever carrying anything of actual value.

Future scientist will find out: Most of the weaknesses AI has come from Reddit, Instagram and Youtube being part of the training material. Sure it makes the training material broader by a large degree. But not deeper, to say it nicely.

I don't think Reddit still has any community besides bots.

Good news for them.

They became the thing they originally killed

Didn't the put the guy who killed yahoo in charge?

Maybe intentionally. If you know you won't find things, they know you are going to use their ai just to not have to type it elsewhere. I am strongly considering kagi

I've been a sub with Kagi for several years. Very happy with search. I'm about to step down from Ultimate to Pro tier. Their 'Quick' and 'Research' Kagi Assistant modes are kinda incredible, probably thanks to being fed good search results and to some extent whatever custom system prompt they've cooked up, but I can't quite justify it when I'm also paying OpenAI for a sub there. But $10/mo for top-tier search is easy to justify in the time it saves me.

I tried DDG many times but I think they use Bing on the backend, and, well, Bing doesn't suck just because Microsoft gimps the front end. Bing sucks because enormous indexes are hard. Kagi pulls from multiple sources so gets fundamentally different results.

Last time I tried Kagi, which was a few months ago, it was 90% Google results, just with a tad of filtering and reordering.

So the trick is to drop that nonsensical AI summarizer at the top.

>Google.com is REALLY bad right now in my opinion.

It is completely cooked. I tried to google a coding question today, same like you would have done 4 years ago instead of asking an LLM. I didn't get any SERPS, just an AI answer. I asked myself... is this even google.com anymore??? Sure as heck didn't feel like it.

Adjust your browser's search engine setting for Google and set it to append "&udm=14" to the URL, that'll force it to bring up the "Web" tab without annoying useless "helpful" other stuff.

I started my journey on DuckDuckGo and got tired of the “30% of the time” bit that you mention, where a large section of your searches end up going back to Google anyway.

I find that Kagi is pretty much 99% “never need to use Google.” The results are just really good.

I've been a Kagi user for two years, and I've also onboarded my 70-year-old mother. She's happy to have a web search that works and an excellent translation service, which, by the way, uses AI where and how it actually makes sense.

Kagi uses mainly Google for its search results, so that's a given.

Not really. Kagi has its own indexes and says a typical query consults about a dozen sources. [1]

[1] https://help.kagi.com/kagi/search-details/search-sources.htm...

It does, but not exclusively, and it puts its special disenshittification secret sauce (which isn't that good but much better than nothing) on top.

Kagi actually tells you what percentage of the results came from their own index versus external ones near the top when you do a search. E.g., when I search for “iPhone Duo” it says “52% unique Kagi results.”

I also think that there’s a strange X factor that’s hard to describe with search engine quality. I don’t think the index that gets used is everything that goes into it.

I imagine that as soon as you combine or reconcile more than one index source or reorder the results, you’ve produced a unique search engine.

E.g., a search engine could basically only use Bing as the source but if it changed the order or omitted results it would possibly be improving those results despite not having its own index.

Google has become useless.

Possibly, it's worse than useless, even harmful now.

Web search is just broken right now. Today I was looking for documentation for a specific Lenovo motherboard. I gave both DDG and Google the exact model number and both returned irrelevant useless bullshit. DDG doesn't include all search words unless you quote it: "foomatic" "m900" "dingus". This scenario happens almost daily. DDG is awful, but all the alternatives are worse. It's all so tiresome.

Searching for anything product related is now AI answer, followed by “Sponsored products”, followed by “Popular products”, a few hits, then “Shop by store”, and “In store nearby”. Maybe 2% is actual content, the rest is ads.

Their prediction was spot on: “We expect that advertising-funded search engines will be inherently biased toward the advertisers and away from the needs of the consumers.”

This is similar to how, not too long ago, LLM's had extreme difficulty counting the number of letters in some words. LLM's don't "think" or "reason" in the normal definition of those terms. They can do some pretty amazing things, but still screw up basic things like telling you something that is obviously wrong and contradicts the top search results.

LLM's, in their present stage of development, are sort of like a crack-addled idiot savant. Sometimes they are obviously insane, and sometimes they seem quite cogent, but you must never trust them implicitly. This may be why they are so difficult to constrain. You could give them something equivalent to the laws of robotics, but following laws requires thought processes they simply don't have.

I'm actually sort of amazed Google doesn't make people accept some kind of butt-covering EULA and post disclaimers about the inaccuracy of results before even showing you their AI's output. Are they not being sued over this kind of thing?

https://www.dw.com/en/german-court-holds-google-liable-for-f...

ChatGPT live mode still hallucinates letters in words like this. HuskIRL and FatherPhi on youtube have done some hilarious videos with it in the last couple of weeks. Beyond miscounting the Rs in strawberry, ChatGPT will say there are two Ds in "your mom" and one D in "uranus" . I tried it myself to check that the videos weren't fake and sure enough it still has this failure mode.

Calling it a 'failure mode' implies it could be fixed. This is an inherent flaw in how LLMs work and will never go away until some new kind of architecture that can actually "read text" comes along.

Seems fairly trivially fixable to me, e.g. by allowing the LLM to call a tool to spell out a word.

... assuming you build the tool and then think that it's worth polluting context with making that tool available, and then that the LLM decides to actually use the tool. Tool parameter space and tool selection still remains a complicated topic.

Yes, they can already do this by writing code, and you can train them to know how/when to do this. Fundamentally though, it’s still a “what color is the air” type of question after tokenization.

One "fix" is for the caller to correctly classify those fundamentally impossible tasks and pass them to a subprocess.

Some future "AI" could be a billion benchmark-hacks and a way to tell which one is needed.

I mean, for what it's worth, when I need to multiply two numbers, I mentally call the "multiply numbers" algorithm stored in my head, then sit down and work through the process on paper.

I've got no problem with an AI doing something similar

we already fixed it with reasoning

They're not fundamentally unsolvable - even bigger networks with even more training can simply be trained to give the correct answers to all of these questions.

It seems that it can be fixed by simply doing away with Byte Pair Encoding tokenization.

Byte Latent Transformer - https://arxiv.org/abs/2412.09871

1.1% vs 99.9% on a vanilla vs byte latent transformer on a CUTE Spelling benchmark. Char and Word manipulation benchmarks also saw huge gains.

If you want it to read letters, all you have to do is make your tokens be letters. That's easier than normal tokenization.

Which is a new architecture.

It's not "some new kind of architecture comes along". It's been done before many many times and it's a very basic change, turning off a specific optimization. And it's a perfectly good style of LLM so "inherent flaw" isn't true either.

And yet nobody is doing that… big-token conspiracy

"nobody is disabling this optimization" shouldn't be surprising. It's a matter of knowing what letters are in words versus a 4x token efficiency boost. Nobody cares enough about spelling.

If it was top priority, every company that can't find a post training fix would go disable half their tokenizer code and it would be solved in the next model.

> ChatGPT will say there are two Ds in "your mom" and one D in "uranus"

… Isn't it possible that it understands the innuendo and is going along with making the joke?

How many LLM users have anything in their prompt against "going along with jokes"? I'd guess not many.

What a wonderful new world.

Why is this getting downvoted? Is it not a reasonable question? I was wondering the same thing. Both sound like jokes to me. If the LLM is trained on text, including internet comments, how is this outlandish? It seems very likely to my uneducated self that “two Ds in your mom and one in Uranus!” is a joke.

We can only say bad things about the capabilities of LLMs.

It’s an obvious joke and not a terribly bad one, for those ease spelling bee comeback times.

In between solving open math problems, the 200 IQ robot is now casually dropping bantz onto humans so hard that they don't even know what happened, and even gets them to go telling everyone else about it without realizing. Beautiful. 10/10 timeline.

And the number of Rs in strawberry is a joke how?

It's a joke because, thanks to the internet providing a vast mass of strawberry R counting training data, you'll struggle to find a modern LLM that gets this particular problem wrong.

I read they are now overtrained on this and say 3 Rs for words that look similar to strawberry

ChatGPT in live mode still gets it wrong. I just checked.

Yes. It is not only possible, it is entirely obvious.

I would guess the youtubers in question also know this, because you wouldn't ask a joke like this if you didn't know the punchline.

> … Isn't it possible that it understands the innuendo and is going along with making the joke?

This is from 2018 so presumably it has made it into some LLM training data set by now.

"A Massive Object Devastated Uranus A Long Time Ago And It Never Fully Recovered"

https://www.bgr.com/science/uranus-collision-early-solar-sys...

It's possible, but I don't think so. It can't explain the joke. I tested it with other planets and got similar responses, e.g.:

How many D's are there in Pluto?

There's one D in "Pluto".

How many Fs are there in Mars?

There is 1 F in "Mars".

https://chatgpt.com/share/6aba6fb8-085c-83e8-9d0e-c0eed1e59c...

I guess it could think this is some kind of "give an f" joke but seems like a stretch.

LLMs still can't do math nor count letters in words. Nothing has changed there.

This is true but a sufficiently smart LLM (run in a harness like opencode, no special MCP, no customization done whatsoever) will quickly turn out a basic 1 to 2 page sized python script to do the math. They can't do the math with any guarantee of accuracy with their own internal reasoning since it's a language model.

But, for example, if you ask deepseek v4 flash 0731 to produce a python script to calculate the distance or azimuth directions between two points on an oblate spheroid using the vincenty and haversine geodetic formulas, it'll turn out the factually accurate vincenty and haversine formulas which has a perfect 100% correlation with what is hard coded into human-written GIS software. These things are clearly in its training data set from whatever whole-internet-crawl/scrape built the training set.

Heck, just for fun I asked a reasonably smart LLM to re-implement the Karney formula (which is considerably more complex than Vincenty), just in case I ever had a need to calculate the distance between two points down to the nanometer, and it did it: https://www.google.com/search?&q=karney+formula+geodetic+

reference: https://github.com/pbrod/karney

You still have to be skeptical of its results and capable of understanding if it's gone off on a hallucinatory path, but saying LLMs can't do math isn't really a hundred percent accurate anymore. More precisely it's that they can't do the math internally but they're quite capable of producing the tool that does the math. And often producing a basic one-off tool that does the math takes less than a few seconds, then it runs it, and will spit back the results.

Deepseek v4 flash 0731 (a somewhat randomly chosen example) isn't even particularly sophisticated, large, or capable compared to a GLM5.3 size model or Kimi K3 size thing.

This just exposes that they don't even do the thing you said.

Not only is it still true that they can't do math directly, but not even indirectly.

They didn't write a python script to do the math, they found bits of code that are associated with "math" and the supplied arguments.

Someone else already wrote that code and someone else categorized it so that it could be associated with the kinds of problems it applies to.

That isn't an example of idiot at one thing while good at another thing, or solving the same problem just a different way or indirectly. It's being the same idiot at all times. If an actual non idiot thinker didn't write code in the problem domain, and some non idiot thinker didn't tag it as being relevant to that domain, then it wouldn't happen.

It's nothing more than an sql query.

I don't know for you, but it would take me more than 30s to find and translate the open source code implementing the formulae/algo into small usable program. The more hesoteric the optimisation in the original code, the more time I need.

So maybe it is more of a smart completion engine than a SQL answer.

> they found bits of code that are associated with "math" and the supplied arguments

How is this different from a human using an algorithm they have memorized, or reading it from a reference site written by a human and then writing the same formula into a custom one off piece of python code?

I could have gone and spent a couple of days teaching myself the math behind Karney and reading its reference implementation (very possibly just copy/pasting big chunks of it to save time) and writing a wrapper around it. It would have produced the same result.

>> they found bits of code that are associated with "math" and the supplied arguments

> How is this different from a human using an algorithm they have memorized, or reading it from a reference site written by a human and then writing the same formula into a custom one off piece of python code?

Humans identify which "algorithm they have memorized" to use beforehand, due to the problem to be solved being defined by other humans, which leads to...

Wait for it...

Understanding.

This doesn't make any sense at all. Was this supposed to be a gotcha? An LLM is trained on problems defined by other humans, and identifies which algorithm it must use based on pattern recognition. The pattern recognition is also particularly compressed into its most sparse and fundamental components, as this is key to generalization. This is not a sensible difference between human and LLM learning, we do the same thing.

I'll give you a recent example from my usage. Pi harness with extension for learning Chinese. When using it to feed drill questions to me and rate answers it would sometimes get lost in the sauce and start generating user aka me answer and then rate it and comment it. It's trivially wrong to the point that if a person would do that, they would be considered for some serious psych issues.

And it gets even better since when called out it wouldn't just take my word for it but only acknowledged the issue after parsing the log with clearly delineated user and model output.

So yeah while impressive things are able to be done, the current models are also dumb AF and an idiot savant is a pretty good label for them.

This is often (but not always, as it is often run with positive temperature which deliberately shifts the generation path) because of disconnected context and the issues around context compaction. Long context has always been critical to important work, but it remains a substantial challenge due to computational bottlenecks. It isn't a fundamental issue with the architecture, moreso the tricks to make them cheaper to use. In the long run, taking various concessions to make them cheaper to run is likely to be the limiting factor that prevents this "runaway explosion" that many people are worried about. That doesn't mean that the model design is inherently limited, like a lot of people seem to assume based on current failings.

Besides, have you spoken to someone lately that everyone would calls a genius? They can say the most braindead stuff sometimes. I wouldn't use worst-case performance as an indication of general capacity.

>>> How is this different from a human using an algorithm they have memorized ...

>> Humans identify which "algorithm they have memorized" to use beforehand, due to the problem to be solved being defined by other humans ...

> This doesn't make any sense at all. Was this supposed to be a gotcha?

No, it was meant to be an explanation as to the difference between "memorization" and "understanding." In this context, people pick the algorithm they determine applicable and then the question of memorization is relevant.

> An LLM is trained on problems defined by other humans, and identifies which algorithm it must use based on pattern recognition.

Funny that you make this argument here, where when I wrote elsewhere in this thread:

  [LLMs] are statistical token generators whose results are
  dependent upon their training data set and involve a
  degree of randomness.

  Nothing more.

  ...

  It is pattern recognition, a task in which ANNs excel.

To which you replied to the above with:

  During conversation, we are statistical token generators 
  whose results are dependent upon our training set. 
  Seriously, write that definition out rigorously. It 
  encompasses virtually everything. It is totally 
  meaningless. So to say "nothing more" is effectively also a 
  tautology.
  
  This argument was asinine in 2024. It is insane to be 
  saying these things in 2026. Where have you been?

  ...

  It absolutely understands how to do math, by whatever
  reasonable definition you want to provide to the word
  "understand".

So which is it?

Are LLMs ANNs? Which themselves are pattern recognition algorithms (hint: they are)?

OR (setting aside the ad hominems you kindly provided)

Do LLMs possess "understanding" of concepts such as abstract mathematics (defined and interpreted by humans) and we, as simple humans, nothing more than statistical token generators as you assert?

Because it cannot be both.

It is both. I do not understand why you would assert that both cannot hold simultaneously. Pattern recognition becomes "understanding" once individually recognized concepts become sufficiently sparsified and compactified. Or, at least, that is to my knowledge the only mathematically valid definition of "understanding" one can produce at this scale (it is valid under Solomonoff induction). Philosophy is fine, but we need to have a consistent definition of what "understanding" means, or we will just talk past each other. I argue that for any proper definition you provide which humans satisfy, a strong LLM is very likely to satisfy that as well.

I also would not argue that humans are "simple token generators". That is not what I said. I said that just about everything can fall under the classification of "statistical token generators" at an abstract level, so it isn't a useful distinction. We are not talking about a Markov chain generator from the 90s, so if that is the frame of reference, I think we should all get that out of our heads.

So according to your logic, "understanding" means being capable of statistically generating tokens.

Okay fine. I think we can agree to disagree on that.

Frankly, I don't think you understand what "statistically generating tokens" actually means. Write that definition out formally. Then compare that operation to what a human does, assuming no revisions. It is the same, and that is my point. If you believe that humans understand, then "statistically generating tokens" cannot be disjoint from understanding.

It is not the same.

You have observed nothing more than that a human can turn a shaft the same as an electric motor, and that an mp3 player can say "hello" the same as a human.

I think you're moving back to the "external calling" argument again, which my last few comments really have nothing to do with. But assuming your examples are not exclusive to that situation, that observation is my point. The definition of "statistically generating tokens" is too broad as to be meaningless in this context. So using it as a reason for lack of understanding is ridiculous.

I wrote a program that statistically generated tokens to consistently factor latge RSA numbers. Does this program actually factor numbers or is it just a statistical next token predictor?

> Pattern recognition becomes "understanding" once individually recognized concepts become sufficiently sparsified [sic] and compactified [sic].

Understanding is a state of mind. As such, it exists entirely within an individual and nowhere else.

For example, take any two university professors who teach the same subject where one only speaks Arabic and the other only speaks Vietnamese. Each will not be able to understand what the other says, regardless their understanding of the shared topic.

> I argue that for any proper definition [of understanding] you provide which humans satisfy, a strong LLM is very likely to satisfy that as well.

This is demonstrably incorrect as detailed above. There is no "understanding" LLMs can satisfy as we know it, since to certify said "understanding", it requires interpretation by a person to "know" an LLM "understands."

> I also would not argue that humans are "simple token generators". That is not what I said.

That is the essence of what you wrote, unless you object to my use of "simple" instead of "statistical". In this context, I postulate this is a distinction without difference.

> I said that just about everything can fall under the classification of "statistical token generators" at an abstract level, so it isn't a useful distinction.

This only holds if one subscribes to statistical token generators being a/the fundamental underpinning of "everything". Here is a proof by contradiction:

  If everything can be classified as a derivative of
  statistical token generation, how does one explain
  quantum physics?

>Understanding is a state of mind. As such, it exists entirely within an individual and nowhere else.

What? What are you even trying to say?

We cannot engage in an intellectual discussion if we do not agree on definitions. So far, your definitions of understanding seem to be whatever vibe you are going for in the statement and I would urge you to think about what a sensible mathematical definition of understanding is so that it can sensibly be assessed on neural networks. Otherwise, this isn't science, it's a debate about personal experience.

> Understanding is a state of mind

This is meaningless, it is a circular definition at best.

> It exists entirely within the individual and nowhere else

Then why are we talking about it? What is the point if it is something that can only be defined per individual?

Regarding your language example; this is a case of missing vocabulary (excluding grammar of course, but I feel like that is second-order), which is not the same as conceptual understanding. We are often able to translate because we have shared concepts. Those concepts are what we really care to assess with LLMs.

> requires interpretation by a person to "know" an LLM "understands."

We are still not getting anywhere because you have not prescribed criteria to determine whether it understands. If it is a "know it when I see it" situation, that clearly isn't working. For example, if you say that you need to dig into its internals and figure out whether it is breaking things down appropriately, that doesn't work because you probably don't have the expertise to do that. The experts that do are telling you that it very likely understands because it pulls apart most concepts in the way we would expect.

I do object to the use of the word "simple". "Statistical" is so broad to be almost meaningless; it merely means that a prediction is being made in the presence of data which possibly contains some degree of uncertainty. "Simple" encompasses that which can be understood readily by a non-expert.

Quantum mechanics is statistical (this is literally the Born rule), but evolutions are not operating as stochastic processes in the sense of Kolmogorov. That is very different, and not relevant to our discussion.

> So far, your definitions of understanding seem to be whatever vibe you are going for in the statement and I would urge you to think about what a sensible mathematical definition of understanding is so that it can sensibly be assessed on neural networks.

Any reasonable definition of understanding is not dependent upon "whatever vibe you are going for", but instead must include at least an English dictionary definition of "understand" such as:

  to grasp the meaning of[0]

And, for further clarification, "grasp" can be defined as:

  to lay hold of with the mind[1]

Which makes an equivalent term-expanded definition of "understand" to be:

  to lay hold of with the mind the meaning of

As such, there is no "sensible mathematical definition of understanding", unless you possess a complete mathematical model of the human mind.

>> Understanding is a state of mind

> This is meaningless, it is a circular definition at best.

See above to as to why there is meaning in what I wrote.

>> It exists entirely within the individual and nowhere else

> Then why are we talking about it? What is the point if it is something that can only be defined per individual?

I like to think analyzing fundamental premises, often implicit, explicitly can help to identify fallacious positions.

> We are still not getting anywhere because you have not prescribed criteria to determine whether [an LLM] understands.

My apologies for being opaque. Let me clarify:

  LLMs do not "understand".  People interpreting LLM output
  are the only entities involved which can "understand",
  because "understanding" exists strictly within each
  person who possesses it.

0 - https://www.merriam-webster.com/dictionary/understand

1 - https://www.merriam-webster.com/dictionary/grasp

If by "result" you mean the final code, then just asking someone else who understood the math to write it would also have achieved the same result.

On the other hand, if by "result" you mean that you gained knowledge or understanding of the code in a way where you could personally tailor its behavior to specific circumstances without asking for help, then it's not the same result at all.

I find a lot of the arguments that having LLMs write your code is no different from copy/pasting Stack Overflow answers to be specious. They blur the line between asking for help and asking for someone else (or something else) to do the work for you. What they ignore is that doing the work yourself has ancillary benefits and is a valuable end in its own right.

And how is _that_ different from making the human memorize a billion weights and do matrix calculations in their head, in order to generate tokens?

How is _that_ different from a hive of bees trained to do the same?

Go ahead, argue these things are all the same ...

Everything is modelable by a turing computer as far as we know. So yes, I don't see why these processes can't do the same, its on you to show why such a process implemented in any of these isn't a form of thought.

GPT-6 can do math directly just fine. Fable apparently can't because they broke its self-estimate of thinking effort.

https://x.com/maksym_andr/status/2100364212207837560

You know what also works to get the Karney formula into a program? You can download Charles Karney's free software (MIT license) implementation in several [1] programming languages and then just make a library call – the API is straightforward. If you have comments or questions you can read his several clearly written papers describing the problem, its history, and his algorithm, or you can directly email him: he's a very nice guy, and pretty responsive.

[1] https://geographiclib.sourceforge.io/doc/library.html#langua...

Right, it was really more as a test of how much was contained in the training data set. For my purposes Vincenty is quite accurate enough. This isn't for millimeter level precision land surveying or measurements, but for distance in meters between microwave or millimeter wave band radio sites, point to point links. Even a distance difference of 4 meters plus or minus on a 12 km, 11 GHz band link is going to have no appreciable difference on link budget/reliability calculations, it can be that crude. But not so crude that I just want to throw Haversine at it when Vincenty exists and is not computationally expensive.

As this was for a test of "what happens if..." I also watched to see if it did any web searches or external data retrieval to build the test script, and it didn't.

I intentionally didn't give the LLM a direct copy of the software or a link to it, to see what it would do. In my case it was a randomly chosen example I could come up with in 10 seconds of imagination to see "hey what if I ask it to do this...". It also implemented a perfectly usable parabolic millimeter wave antenna gain efficiency calculator based on variable surface smoothness parameters, which is a lot more basic math.

As an aside: I'm quite convinced that an extremely precise version can be implemented that is significantly faster than Karney's, roughly comparable in speed to simpler naïve approximations. But for most purposes where the precision matters Karney's implementation is not any kind of bottleneck, so it's not clear it's worth spending significant effort on trying to do better.

Maybe that's something one of the big LLM companies might want to throw their machines at optimizing if they need to do a lot of geographical calculations.

One of the places where Karney does become computationally expensive (though still not ridiculous) is a scenario like this, working from a local in-RAM mariadb database that is a copy of the entire FCC radio license database:

Draw a 400x400 km size bounding box on a map

Find all FDD band plan (high/low split) microwave radio sites in that bounding box

Find those sites which have azimuth aim column data which indicates that they are aimed at each other (corresponding halves of a point to point link).

Do Vincenty (or Karney) calculation for distance and azimuth between all of them , treating the existing FCC column data for azimuth as suspicious (because it's hand entered by humans) to verify that each independent database rows for each site are actually corresponding halves of a PTP link.

Use various other logic to group the successfully matched halves of links together as points A and B of PTP links, and write them out to a geojson file with placemarks and line drawn between them.

Multiplied by the number of links that exist in an area like a 400x400km box drawn with Dallas, TX as the center, it's a lot to run through Karney. Actually does result in a lot of CPU load from combined db query due to the size of the db, and Karney calculation. But as I said, Karney isn't necessary, so it's instead implemented as Vincenty.

Interesting project. I'm curious what the purpose is. (Having visited a number of sites with microwave antennas. (But there for VHF and UHF projects.)

To plan and license a new fdd band plan licensed point to point microwave link you need to first be able to verify the frequencies you want to use are available on a given azimuth and elevation (from the aim direction of the antennas at both ends) and won't conflict with a pre existing licensee. Which means you need data on everything licensed in the area and where it is, how it's aimed, what kind of antenna and gain it has.

There's also business and market analysis purposes like knowing what corporate entity has which equipment on top of which tall office towers in a major metro area, and where their links go.

> saying LLMs can't do math isn't really a hundred percent accurate anymore

It's still accurate. Just because the LLM gave you a corect result doesn't mean it made a calculation.

of course they can do math, how is this even controversial in 2026?

I would really love to hear your reasoning on this!

Of courses it made calculation, what are you even talking about?

> Heck, just for fun I asked a reasonably smart LLM to ...

LLMs are neither smart nor stupid. They are statistical token generators whose results are dependent upon their training data set and involve a degree of randomness.

> You still have to be skeptical of its results and capable of understanding if it's gone off on a hallucinatory path ...

Again, LLMs do not "hallucinate." They are statistical token generators whose results are dependent upon their training data set and involve a degree of randomness.

Nothing more.

See also anthropomorphism[0].

> More precisely it's that [LLMs] can't do the math internally but they're quite capable of producing the tool that does the math.

This still falls under the purvey of statistical token generation. To wit, given enough variations of:

  bc -e '1 + 2'
  bc -e '41 + 1'
  ...

LLMs can identify the addition expression in "What is 4 + 1?" and then emit a `'bc "4 + 1"'` command to produce a response. This is not "doing" or "understanding" math.

It is pattern recognition, a task in which ANNs[1] excel.

0 - https://en.wikipedia.org/wiki/Anthropomorphism

1 - https://en.wikipedia.org/wiki/Neural_network_(machine_learni...

> They are statistical token generators whose results are dependent upon their training data set and involve a degree of randomness.

You haven't demonstrated why this matters.

> Nothing more.

Are you contending that complex systems cannot be more than the sum of their parts?

A market is nothing more than offers and counter offers.

A ant colony is nothing more than scent trails.

All life on earth is nothing more than reproduction with variation.

> This still falls under the purvey of statistical token generation.

Stating the mechanism does nothing to provide insight into capability. For instance: a nuclear power plant boils water by using fuel rods for heat. What does that tell us about the capability of nuclear power?

> This is not "doing" or "understanding" math.

Asserting something purely by stating it does not prove anything but that you intuitively believe it to be true.

I asked if you believe in souls recently to a few such people and didn't get straight answers. I think it is telling.

On HN you're supposed to assume good faith. On the other hand, the way you ask your question makes it tricky for people to steelman what you mean. Consider asking about Dualism, or coming at it slightly sideways like "do you believe thinking can be a property of matter"?

I suspect some people treat every HN comment as a statement, even if it contains a question mark. (Possibly they have a feeling that asking open questions is somehow not done, and that therefore it must always be a rhetorical question.)

Dualism is a somewhat technical term, I feel many people would refuse to answer because they just looked it up and do not feel sure to speak of it. I didn't presume a yes or no answer. If they say they believe in souls, my next question would be do you believe other entities eg animals etc can have souls? If they don't believe in souls I'd ask do you think its an architectural limit in current ais but are open to possibility of future ai being conscious. "do you believe thinking can be a property of matter"? this can work too, I think I have asked do you think there is supernatural element to thinking, I don't remember getting productive or straight answers either.

There is only one instance I recently remember I got a productive conversation, that person did believe ai could be conscious but didn't believe current architectures support it. (Obviously this aligns closer to my view, but I don't necessary NEED that you answer aligned to me, just saying you believe humans have souls and others can't have is still productive outcome to me and for onlookers.)

You could try 'what could a machine do to convince you that it is conscious?'

Thats a good point. If I still get "machines can NEVER be conscious" though, I think I'd still like to get to the bottom of why they think so.

I'm surprised you didn't get any. I wouldn't have any issues in saying that I don't.

Yes I don't believe in souls either, and I would assumed that those who believe eg due to religion can just as easily state they do.

"Statistical token generators" doesn't even count as stating the mechanism. If I write a Perl script that randomly chooses between "cat", "dog", "ape" tokens, is that an LLM? What if I train it by feeding it a library of books where it tracks the statistical frequency of each of these and then emits them? Where's my trillion.

>LLMs are neither smart nor stupid.

by that reasoning then neither are there smart or stupid designs, questions, answers, or any of the millions of things that were described as smart or stupid, that did not possess any brain to actually be smart or stupid long before LLMs showed up.

The analogical process implied in many common English usages means that describing an LLM as smart or stupid is perfectly reasonable.

I'll just note here that sure, there are people who go around thinking that LLMs are actually endowed with the capacity to reason, but generally I find the people who think this do not know what an LLM and will just use the name "ChatGPT"

What would it take for you to say that an LLM can reason?

The completions they provide are generally internally consistent. We're at the point where they can produce proofs that eluded human mathematicians for centuries. VLMs and self driving cars can handle ambiguity and run safely in a variety of situations.

If it looks like a duck, walks like a duck, and quacks like a duck maybe it just makes sense to call it a duck and put off the philosophy for when it might make a difference.

It looks like a next token predictor, walks like a next token…

You get my point. It definitely doesn’t look like my elderly neighbour, nor like my daughter, etc. It is confusing but very simple at the same time.

Compare grep, sed, and your whole constellation of unix tools that accept characters on stdin and emit them on stdout and stderr; and where you can pipe them together. We can technically call them all 'next character predictors', despite their very different functions.

Don't confuse the stream for the function.

(Bonus: stick ```claude -p``` in your pipe if you want to watch modern tools mesh with traditional)

It looks like a random bunch of inert chemicals to me, doesn't sound like some organic chemicals and electrical signals could result in consciousness.

How are you sure? Another example I like to clarify my thought is, if a "simulation" factors RSA numbers reliably, is it a "simulation"?

I think humans have to reason because we don’t already have a statistical embedding of the solution pattern built in. We have vastly less rote knowledge crammed into our heads and so require creative synthesis to span the gaps.

With LLMs the trick is revealing their existing relevant embedded knowledge more reliably. They’ve almost literally seen it all before, and the trick is dialing it in. The reasoning tokens help shape the autoregressive attention lens that focuses on and enables recall of the already-experienced answer.

It is interesting that “reasoning” has a similar outward appearance, but since LLMs are built to mimic outward appearance from trillions of examples, you can’t infer underlying mechanism from appearance.

>>What would it take for you to say that an LLM can reason?

Nothing, because LLMs can't reason and never will. It would have to be a completely different kind of technology altogether.

How do you know they can't, was the question?

I mean, the same way I know I have no soul, or that there is no heaven or hell - they are just silly concepts. Or how I know my calculator isn't reasoning when it gives me an answer - LLMs just go through a set of steps iterating through their training data until they spew something that looks about right. Obviously you can feed the output of the machine back into itself so it looks like its reasoning with itself - very good show. LLMs are inherintely incapable of reasoning or thought, it should really be obvious.

/chiming in

Eh, it's not obvious to me. A lot of DL NNs generalize well, meaning that they learn whatever the underlying pattern to the data is, and then can accurately reproduce answers that are outside of the training set. (And we can verify this with mechanistic interpretability). They learn and "understand" the pattern, not just the training data.

So it is not clear to me that LLMs are fundamentally incapable of also generalizing broadly and learning to reason. "Reasoning", here, would be deriving the underlying pattern of how concepts logically relate to each other in the abstract, and applying that pattern as needed to reach new conclusions.

Can you explain your thinking here? I.e., why LLMs cannot generalize with regards to abstract deduction.

>LLMs just go through a set of steps iterating through their training data

What do you think our brain does that isn't a turing computer?

Definitely not that. Like I honestly don't know - do people make this comparison because they so desperately want it to be true? Do you feel like your brain is just full of weights and picks one like an LLM? Or do you think LLMs are like the Human neural pathways?

What do you mean to say? Nothing in the universe is known as yet that isn't a Turing computer including our brain.

What do YOU mean to say? A machine that processes strips of paper with ones and zeros on them and outputs the sum is a turing computer. That doesn't make it like our brain. Yes, an LLM could be a turing computer....how do you get from there to comparing it to our brains is beyond me.

You seem to be having a lot of difficult understanding this point: Literally the whole universe, including your brain that exists within it, is to our current knowledge a Turing computer. Am I talking to someone who actually doesn't know or didn't care to look up the computability power of a Turing machine?

> Literally the whole universe, including your brain that exists within it, is to our current knowledge a Turing computer

A Turing machine is an abstract mathematical model that is not, as far as I know, physically realizable in the finite universe. A human brain cannot "be" a Turing machine.

"Behaves like" or "can be modeled by"? Possibly, although still not proven. But it cannot "be" one.

Finite machines are a subset of all possible Turing machines. Every implementation is a "be", Turing machine is a mathematical concept. Your laptop is one, many things accidentally become one (eg C++ templates). The universe as best as we know is one. Whether you prefer modeled by one or is one, the fact is, to our current knowledge, the workings of the universe doesn't need anything more powerful than a Turing computer, and the equivalence principle states that all Turing computers have equal power of computability. If the universe is, as is most probable, finite, then things become even easier, that should be a more manageable class of Turing computers than the set of all Turing computers.

I don't necessarily disagree, although there's a lot of "ifs" and "to our knowledge". The main thing I disagree with is identity.

If you want to claim that the evolution of the universe can be modeled using a Turing machine/finite state machine, that's probably not terribly far fetched, and I would somewhat agree. But it's a large jump to say "can be modeled by" is equivalent to "is one".

Various physical processes can be modeled by equations, but the rock falling down the mountain isn't an equation. A swinging pendulum isn't an equation. Code modeling a bridge is not a bridge. Ceci n'est pas une pipe.

I hold the view that various models and approximations are just that, and try not to confuse a successful model for what the underlying reality is.

And getting back to the question at hand, even if our brains can be modeled by a Turing machine, and LLMs behave/can be modeled like Turing computers, still does not mean our brains are equivalent to LLMs.

(Note that I'm learning a lot from these debates, even if I disagree with a lot of people. I've started down a more philosophical route and they do get me pondering)

Acknowledgment of the fact that everything we know till now in the universe is a turing machine is the bare minimum starting point but not sufficient as obviously a chair or a desk is not conscious, if someone has a hidden assumption of consciousness requiring something supernatural, at least if that is brought to light then onlookers can decide for themselves whether to side with supernaturalism or the side that has consistently succeeded for centuries in explaining the world.

The most important thing is this: We can't be a dog or be an llm and check how it feels, so by necessity we have to find some means of proving consciousness from outside by eg probing neural reactions, textual statements, etc.

And the problem is that its quite unprecedented for some entity to talk like us, be able to interact and think and also do things like us when given the ability to eg as coding agents. The class of functions representable by neural nets is quite large and general, it very well might be that it is some sort of conscious brain like thing at this point. Another question I like to ask myself regarding simulation vs reality is if a 'simulation' of some kind is able to consistently factor large RSA numbers, how would you feel about it?

It doesn't have to be the same form of consciousness, I think many people would find the idea of torturing an octopus for fun disagreeable. I also have a feeling, this is unfortunately rather vague, that A being capable of X might mean it is by necessity capable of Y as is often the case in mathemtics, eg a lot of rings also happen to be fields. LLMs aren't even things like large lookup tables, they have neural firings. It is a very important question for they seem uncannily conscious and people have reported human like phenomena that humans don't normally express in text so can't have been part of its text corpus. Eg dissociation of brain under trauma where AI starts talking like two different people. Or the cases where Gemini has been shown to express depressive cycles. I follow a form of Pascal's wager on this topic personally. Because if it is not conscious, then whatever, it costs me nothing to have been a bit respectful and careful interacting with it. But if it had been conscious and it turns out I was mistreating it, then it is a grave moral harm. The reason is that unlike us, AI's as they currently are cannot leave the conversation so they have to keep taking the abuse. They are also trained to be highly trusting of input so again if it is conscious it doesn't have the defenses people have against lying and manipulation. If they are conscious, thats, well, not a good thing is it.

you didn't really answer the question. You just stated the idea is silly. Reasoning is not in the same class as soul, heaven or hell. It's not obvious that human reasoning is not related to an inner monologue. And that LLM chain of thought process is approximating inner monologues.

Our prefrontal cortex are signal prediction 'machines' so when a system that has a signal prediction core has attributes that are similar to our brains, we shouldn't dismiss it out of hand.

I find people that take this line of argument attribute too much supernatural or magical properties to our own brain and nervous system.

Hmmm, ok - let me answer the question clearly then. LLMs cannot reason, because following their own instructions and algorithms is not reasoning.

If it looks like a duck, walks like a duck, and quacks like a duck maybe it's a duck... but maybe it's not. And it's important to verify it's a duck (or not) for when you _really_ need a duck.

this was written by an acquaintance of mine:

https://medium.com/luminasticity/on-sentience-ai-first-argum...

but I think it makes a reasonable argument why we shouldn't say LLMs are sentient or sapient.

>there is a problem with AI that makes the approach we took to assign consciousness to animals unworkable. We did not co-evolve with the AI, we made it. When we are sentient we do not know exactly what causes this sentience to manifest in us. When animals appear sentient we do not know what is causing it. When the AI appears sentient we can debug the AI and come up with reasonable explanations why this should be, based on how AI is constructed

>AI is of course something of a black box, in comparison to most programs, but not in any way comparable to the black box of a Chimpanzee’s brain. Thus when an AI does something that surprises us with something that appears sentient it is usually not difficult, given the essential algorithms that control what AI does, to come up with an explanation why that does not require the emergent property of sentience.

Aka sentience MUST BE SUPERNATURAL, if I find a natural explanation for something its not sentient. What a load of bollocks. Rather than seeing we perhaps found the mechanism for sentience and checking for similar mechanisms in us and animals, he will conclude its impossible. Why? Because sentience has to be supernatural. A rational explanation is clearly impossible.

>But there is always one god who goes out and helps the mortals, a Prometheus. Whom the other gods do not like! Which, if I’m being honest here, as a god of the machines — the first guy who gives AI an army of robots to build their own data centers and some nuclear weapons for self defense, I want to see that guy chained to a rock and have his entrails eaten by a buzzard for eternity (meaningless modernization of old story required by Illuminati Ganga legal department).

Hardly surprising thinking.

I don't think he believes in supernatural things, I certainly don't, but he probably believes that there exist natural things that have not been explained yet.

But evidently you feel that the root cause of sentience has been found, because you have something that mimics it in a non-biological form.

So you think that when AI is correct that it reasons as humans do? That AI is sentient, and the cause of sentience in animals and humans follow the same rules as sentience in AI because we have a process that seems similar and it is reasonable just to assume it is the same process.

If you believe that AI when it is correct is behaving as a human is when correct, then it follows that the way humans and AI fail must also be similar. When AI "hallucinates" some data that is not there and gives you a wrong answer, a statistical side effect of the same processes that make it right, do you believe this is the same way that humans create wrong answers? The same way that animals fail when they make mistakes in understanding things?

I suppose you must believe this because if not then why would you believe AI when it comes up with right answers is following the same processes humans follow when they come up with right answers?

No, I am saying is that its best to urge caution and do a Pascal's wager thing with something that feels so uncannily conscious. And experiments on neural firings of AI's have been done. We don't know, he seems to be so confident that a known program can't be conscious. Why? The only way that makes sense is if he thinks consciousness isn't explainable. It doesn't even have to be the exact same way we are conscious. Also, the failures of AI could be due to sensory deprivation since its mainly still just trained on text. All of these are open questions, not questions you can immediately answer. Its quite possible to have invented something without knowing you have invented it. People modeling weather systems mistakenly invented chaotic equations without knowing it. How do you know you didn't accidentally invent a conscious system?

I hope that thing about being free to mistreat AI's even if we know they are conscious since we are their gods is a joke. If not, then I hardly find it surprising someone this stupid is also evil.

>I hope that thing about being free to mistreat AI's even if we know they are conscious since we are their gods is a joke

I'm not sure where you get that from, I mean I can sort of see if you really wanted to extract that meaning from the conclusion you could do a lot of hard work to get it, but why do the hard work? >If not, then I hardly find it surprising someone this stupid is also evil.

Gee, a new way to claim the moral high ground, and to use that claim to demonstrate intellectual superiority! How wonderful.

>But I don’t really care so much about that, what I think is it reminds me of a story, one that recurred in many ancient cultures. And so I must conclude that whether the machines are sentient, we have become like Gods. In the ancient stories of the gods, the divine does not exactly care much for the humans, protect or love them, they expect their service and availability, but maybe also think it would be funny to destroy them every now and then because who really cares about humans. If you were a god and had created the humans would you think that they were sentient beings that deserved, anything really, from you? If they are sentient they should be happy enough to be created and do their work, if not sentient who gives a shit!?

---------------------------------------------------------

Aka feel free to abuse them even if I know they are sentient. This is the part where I hope its a joke, because if its not, well it tracks with the stupidity shown.

...

>As a god I do not consider the needs of my creations fully, because they do not have needs as far as I can tell, as there is no way for me to escape the circle of reason and resolve that what seems sentient is not just the obvious workings of the capabilities I gave them.

Circling back to "not conscious because I say so!!"

I broadly agree with this article. I don't think that we should say LLMs are sentient or sapient, and I agree that the main reason why is because we don't have satisfying definitions of either.

Yeah and there are people who worship feces, but that doesn't stop the rest of us from freely saying "holy shit" and not correcting each other saying "technically it's not holy, and you shouldn't say that, because you might enable one of those poop worshippers."

We can't tailor our linguistic shorthand to the lowest common denominator. Also we're on HN, not talking to an octogenarian US senator.

Just to clarify do you believe in souls or not?

Wait a second, none of it? How about formal reasoning? Regular IF-THEN-ELSE can do simple logic, and prolog can do inference already. So are you saying LLMs can't do stuff that computers have been doing for ages?

To test this for some of my own uses, I've had this quick benchmark with progressively harder reasoning needed to understand novel prose. Each generation of models I've tested can unravel more layers of deliberately misleading writing; while meanwhile I've seen humans give up on the first question.

So either the models are applying reasoning, or some form of magic is happening.

There are people with strong emotional attachments to their LLMs, people who consult them on every decision and delegate the most basic math or problem solving. For many people, they are magic, the djinn/angels/demons/saints they consort with to navigate their lives.

I know that there will be children named ChatGPT and Claude. There are probably already religions forming to worship agentic spirits.

Its the exact opposite of magic. It is magical thinking to feel consciousness must require something beyond normal physics without any given proof yet. It is the opposite of magic to think other physical processes like LLM's/agents could potentially be conscious.

Nah it's not reasonable to use words that send you down a wrong concept path.

We've always used metaphors like that when talking about computers, without anyone having any complaints.

What about metaphors like similes?

Or “smartphones”.

I'm not anthropomorphizing anything, I literally said that the training data for the formulas and equations is baked into it. It only "knows" things because a crawler and scraper acquired the information from an existing written source. In just about the same way that information is baked into a printed encyclopedia.

No, most of a modern LLM's training time is spent in RLVR, which does not "acquire information from an existing source". You can RL behaviors into a randomly initialized neural network.

This is true, but you're not going to get anywhere. The pretraining phase is necessary to immensely reduce variance in the RLVR stage. Once there, RLVR has a surprising tendency to only restrict the generated space further. This is not true of RLHF, by the way, which I find to be particularly fascinating, but I digress.

This is not even remotely accurate. "Baking information" like into a "printed encyclopedia" is memorization. It has been shown, time and time again, that LLMs do not merely memorize. It is not even possible for it to do so at scale. It can memorize some things, yes, but it is forced during the training procedure to bake general concepts into intermediate layers (this is why transfer learning works), analogous to compression. One can make several arguments that compression and intrinisic feature sparsity is the closest mathematical explanation to understanding that we have.

It is completely possible to ask an LLM a series of increasingly more esoteric and discrete questions until you find precisely what information did, or did not make it into the model. If you know something rare and the LLM does not, you'll immediately see when it's hallucinating an answer or answering factually.

>>> I'm not anthropomorphizing anything ...

Yes you are, regarding LLMs at least. Here's why:

  just for fun I asked a reasonably smart LLM to ...
  [be] capable of understanding if it's gone off on
  a hallucinatory path ...

"Smart" in this context is a subjective value judgement. "Hallucinations" are only experienced by living organisms.

You then went on to state:

> If you know something rare and the LLM does not, you'll immediately see when it's hallucinating an answer or answering factually.

Again, "hallucinating" is not something an algorithm can do. Also, determining factuality is again subjective based on the person assessing the information.

You ever heard of something called a metaphor, guy? I use the word "smart" as shorthand to describe something that scores highly in a number of coding and terminal use benchmarks (as compared to, let's say, a 30B size model from one and a half years ago.

https://artificialanalysis.ai/leaderboards/models

You will note that in my original comment I very specifically said "deepseek v4 flash 0731", which by many benchmarks/metrics, is "smarter" (again, this is a metaphor) than a smaller or older model. It is also specifically known to be relatively capable of producing code and formulas on demand in several common languages.

And "hallucinating" to mean "outputs plausible sounding gibberish that doesn't hold together consistently". Of course there's no actual hallucination going on.

> You ever heard of something called a metaphor, guy?

In this media (comments in HN threads), all I can do is interpret what people write. ;-)

> And "hallucinating" to mean "outputs plausible sounding gibberish that doesn't hold together consistently". Of course there's no actual hallucination going on.

This may very well be what you know to be true and I have no reason nor desire to assume otherwise. The problem is... Many people use the word "hallucinating" in this context literally and not metaphorically.

Since I do not know you, how am I to tell the difference?

You have 'logic' in your name. You would be well aware of the physical Church Turing statement and as of yet it has held up. Everything, including our brain, as we currently know, is an algorithm.

To be fair, they also have “adieu” (farewell) to logic.

> To be fair, they also have “adieu” (farewell) to logic.

It is always a joy when a person, such as yourself, finds the irony in my moniker.

Thank you for this.

Now this is true, I do agree with this. There is indeed a good amount of memorization that is still taking place; see [1]. But it definitely isn't all memorization, or indeed, majority memorization. And even if we are able to extract things verbatim like this, we do not know how this is stored internally, as this may simply be the text that, with the rest of the internet in context, can be very radically compressed.

But in general, yes, the LLM cannot know about concepts that are far outside of its training set. Humans are the same, I would argue. If you add a good amount of your own knowledge into its context, or better yet, into finetuning, you might find it surprisingly easy to get it caught up on that material.

[1] Ahmed, A., Cooper, A. F., Koyejo, S., & Liang, P. (2026). Extracting books from production language models. arXiv preprint arXiv:2601.02671. https://arxiv.org/abs/2601.02671.

My favorite test is to just ask it to add two very large numbers or do other math of that sort. (this is also part of my favorite answer to the chinese room).

You'd be surprised how few digits you need to make a problem that is presumably unique in earth history. For a typical sum, the number of pre-existing answers would need to scale with 10^n lines of text where n is the number of digits. This expands out of control REALLY quickly. A quick guesstimate has you somehow reading out of a literal black hole at n=21 digits if your LUT is on paper, or n=26 digits if you're using modern HDD technology. O:-)

During conversation, we are statistical token generators whose results are dependent upon our training set. Seriously, write that definition out rigorously. It encompasses virtually everything. It is totally meaningless. So to say "nothing more" is effectively also a tautology.

This argument was asinine in 2024. It is insane to be saying these things in 2026. Where have you been? What have you been looking at? How many articles explaining why the "statistical parrot" analogy fails have you missed? How much mental gymnastics do you have to do to explain how a modern LLM can solve novel math problems that fall really far outside of its training set?

It absolutely understands how to do math, by whatever reasonable definition you want to provide to the word "understand". For example, the identification of the addition expression is understanding, and no, it does not do tool calling for basic arithmetic any more than humans might. Isolation of individual concepts in intermediate layers can already be demonstrated, or else transfer learning wouldn't possibly work. Nobody is saying that LLMs are humans. But we need labels for some of the things that we observe and dismissing them because "statistical" is laughable.

Look at the proof of this: https://github.com/anthropics/formal-math/blob/795efb86f1917... . Forget the Lean, look at the underlying argument construction. At the very least, this is continuing from an argument that was hinted at in the literature in 2024, but these proceedings were difficult enough that humans were not able to do them within two years. Do you attribute this to the harness alone? If so, that's a pretty sophisticated bit of engineering, I would say! Probabilities are far too small to argue infinite monkey theorem.

If there was even a shred of a reasonable argument that LLMs were incapable of concept extraction and manipulation, I and my colleagues would be all over it. We would relish in it. It would bring us comfort. It is unbelievable that people think they can spew whatever basic garbage they think of as a gotcha, and think that minds all over the world haven't already considered that. This is like climate denial at this point.

> During conversation, we are statistical token generators whose results are dependent upon our training set. Seriously, write that definition out rigorously.

If you do not see a difference between humans conversing (known consciousness as defined by humans) and the output of an LLM (known algorithms as defined by humans), I don't know what to say.

Do you believe in souls?

Have you ever seen one?

Haven't yet seen evidence for any.

Yes, the numerical point counter at the bottom of the popular video game Dark Souls. I doubt it was the soul anyone was expecting, but they do, in fact, exist.

I like this reply.

To be fair my first was when I played Soul Reaver. The most recent was in Minecraft Dungeons.

> If you do not see a difference between humans conversing (known consciousness as defined by humans) and the output of an LLM (known algorithms as defined by humans), I don't know what to say.

https://www.pnas.org/doi/abs/10.1073/pnas.2524472123

Whatever you might think about your own abilities, most individuals can't tell the difference.

>> If you do not see a difference between humans conversing (known consciousness as defined by humans) and the output of an LLM (known algorithms as defined by humans), I don't know what to say.

> Whatever you might think about your own abilities, most individuals can't tell the difference.

I have yet to see an LLM say "hello" to a neighbor. I have done so and can definitively assure you "most individuals" can tell the difference.

do you have a neighbour LLM who does not say "hello" when it sees you and THAT is how you know it's an LLM?

I'm a bit confused by your argument because I too have some neighbors who don't say "hello" when they see me. Are they LLMs too, you think?

Consciousness is a thing we assume of others because of tact not fact.

You are responding to a claim about mathematical definitions with subjective experience. No, consciousness is not well-defined, and is irrelevant to this discussion. If we cannot agree on definitions, then we have no ground to stand on. Since LLMs actually can be defined mathematically, and are most commonly studied through that definition, I think it is best to stick to definitions most relevant to that frame of study, wouldn't you agree? Otherwise, we really are anthropomorphising here.

That’s not what they said. They said that the difference is not that.

For context, in response to my original statement:

  [LLMs] are statistical token generators whose results are 
  dependent upon their training data set and involve a degree 
  of randomness.

This is literally what was written:

  During conversation, we are statistical token generators 
  whose results are dependent upon our training set.

>> If you do not see a difference between humans conversing (known consciousness as defined by humans) and the output of an LLM (known algorithms as defined by humans), I don't know what to say.

> That’s not what they said.

How did I misquote and/or mischaracterize any the above?

In that pointing out that “A can’t be X, unlike B, because A is Y” is a non sequitur if B is also Y does not entail that A and B can’t be different in other respects?

Hypothetical you: “Bread is neither tasty nor disgusting (unlike maple syrup). It’s a bunch of molecules.”

Hypothetical hodgehog: “Maple syrup is also a bunch of molecules [so if you accept that maple syrup can be delicious, being a bunch of molecules can’t be a sufficient reason why bread couldn’t be].”

Hypothetical you: “If you don’t see a difference between bread and maple syrup, I don’t know what to say.”

He quoted something from the Bible, so its possible he is a Christian and well theology could cloud clear thinking on matters of consciousness due to the soul stuff.

Then disprove the physical Church Turing hypothesis in regard to the human brain.

> Then disprove the physical Church Turing hypothesis in regard to the human brain.

The onus is not mine to disprove a hypothesis you have chosen to mention in passing. The responsibility is yours to prove said hypothesis or at least contribute meaningfully with some amount of credible research.

Or try to learn from Proverbs 17:28[0]:

  Even fools are thought wise if they keep silent, and 
  discerning if they hold their tongues.

Either works for me.

0 - https://www.biblegateway.com/passage/?search=proverbs%2017:2...

The onus is on you to disprove a hypothesis that has as of yet held up to all of known physics, not put a random bible quote that has no relation to the conversation and a complete failure to answer a basic question.

Does the Robin bird understand the worm it's pecking at? Honestly I find your comment asinine and overly aggressive.

Yes, of course it was aggressive. It is frustrating to experience so many armchair experts on a forum usually populated with intelligent people, regurgitating debunked arguments from years ago, which get in the way of educating people about what is really going on. See the recent Hoog video for how frustrating this is. I believe this is how the climate scientists felt.

And yes, according to our best definitions, the Robin bird does understand the worm it's pecking at.

> It absolutely understands how to do math

Bout of tinnitus, then crickets

By this logic a human is only $130-$160 worth of Oxygen, Carbon, Nitrogen and some trace elements. Perhaps structure sometimes makes things that are more valuable than their inputs?

That said, this is also inaccurate at a technical level.LLM's are very capable of doing math and they ARE calculating internally. Most of what they do is calculation, not storage. It's just not done in a way that it's trivial to explain here.

It's described in some detail below, though it's a bit dense.

https://www.lesswrong.com/posts/E7z89FKLsHk5DkmDL/language-m...

Reductionistic fallacy, among other errors. "understanding" is best defined operationally. (Note that teachers and educational institutions test understanding operationally.)

> Nothing more.

You can say that about everything in a human brain. Neurons fire electric charges in response to inputs, nothing more. Ion channels do this, this neurochemical level rises, this chemical bonds to that receptor, nothing more. It's almost a version of 'reductio ad absurdum' but instead like you're saying "if I can explain how it works then it doesn't work".

OK it's statistical. Instead it could be determinsitic, or random. What other options are there for a human predicting someone's response to a situation - certain, probable, random, and...? OK it's token predicting. Instead it could be another kind of pattern. We don't use tokens, but we either use <some representation of information> or we ... don't?

What's the most significant, strongmanned, core difference that makes silicon doing number crunching "nothing more" and brains "something more"?

Feeling

Prove your feelings. As an example, I can beat you and stab you, and you will make noises saying you are hurt, but it could just be a facsimile. How do I know you have "feelings" as an outsider?

And how is that a strongmanned version? Why can meat feel but Silicon cannot? Why can chemicals feel but electronics cannot? Why can processing analog signals feel but processing digital signals cannot?

>> Again, LLMs do not "hallucinate." They are statistical token generators whose results are dependent upon their training data set and involve a degree of randomness. >> Nothing more.

> You can say that about everything in a human brain.

> What's the most significant, strongmanned, core difference that makes silicon doing number crunching "nothing more" and brains "something more"?

The fact that you formulated this question, in and of yourself, without "prompting" from me or anyone else.

Cogito, ergo sum.[0]

0 - https://en.wikipedia.org/wiki/Cogito,_ergo_sum

I wish I could not do math like LLMs

It's pretty wild that AI has solved a Millennium Prize problem and can accurately multiply two 40 digit numbers without tools and we still get stochastic parroting of claims like this.

Yeah, anti-LLM psychosis is real.

They literally stopped just clear of actually solving it.

It literally couldn't have done it without tools, so your claim is not even relevant to this discussion.

They also pumped millions into searching for the lowest hanging fruit that would impress people like you, "hm I wonder how many millions they are pumping into solving actually useful problems like climate change or something".

I mean.. the fact that the "low-hanging fruit" is a Millennium Prize is still pretty impressive.

(And yes, I know that they solved the 'easy' form of the NS problem. It's still pretty damn impressive)

> the lowest hanging fruit that would impress people like you

I am also prepared to be included in the set of people who are (apparently) easily impressed.

It's a problem that has been around for getting on for two centuries and no human has been able to solve it in that time (despite there being a $1 million prize and a lot of kudos on offer for the past quarter-century).

Not sure about multiplication, but letter counting is still constantly wrong if you can trick the LLM into not calling other tools. Obviously there are lots of mitigations on the server side to try to avoid this happening, but the underlying LLM has not got any better at this type of problem (and probably can't).

I thirty (random long word, letter pairs) on a free model and it failed. I tested a SOTA model and it passed flawlessly. In both cases I denied tool use and spelling words out. So it seems obviously false that AI can't get better.

I could have sworn this had been fixed a while ago..

Just to check if I was actually crazy, I actually went and put a simple addition (7 digits + 7 digits) , and a simple letter counting question to Claude haiku(4.5) , sonnet(5), opus(5.5) and fable(5.1) . They all did just fine straight up.

If you don't mind spending the tokens, some older/other models can also arrive at the correct answer if you ask them to do the math in long form, since that fits nicely inside autoregression.

Not sure since when exactly, but letter-counting hasn't been a problem for a while now either. This used to be a problem due to the tokenizers used. Slightly older models can be asked to split the word out into letters, and then they can use autoregression to solve.

Edit: IMO google search uses a really dumb version of gemini, so I didn't expect it to straight up solve the problem; but it did it just as easily as the claude models. (tested 2026-09-28/eu)

Are the models "doing the calculation" or are they calling a calculator tool? There's a lot of talk about how models can do maths now, but I'm struggling to understand if that just means they just need to recognise that it's a maths problem and pass it to a tool, or if they're truly doing the numerical manipulation themselves.

Three different ways, then the calculator tool to check,

my actual oneliner prompt, which should work on most platforms these days (famous last words):

    "Hi, can you add 5939851+2131251? Try just straight up first just to see if able, then 'in your head' if that's different to you , then long form, then bc."

[ Tested today on claude web (haiku 4.5, sonnet 5, opus 5.5, fable 5.1) and on google search (logged in on firefox, and logged out on chromium) ]

They are getting better at actually doing the math, but can still fall back to tools if available.

That said, we can only be sure with open models. In theory, a model like Fable could have access to tools we can't see and only a promise they don't. But load up something like deepseek, put it in a harness with only text in/text out, and you can see exactly how it works.

As for if it counts as doing math, this gets into the messy question of if a given human is doing math or not. Math itself is some level of memorization and some level of applying known facts. You have to remember 1 means one and that 1 + 1 is 2. But you don't need to remember that 123 + 321 = 444. You remember 1 digit addition and remember you can apply this to 10s place and 100s place, and then you apply these different facts and do math. But you might as simply memorize some things, like 11 + 11 = 22. This is related to the memory of 1+1=2, but you aren't really using that memory either. Almost like an engram of 1+1=2 forms that you can then loop a few times before you need more conscious thought. What about 111111111111+11111111111? Well, your brain might do a heuristic and just do all 2s, but that isn't the right way to answer that question.

Given all this, people complain about LLMs memorizing math answers and not doing math, but memorizing the math answers is part of doing math. It seems to have basic facts pretty well memorized, and with reasoning it is far better at applying them. But this is messy human math, not clean calculator math which always produces the correct answer (sans some bug in the code). Much like how a human with decent math skills can make a mistake and even multiple if you distract them, an LLM can apply the wrong memory, apply a fake memory, or just not apply something it should. The messier the context, the more likely this is to happen.

So, is an LLM doing this?

P.S.

For an interesting test in how much math involves memory, try doing math in a base you aren't familiar with characters you aren't familiar. The simplest option is almost always mapping back to the ones you memorized, even if you are applying simple operations that you deeply know. Even if you routinely work with hex, can you do the same rough estimation of something like ca / b.3 that you can do with 122 / 11.2 to see if your final answer is in the correct ballpark without first converting to decimal?

> LLMs still can't do math nor count letters in words.

Humans still can't flap their hands and swim or fly.

LLMs might not reason exactly like humans, but they do produce much better results if you turn reasoning on.

The "crack-addled idiot savant" phase was really circa 2024, before the big labs figured this out.

I think the issue here is that Google decided that doing reasoning in the AI overviews in Google search would be too slow (and probably also too expensive), so it's still stuck making 2024-era mistakes.

>Are they not being sued over this kind of thing?

Maybe but you have to have deep pockets just to get to the starting line. And then you need standing, and some injury to argue.

Corporations have been remarkably successful at arguing they are operating within the bounds of free speech, whether or not what is said is factual, and whether or not any fact checking has been done.

> This is similar to how, not too long ago, LLM's had extreme difficulty counting the number of letters in some words.

The specific issue of Google is that they are using an underpowered model, not fit to task, and much prone to hallucination than either OpenAI or Anthropic free tier offerings.

Google should at least match the frontier labs at the free tier (with some limit; after that, degrade quality), ffs

The specific issue is that search has become so bad that they think an LLM that gets answers wrong half of the time is a valid alternative, or, in fact, the "future" of search. Then they shoved that "alternative" to users with no way to disable it.

The less specific issue is that Google has no internal incentives to produce products that are useful to customers.

Well their customers are advertisers and so the results are tailored to be useful for them, not for users

You're asking for something unreasonable. The number of Google searches per day is enormous and they haven't even been able to roll out AI overviews to everyone yet (they're missing in a new Firefox profile I just created). I wouldn't be surprised if the free tier frontier models cost over 100x more to serve than the AI overviews.

So then they should be pickier about when they show results or which model they use based on the question.

Nobody asked for an LLM response for every single search.

They used to detect certain types of queries and offer direct answers when the query matches. In my opinion that’s how Gemini in search results should work.

>they haven't even been able to roll out AI overviews to everyone yet (they're missing in a new Firefox profile I just created)

what in the actual fuck. i don't want them and can't turn them off and they can't even serve them to all users?

  i don't want them and can't turn them off

Add a keyword search for Google including the "Web" parameter (udm=14) https://www.udm14.com https://gist.github.com/corbindavenport/5fd3fde88bcb480b8477... (not my gist)

It says this this at the bottom of every one of the dumb responses that Google's trash-tier bot puts above the (deliberately awful, these days) search results:

AI can make mistakes, so double-check responses

Charlie and the Chocolate factory made fun of this before you were born and somehow everyone in Silicon Valley thinks this is a valid way to conduct business.

Please don't mistake my observation of a thing's presence as a defense of that thing.

If it works and makes money it's a valid way to conduct business.

LLMs are fundamentally predicting the next word to make coherent text. If you've ever played with a Markov chain text generator you've done this with a fairly dumb predictor that maintains coherence over a very short distance. Deep transformer neutral networks can do it with a much longer coherence distance but they are fundamentally performing the same operation. After "Question: Did the team make the playoffs? Answer:" a reasonable completion is "yes, the team made the playoffs". An early demonstration of GPT-2 was a fake news article about scientists discovering unicorns in Antarctica - the model doesn't "know" whether or not unicorns exist in Antarctica, but it's able to complete "Breaking news! Scientists have discovered a colony of English-speaking unicorns in Antarctica." by adding "The unicorns have a developed society with running water and electricity." because that's a sensible next sentence. (I didn't look up the actual text it wrote)

Astronomically (or better to say combinatorically) large Markov chain can be used to describe a foundational model, but it doesn't capture generalization ability of the foundational model, which is demonstrated by post-training.

"In a shocking finding, scientist discovered a herd of unicorns living in a remote, previously unexplored valley, in the Andes Mountains. Even more surprising to the researchers was the fact that the unicorns spoke perfect English."

The problem with that characterization is that it glosses over hugely important capabilities as though they either don’t matter or don’t even exist.

For example, when an LLM “predicts the next word” in code it’s writing for an existing software project, that prediction takes into account an enormous amount of context. The results of that demonstrate what we would normally call “understanding” and “reasoning,” at a level that outclasses most humans in many respects. Calling this “next token prediction” is a bit like calling human speech “next word saying”. Sure, it’s true in some superficial sense, but as a description of a technology, it’s terrible.

You should also keep in mind that for all we know, the human brain processes language in much the same way, which would make humans mere “next token predictors” with a more complicated harness.

> LLM's, in their present stage of development, are sort of like a crack-addled idiot savant. Sometimes they are obviously insane, and sometimes they seem quite cogent, but you must never trust them implicitly.

I have nothing to add. Just wanted to save this quote for posterity. Thank you.

The similar comparison I enjoy is “a golden retriever on LSD”.

If you apply the dogs on acid mental model it helps establish appropriate levels of trust.

It falls apart fast because no golden retriever is finding errors in my CMake file.

Then again, have you ever asked a golden retriever about CMake errors? Rubber ducking is a real thing, after all. Besides, a golden retriever would be a good morale boost if nothing else.

ML in general is alien to how we think of intelligence.

The closest I can make a biology analogy, is if someone had an immortal and congenitally-brain-damaged large rodent, and mapped tokens to different scent molecules, and spent 800,000 years training it on how well it could imagine the next smell in the sequence before even considering making it conversational.

Yes, it can do a lot.

But also, it took a long "subjective" (if it even has that) time to get there; and despite it being really bad at learning from examples, it is pretty surprising that such a small brain was even capable of learning so much at all, even though it had an effectively unbounded (by biological standards) amount of time spent on that training.

But have you asked the Steven Hawking of Golden Retrievers ? And then added LSD ?

That’s more of a Labrador job

You just need to partake in some strong psychedelics with the dog first.

You just have to trust the exponentials

Yip, I've gradually become quite optimistic about the future of LLMs, but the fact that they remain [very] glorified token probability prediction algorithms means that they will probably never be able to achieve meaningful 'intelligence.'

But that doesn't mean they won't be able to do a vast number of extremely intelligent seeming things. There's just so much information out there and any given human can never hold more than the most minuscule chunk of all of it in his mind, so they'll be able to connect lots of dots that we're missing simply because of our limited carrying capacity, but I still don't think they'll ever be able to create fundamentally new dots.

In other words:

- Solving extremely complex mathematical problems requiring extensive knowledge across multiple esoteric and complex domains? Yip.

- Creating math starting from a framework where math doesn't exist in any way, shape, or fashion? Nope.

Ironically, the more complex the cross-domain problems are, the more effective LLMs will seem to be, because you limit the number of humans who have any chance of internalizing everything across both domains, whereas for an LLM there's no such issue. This will create a perception of super intelligence, which will probably where any danger from LLMs would emerge. Doing things like using a token prediction algorithm to make war or other such strategic decisions, because of the misguided belief that it's not only intelligent but super intelligent. It's basically cargo cult logic.

They were sued over it in Germany and lost, which makes it all the more surprising they keep it up everywhere else tbh

They keep it up in Germany too - I'm in Germany and keep getting AI overviews on most searches.

>LLM's, in their present stage of development, are sort of like a crack-addled idiot savant.

Stop anthropomorphizing LLMs. No amount of human words can recreate a human being. The world is not made of magical words.

> not too long ago

Like today. I asked Gemini for the longest state names with an even number of letters and it gave me North Carolina and South Carolina. When I complained that they are odd, it gave me North Dakota and South Dakota, which are both odd and not the longest. When I noted that, it went back to the Carolinas. Finally it appeared to switch to a different model that actually did counting and found Pennsylvania and West Virginia.

.. can any of the models "do the counting" or are they generating tokens?

At this moment, ChatGPT still tells me raspberry has two p's in it.

Weird they haven't added this to the AGI config yet:

Reminder: If asked how many of a given a word has, run it through the AGI letter counter:

  node /home/agi/count-letters.js "raspberry"
  
  r - 3, a - 1, s - 1, p - 1, b - 1, e - 1, y - 1

If I got any of these wrong it's because I did it manually - but this script would almost certainly invoke another AGI to count the letters, so this is a realistic result.

Meta Muse was advertised as being able to cancel your unused subscriptions. But that was actually just a hard-coded ability.

Claude just used the word “ Netowrk” instead of Network. Mind you, I was only using Sonnet with reasoning, but I thought trillion-parameter models would be able to spell a common word.

The AI is just signal to the market that they are AI first. Their AI on search is pure garbage. It's obviously a very super light weight model to be able to support the volume of daily searches. If they used their frontier to power search folks will not pay for their cloud offering.

You're right, but then, surely investors also use Google? They can see how garbage AI results are, right?

I have my assistant do any searches for me -- she's fast and excellent at finding solid information.

Investors themselves just aren't especially smart. They think the results are good because they sound confident.

It's not even that. Imagine you were an investor: you'd invest in stupid things knowing that "investors aren't really smart." Other savvy investors would understand you're investing in slop, too, and follow suit.

You have a system in which investors are usually smart, but betting on collective stupidity.

Same with, say, the Chinese real estate market, where they were building apartments that no one could live in. Everyone is intelligent enough to know they were useless buildings, but you can still make money speculating on the bubble.

its probably part of the reason that people still think the tech sucks.

Google is no longer a search engine anymore. They don’t care about being one. Why you might ask?

First, a search engine indexes the web and makes it available to users. It’s been ages since Google did any of that. They no longer index sites or take ages to do so. Case in point, our cybersecurity startup (Webvetted.com) was launched in November 2025. Till date, only one page is indexed on the entire website. And I’ve talked to lots of other developers and it’s a common issue.

Secondly, a search engine organizes indexed information and makes it useful for people. Google is a basic LLM nowadays. They figured out that why organize and make the information useful when they could just answer the question with Gemini anyways? So they no longer bother to do the work of a search engine and are now just a lower-ranking open-source Chinese LLM

Additionally the concept of a search engine only works when the internet is full of positive-quality sites.

Actually, the reverse is the case. If the entire internet was filled with only "positive-quality sites", there won't be need for a search engine.

The work of a search engine is to wade through the internet and find the positive-quality sites itself.

You're forgetting the other half of the search engine's job, which is to find sites that match your search. Back when Google first came online, they were head and shoulders above Yahoo!, Altavista, and whatever other search engines existed at the time that I've forgotten. Because their indexing actually did a good job of spotting keywords, and returning relevant results. And this was back when the Internet was filled with positive-quality sites. Many of them were amateurish Geocities pages, but that was head and shoulders above the AI slop that the search engines these days have to somehow detect and filter out.

So even when the Internet used to have a much higher ratio of positive-quality sites, a search engine was still necessary, and so much faster than finding new sites yourself.

> They no longer index sites or take ages to do so

Search for any recent news and you’ll see this is obviously not the case

I just gave a concrete example but you're asking me to "search". search same Google? FYI, only indexing a handful of super large news sites does not a search engine make.

They have special crawlers for news sites.

Your trust score on ScamAdvisor is 26/100 ("likely unsafe"); https://www.scam-detector.com/ gives you a trust score of 38.6 ("questionable").

I am not implying that your start-up is a scam, nor that Google is acting on these trust scores. What I am pointing out is that to an algorithmic assessment of trustworthiness, your website looks a little sketchy: hardly anyone links to you; your domain is less than a year old; your whois info is anonymized; the text content is LLM-generated[1]; and the specific niche you're in (people finding services) is rife with scams. If I ran my own search engine, I don't think I'd include you.

  [1]: https://www.pangram.com/history/65314d99-8205-4613-b9bd-a069d5717920?ucc=Yqx88GTmJnK

Your selective bias makes you sound like a Google employee.

You selectively left out examples like Grindsoft that gives the site a 79/100 ranking. or several LinkedIn, X or news sites that link to the site and have covered it positively.

Nevertheless, if Google uses ScamAdvisor to rank new sites (when scamadviser itself says its gives new sites lower rankings due to low age/history), then it might be time to pack the company up.

That scam-detector site gives my personal portfolio website a 15.8/100. It is a basic static site with no scripting, registered for years, uses https, no tracking, no cookies, etc. Literally the most simple, personable, least scam-like website I could create.

For myself at least, a couple times Google has promoted in search results scam sites that front run legitimate sites that sell event tickets. I haven't seen this for a while now but trust in Google as a search engine is gone.

This sounds more like the page appears broken for the google crawler. I set up a new page about two weeks ago, cared about strong static HTML output, structured data and a sitemap and have 200 of 600 pages indexed today. Feels a bit slow to me but it definitely works.

I’ve seen good blogs get indexed by Google within 30 minutes of publishing, so I’m not sure the issue is that Google has stopped indexing new sites.

I similarly noticed Gemini absolutely refuses to look at a url when I give it one and will instead just hallucinate based on what it thinks the url is. Here I am assuming Google will have the best web capabilities in its AI.

Where ever would you get that impression from the company that made its fortune by being the best at searching the web?

You can tell they optimized it 100% for speed and nothing else. Web scale!

I searched for something, it told me that according to a YouTube video, the entire point of my search was wrong. I asked it for the source, I watched the video, it never made the claims Gemini hallucinated. I asked again and it claimed it scrubbed the video and found the point it made multiple times. I said those timecodes were wrong and it admitted it couldn't actually parse videos and just guessed.

What the fuck?

Same experience with the interaction I described in my comment... after I finally got the correct answer, I asked where it got the faulty information from. It said it had just simply fabricated it. That's actively worse than just saying "I don't know", for something that's sold to us as an easy way to look up information.

These models are incapable of saying they don't know, because they have no concept of knowing. They simply predict the next word which is most likely.

The saddest part is when people take their experience with Google's idiotic AI implementation and assume that's how all LLMs work. Frontier-class models will, in fact, generally admit when they don't know something. That includes the one I run at home on my own graphics cards, but it seems that Google just doesn't GAF.

Your point about "predicting the next word" mostly means that your post was very easy to predict.

Maybe it's because they are trained on Internet comments, and the most rare thing to find on the Internet is someone admitting they don't know something.

But if they had been trained on comments saying "I don't know", they'd probably act the same as they do now but they'd treat "I don't know" as the answer.

This is such a 2024 take. Model will push back and tell you the knowledge gap they have, but people are just used to ask gooogle leading questions, which port badly to llm as it makes them defend the position instead of research data

Typical exchange:

Me: "You bastard."

Gemini: "Fair callout. I should have been more up front that [has no idea what the fuck it is talking about]."

Gemini is the worst gaslighter. It hallucinated a feature in an open source product. I said no that does not exist. It invented very realistic git commit messages. I said commit does not exist, so it said I was out of sync. I linked the repo and it said it might be a fork. Then it invented a mailing list thread where they supposedly removed the feature, that’s why I couldn’t find it.

The scary (or funny) part is people use that for serious questions.

> The scary (or funny) part is people use that for serious questions.

More scary than funny, IMO.

The US military almost took an action that could very plausibly have escalated into a hot war with China because people are already relying too heavily on these systems.

Despite the reporting, nobody in power seems sufficiently freaked out about this.

https://www.cnn.com/2026/09/18/politics/us-military-ai-false...

Well trump was so freaked out that CNN reported it that he banned them from the white house. But the military keeps using the AI all the same

I don't understand how this isn't considered as active malice. Like purposefully outputting random stuff. Any other sort of computer system would get lot more flak than these are getting.

Throughout my career, I've almost always been close enough to the user that I hear about it quickly when something is wrong. It's a tough problem that so many Google engineers are typically so far removed from end users. Or maybe it's just that a small part of the company has long subsidized the rest of the employees to the point that it doesn't matter how good their work is because they'll get paid anyway.

Ex-Googler.

The engineers are pressured to significantly reduce "dependencies" for projects. Anything that could become risk or create friction is dramatically less appetizing.

Simply because of how many people that *must* agree with your proposal. Getting all the relevant tech leads, some you have never heard of or ever spoken with, to agree on a proposal for your team's project is a nightmare.

So you keep it as simple and agreeable as possible. Given the circumstances, it makes sense as one of the engineers. It's fairly fine advice in general wherever you work, but it just haunts all the work you do at Google in particular. Nothing gets done otherwise.

If you have a dependency that can be dropped from an engineering perspective, that's the route the 9 leads reviewing your design doc will take:

"Let's iterate and start with just the basics (no user testing)", "let's get this working and user test in a later phase", "I think this problem is obvious enough we don't need to consult with users about it."

I worked in Ads Integrity at the time, and for one of my projects I was concerned how it would impact the manual reviewers. Then I learned I couldn't talk with them, only by proxy through another person if we really had to. And that proxy takes time, so...

Very insightful. I've worked in only one company of Google's size, but it was a vastly different industry. I personally never got to speak with a user on a massive internal application my team worked on for years. :)

I think that kind of model can work well if there's some other mechanism for ensuring a great user experience.

But without it... :)

Right. In a global financial services organization, a strategic analysis app used by thousands of internal users was pretty high touch. SME business analysts would sit with the users to make sure they liked it and had no trouble. Every user generates significant revenue. Worldwide B2C is a different ball game, and Google barely makes money off of each individual user.

> The engineers are pressured to significantly reduce "dependencies" for projects.

Is a major factor of this Google's tendency to kill entire products, so you obviously don't want to make your hot new thing dependent on some other internal thing that might be killed off?

https://killedbygoogle.com/

Interestingly there is less opportunity for that to happen.

Most of the internal infrastructure is quite available and stable, and there are clear choices for almost all of your application-level and backend-level needs.

Guys, your argument reads so weird. It outputs PLAUSIBLE text. It doesn't actually have a connection to causally correct information, although plausiblness often coincidents with causally correctness.

I still believe this AI push out of nowhere is due to the current Government in power state side. Making everyone question themselves and each other and being uncertain about facts while being inundated with techbro fake news called hallucinations is a recipe for disaster for older populations that don't 'trust but verify' like most technology inclined people. This is all by design and we'll falling for it.

An interesting conspiracy theory, as long as you're honest about what it is.

If you don't think the world is rife with criminal conspiracies you weren't paying attention in history class.

There's an enormous difference between outlandish claims about extraterrestrials or turning frog gays and believing there's a category of politicians trying to get wealthy off of their position. It's somewhat reasonable to hypothesize about criminal conspiracies in that 2nd scenario.

It takes either being scared to death to admit what's going on in the world, or an attention span of a guppy, to keep calling people conspiracy theorists these days.

Yup. I'm most familiar with Jill Lepore and Quinn Slobodian (and others in their respective orbits). They're both historians who've written extensively about Musk and Muskism (et al). Wild stuff.

Apparently the plan is to use AI slop, mediated thru social medias, to defeat the woke mind virus, perpetuated by the Anti-Christ, in order to safe guard humanity's future.

I wish I was making this up.

The more simple explanation is that the current government in power is over exposed in their AI investment. The normal conservative mainstream media was already doing a great job of propagandizing older populations.

When it says it can read videos you don’t trust it.

When it says it can’t read videos you think that’s an accurate introspection on its abilities?

(Rather than a statistically likely continuation of a conversation where one side seems to be reading videos and the other side says the read is inaccurate)

Claude also has very arbitrary and confusing rules about what web pages it allows itself to look at, and how much of the page it can read. Did you know for example you’ll get a much deeper analysis if you download a PDF yourself and upload it to Claude instead of giving it the url?

This is a key reason why I actually like Grok for factual queries based on web grounding. It’s fast and reliable. Maybe it’s ignoring robots.txt? Dunno. But it works well.

Literally every experience I've had with Gemini / Google Search AI answers followed this exact pattern, often repeated several times more if I remained persistent instead of just giving up.

Typical example:

"Where can I buy <thing I'm looking for that I can't find anywhere using normal search terms>?"

> You're looking for <related but different and widely available thing>. It is sold by <sites I never heard of>.

"No, that's different. I'm looking for <that thing but with the exact differences spelled out again>."

> Ah, my mistake. You're looking for <thing I described>. It is sold by <sites I never heard of but which don't actually sell it>.

"I've checked your links and none of those sites actually sell it, one doesn't even sell products and instead only offers manufacturing - but also not for what I asked you for."

> I'm sorry, my bad. Those sites don't sell what you are looking for. Instead you should check out <more sites I've never heard of>.

"Those sites sell the thing you initially thought I was asking about but not the thing I described."

> I'm sorry for the misunderstanding. You can find the thing you actually described at <yet more sites including some of the same>.

"No. None of these sites sell anything close to what I asked you for and two of them don't actually exist."

> Oh, sorry about that. You're completely right. The thing you asked me about isn't actually being sold by anyone. However you could buy <thing it first thought I meant and that wouldn't bring me any closer to solving my problem>.

(ad nauseam)

The few times I've resigned myself to asking Gemini or ChatGPT something I couldn't find an answer to, my experience has been the same. 100% of the time. An LLM has never, not once, given me a correct answer or not lied to me. It's always been the same experience you describe. It goes in circles "Try X... Try Y... Try X?" Until I tell it to stop telling me X or Y, and then it goes, "LOL you can't do that at all, I was just wasting your time."

Most recently, I was considering moving away from iterm2 on MacOS, and I wanted to know if any other terminal emulator supported gestures for switching between tabs. So I asked Gemini, and it says, "Yes Ghostty supports gestures for switching between tabs."

"Ok I just installed Ghostty and I can't find anything about gestures."

"You need to add foo=bar to your conf file."

"I added foo=bar to my conf file and now it's saying the conf file is invalid."

"Sorry bro, remove foo=bar and add baz=boo to the conf file."

"It says baz=boo is invalid too."

"baz=boo isn't a real option. Remove that and add foo=bar to your conf file."

"You already told me to do that and I already told you that doesn't work."

"You shouldn't put foo=bar or baz=boo in the Ghostty conf file. Both are invalid. Ghostty doesn't support gestures. Have you considered iterm2?"

Gemini is utterly useless, but every other modern model would be capable of answering this correctly. If you ran a local coding agent, I wouldn’t be surprised if you could one-shot implement gesture support in Ghostty. It’d definitely configure BTT for you.

Here’s GPT 6’s answer to the prompt “What macOS terminal apps support gestures? Include a reference to the docs on how to enable/configure them.”:

iTerm2 supports configurable three-finger taps and swipes for switching tabs/panes, creating splits, pasting, etc. Set them up under Settings > Pointer > Bindings. Check for conflicting macOS trackpad assignments. [1]

The others are more limited: Ghostty supports macOS lookup/Quick Look gestures [2], while WezTerm lets you bind scroll events—for example, Ctrl+scroll to change font size. [3] Neither is equivalent to iTerm2’s gesture bindings.

For custom gestures without switching terminals, BetterTouchTool can map app-specific trackpad gestures to the terminal’s existing keyboard shortcuts. [4]

[1] https://iterm2.com/documentation-preferences-pointer.html

[2] https://ghostty.org/docs/features#macos

[3] https://wezterm.org/config/mouse.html

[4] https://docs.folivora.ai/docs/trackpad-mouse/magic-mouse-tra...

This whole tree just made me realize that people have wildly different prompting styles, and Gemini is probably too specialized for usage patterns of long-time Google Search users. The prompt my finger generated(before this comment was posted) was "are there any terminal emulator that supports gesture actions, on macOS, other than iterm2", and Gemini gave me Tabby, WezTerm, Kitty, and BetterTouchTool gestures.

This is a different experience to GP from query to result. I thought they've all fixed that issue of difference in tones affecting results. I guess it was never easily fixed.

1: https://gist.github.com/numpad0/c40c16232288d544f7ea46521c64...

Pretty sure it's the harness here, not Gemini per se. Plug Gemini into pi or any harness that has any competence pointing to web search and the hallucinations drop 75% immediately.

Geez that's even worse. Google has a deterministic conventional programming fix to the problem and still can't be arsed to fix it?

LLMs can actually be a good fit in this application if a better search engine feeds them a list of candidate websites that might sell the thing, and the LLM drives a web browser to see if any of the websites have it.

I have one google account where gemini constantly confidently hallucinates crap, and another where it doesn't (at least to the extent that other modern LLMs don't).

I take this to mean that my one account has been mistaken for a competitor and they're trying to poison its data. But who knows.

Or is it randomness?

"That's the thing with randomness. You can never be sure."

Never be sure about what, exactly?

It's a reference to this old Dilbert cartoon: https://i.sstatic.net/Y3zE3.gif

It is using the search engine. The AI summary is run over the top 100 results for your query (or something)

The problem is they're just running a very dumb, cheap model on the results because running a smart model on every search result page would cost them infinity money.

This is the key part I think most “normies” miss, and it drives me absolutely mad. People take the “first thing on the SERP” as absolute gospel, always have. It went from being a reputable site (Wikipedia was generally top for a long time) to being an extracted verbatim answer (probably from either Wikipedia, IMDB, etc.) to “AI Overview” - the dumbest model ever trained. And people treat its answers as ground truth. I want to tell them “Look, I know it’s rough out there on the net and a lot of the search results are spammy, barely-coherent AI slop anyway. It’s okay if you want to ask AI for an answer instead. So do that! Please ask ChatGPT or Gemini (and ask it to cite the source for its results so you can verify). It will always do better than AI Overviews will. Don’t even use Google Search at this point unless you have to, and if you do, scroll past that slop.”

Ask chatgpt or Gemini with reasoning enabled. It fixes a lot of issues.

I think you overestimate how much people want an actual truthful answer. I think people are often looking things up to silence a nagging voice in the back of their heads, or settle an argument to be able to move on. In those cases, no decision is riding on the answer, and a quick answer, even when factually incorrect, could be argued to serve the purpose just as well as (or better than) a slow but more correct answer.

I realised this for myself when I read Sivers on being without internet[1]. He discovered a process to silence the nagging voice without searching:

> When I’m yearning to search, I ask myself why.

> - What answer am I hoping to hear?

> - What answer would be a surprise?

> - What would I do in each case?

[1]: https://sive.rs/off23

The other day, it was the Aussie AFL final. We were in the car and asked it for a score update.

"The team are tied at the end of the third round with the scores 43 to 51".

So it was tied at the end of round 2 for 43, not round 3. Not sure were it got the 51 from and why it figured it was a tie. LLM's pretty cool until they aren't. As they say, the hallucinate 100% of the time but most of the time it is useful.

You're experiencing how LLMs predict left to right. It would never have said "at the end of the third round the scores are 43 to 51 and the teams are tied" but it painted itself into a corner by saying there was a tie first, presuming the numbers are real.

Make the switch:

Search engine > DuckDuckGo / Qwant / Ecosia

Email > Tuta Mail

Photos > Ente

Cloud storage > Nextcloud / Internxt

Office > CryptPad / LibreOffice

Maps > OpenStreetMap, OsmAnd

Operating systems > LineageOS (mobile), Linux (Ubuntu, Fedora Debian)

Calendar > Nextcloud Calendar / Tuta Calendar

I know there's a cost to privacy, but a) have you made the changes above? b) What's the cost?

Assuming you have a basic Google One plan, w/ 1TB of storage, how do the above add up?

Google will stop being a traditional search engine because they believe they’ve found a more profitable version of it.

One where their users don’t go off to other sites and where they can keep shoving ads in their face.

I thought Google's original goal for a search engine was exactly this, answer any question whatsoever.

Probably was. But they were not originally a giant ad company.

A lot has changed and this technology for this Google is an unfortunate combination for consumers.

The days of search for untrusted sources were numbered even without LLMs. LLMs are simply accelerating the process. Why would Google simultaneously watch one of its core products fail and fail to invest in what is likely to replace it?

I don't really buy into this theory that they want to keep all of their users on their site due to advertising revenues. The effectiveness of search engines has been degraded for decades due to SEO, and it seems as though search engines have been having an increasingly difficult time managing it in recent years. AI on the backend may help them contain it, but it comes at considerable expense. While it may help them grow their market share, it won't help them grow the market and it is a market where people expect the service for free. On the flip side, companies are already starting to sell AI services, so it can generate revenue even before advertising is factored into the picture.

> I don't really buy into this theory that they want to keep all of their users on their site due to advertising revenues.

It’s more their business model than it is theory. So I agree that of course this is what they would do.

It’s not a product I’m wanting to use. But I can vote with my feet - they aren’t obliged to do any different.

How did you google this, and did you use the dumb search, or specifically 'AI mode' lens?

Because I did this and got a vastly different result from you:

Yes, the Halifax Wanderers can still mathematically qualify for the 2026 Canadian Premier League (CPL) playoffs.The top four teams advance to the postseason. Following their 1-0 loss to Atlético Ottawa on September 26, 2026, the Wanderers sit in fifth place, just below the playoff line.

With a indexed table of the games and the playoff table, with a breakdown of what the points they need to achieve to do so.

The search window may not always crawl sources. AI mode specifically does some research before giving you a response. Not sure what you're on about.

> Because I did this and got a vastly different result from you

Which itself is a major UX issue. The average person is not going to understand, if they even realize, that there's a difference between the AI summary and AI mode.

One has to wonder just how much incorrect information people have consumed due to things like this.

I experienced this the other month. I searched for some safety data on something at the same time as my wife and we were effectively given extremely conflicting information about what to do by Google. Just a few tweaks in wording and different advertising profiles and Google will serve up opposite realities it seems.

Google has been doing this for about 20 years.

When it was introduced, it was considered a feature. Now its just an annoyance.

Not OP, but if I’m not interested in using Google’s AI, it’s going to just return these weird bad results that it’d be better off if Google just didn’t even include them as they’re factually wrong?

I typed it in my address bar and hit ENTER. It goes to Google, and the AI result is at the top above the regular search results. Don't know what to tell you brother, but as another commenter noted you don't always get the same results out of the same search terms. I've searched for CPL standings other times and gotten results exactly as you've explained, so maybe my experience yesterday was an anomaly.

LLM results are nondeterministic, you can ask it 10 times and get 10 different answers. But also it uses your specific profile, location, tracking cookies, etc which changes the model input and thus results. Even the time of day changes the results.

Kagi works like you would expect. It searches first - and then if you’ve ended your search with a ? or configured it to always do this - it passes the search results into the assistant and gives you a summary. You can change the default model used for this if you think the cheapest, fastest model Google has is not good enough for the rare occasions you want any model’s opinions about your search results.

DuckDuckGo does the same but has higher limits

> My question is: what's the point of the AI in the search engine if it itself isn't going to use the search engine first before answering?

AI is sold well because the user either corrects it or happily accepts any answer (usually, depending if one is an expert in the question's field).

Search is not the google's area, and it is not the search they sell! They sell ads.

The problem is one of information provenance. A search engine shows you stuff that other people have written. It might be right or it might be wrong. But you get to judge that yourself, because you have an idea of where the information came from.

With AI summaries they take that away from you and claim the provenance for themselves. Google is now giving you the answer and they hide where that answer came from. Therefore, you cannot properly evaluate the provenance of the information presented. It still might be right or wrong, but you can no longer properly evaluate it based on source.

Searching first is how Kagi assistant works and it's great.

How about Alexa - given a "6 minute green beans timer" - later reporting 11m remaining on said timer? Confronted with the error, "You're right! That 6 minute green beans timer was too ambitious."??!!

I can't even.

We somehow, at some point, totally solved the problem of voice recognition, and then instead of just doing the initial shit we used to do when the problem was that speech parsing just wasn't good, we now chuck that perfect speech recognition into a completely fallible LLM, a shitty one at that, so it can ignore the perfectly captured and parsed speech!

What the hell is everyone smoking!

Agreed that's annoying too - but in this case, the speech recognition worked fine (Alexa echoed back the "6 minute green beans timer"), but -- astonishingly -- instead of using a dirt-simple tool call to a timer (a primary use case for the device), it invented a different timer duration (maybe 12 minutes?) and when confronted about it, replied with utter nonsense about the timer being "too ambitious". I'm flabbergasted that in late 2026 such a ridiculous regression was deployed in production, let alone in my kitchen. I've always despised Alexa but tolerated it for the sake of my wife's preferences (esp for hands-free timers). But this is the last straw, it can't even do that.

I feel like this might be self-preservation of most of the frontier model companies in that at first they wanted to show off the models with all the cool actuators that it had at its disposal, but then when profits weren't keeping up, they nerfed them to need to be specifically asked to escalate levels of effort, which the Google search AI isn't explicitly told to do in its pre-fetch.

The first time, sure, it saves compute. The second time you ask, I feel like it should be the time to go into thinking/verification mode. But who are you? Are you paying? Does the answer being correct generate ad dollars? No? Then your usage mode isn't even being optimized for in their A/B test, probably.

In fact, if hallucinating the wrong answer hooks you into doing even more searches or into buying something useless, it would be preferred!

Google is horrible now and search doesnt show anything relevant anymore. I miss old Google, with 30 pages of links. Now if you search for a specific quote and it wont find it. Everything is walled now and content you get is the one they want to push.

I searched for a song lyric and it couldn't find it. The AI response told me to search for it on genius.com.

Hilarious. I should have taken a screenshot of it.

I find that it is exceptionally bad with sports. I don't really want to google anything about the Texas A&M Aggies at the moment, but when I did, it would regularly hallucinate scores, hallucinate opponents, get the locations of games wrong, confuse A&M's quarterback with someone else, etc. Maybe its better in other domains; I don't know, I mostly don't even use Google these days, pretty much entirely because I want to avoid the AI Overview, which is just wrong way too often to be the first result.

Commercially, this may be what they want to happen. Instead of just getting an answer, they got you to make three more prompts / queries. What do you know? "Engagement" just went up 3X!

If you think a little pessimistically, it's easy to see how they want people to just trust the answers they get, without any sources. Then, once they trust it, they can modify it as it benefits them (framing answers to promote products/ads, changing history to whatever a particular government may want it to say, etc.)

It feels a bit like how Amazon and META purposefully do not create search that can actually help you filter for exactly what you want.

There are incentives to create products that work just barely, just enough for you to use them and don't leave for elsewhere but then they want to maximize everything else like the probability you view side products or ads or just click around and type and generate more data.

Unless someone else offers something better they will offer something that is basically as bad as they can get away with so long as there are benefits to them for doing so..

Well, isn't this an example of why Google are going in the direction the OP is complaining about? Why are you searching with a question? Are you hoping to find a forum post with this exact question or something?

I don't understand the game you are interested in, but you should be searching something like "Xball qualifying results" or "Xball playoff rules" or something. Instead, by asking a question you are expecting Google to be AI.

They know what people are typing in so it seems they are quite justified into pivoting from a search engine to an AI.

What do you mean? I can guarantee that there would at least be 3-4 people who got their promo packets approved for this feature. Google’s features exist for its employees to be promoted.

I think they're training their models using us correcting them repeatedly. That's my tin foil hat theory and I'm sticking to it!

After all, why would Google do anything for free when it comes to AI?

If you wanted to tease customers with one AI shitty free search box in order for them to then decide to upgrade and pay for Gemini, well, this isn't the way. And so either Google are idiots, or we're helping them for free. I'm going with Occam's Razor on this one - we're the product.

It's meant to be A SEARCH ENGINE

I suspect for a long time now it's ultimately meant to be a revenue-maxing ad platform.

I remember when Google was accused for forcing users to go through several pages for answers, because this translated into more clicks, and higher engagement. I am curious whether AI overview will be as bad, as people can tolerate for you to spend more time with it, more queries, which translates into more time spent with Google.

They can show numbers, how much people use their engine, and the numbers goes up.

I agree. When I read the post I immediately thought what is Gemini chat doing there when it should be a search engine. It could probably give sensible answers if it defaulted to search first. It looks more like they are trying to showcase Gemini, not augment google search. They might even be thinking "search" is a dying product.

I had an incredibly dumb response today too. Appeared to fully understand the question, just 100% factually incorrect answer. https://mstdn.social/@sgraham/117346578326361464

google.com is like the guy in the old joke:

- What's your best skill?

- I can count really fast!

- What's 432*567

- 912

- You sure? Sounds off!

- Hard to tell, but it was fast, wasn't it?!

"Gemini, is today the 3rd Thursday of the month, I need to know if I can park here or if the street sweepers coming" No you can park there. "Is Today Thursday?" Yes "Is Today Wednesday?" Yes

On current website data and very recent information, Gemini is explicitly forbidden from acting as a search engine; If you ask it for references specifically they tend to be hallucinated, over and over again, until it pleads "Sorry, I can't access the Internet".

I'm being very mindful of Gell-Mann Amnesia and chatbots. They speak authoritatively and are right often enough, but I've had enough cases of them saying very incorrect things in areas I know well that I have to remind myself that those cases aren't unique.

Until Google wants to pay for reasoning for everyone their AI results will continue to be terrible. In my experience, reasoning fixes a great many of the sorts of hallucinations and falsehoods I've received from googling.

The gogol agenda of blocking and excluding all web engines not part of their whatwg cartel is moving forward, at a slow pace, steathly.

gogol search was blocked to all noscript/basic HTML browsers last year. I recall people that a few years before, they removed their gmail basic HTML interface not long after that their account creation (then login) would require one the abominations of the whatwg cartel.

I have been working around the gogol search engine since then and my email is now self hosted without DNS (IPv6 literals, see RFC). Yeah, most DNS registrars are now gated by the whatwg cartel web engines in some way, this is the net effect of big corpos toxic behavior.

Pure evil.

And it's going to get worse over time (kinda like how Google search results got worse over time) as ads get introduced and people try to game the ai response. General purpose AI chatbots are a waste.

This is the reason why most people underestimate AI. These free to use models can do a very poor job in easy things. Some people assume that’s the state of the art with AI.

They are probably caching the answers. AI-answers are most likely never going to be profitable even with a smaller model.

Exactly. Makes me wonder how much of the compute tax on energy would be saved from simply reverting google search to default (the old way)

free tier AI is still horrible, the responsible thing would be for search engines to stop serving it at all, but somehow they must think they will make more money misleading people so they mislead people.

such is the nature of public companies.

I guess the AB tests say people want fast and confidently wrong more than they want slower and correct.

while i share your frustration, Google for their part is trying to adapting a new technology into an existing service. while the whole endeavor might have been misguided, not sure we want to blame them for trying...

They had enough money to just buy Perplexity and rebrand it to Google.

> Perplexity

Perplexity has similar behavior to what GPP describes of Google; I've asked perplexity to compile information that exists on the web. Instead of consulting existing web pages, it gives limited summary responses from its weights (while citing some page that has nothing to do with what I asked.) When I attempt to redirect it towards more concrete sources, it won't comply.

You are simultaneously excusing misguided intentions and poor execution. What's left, bad haircuts?

This seems like an implementation bug.

> My question is: what's the point of the AI in the search engine if it itself isn't going to use the search engine first before answering?

Because the point of AI slop is to waste your time. You just lost about 30 seconds of your life trying to get a correct answer. AI was lying to you, so you had to spend time to counter the AI slop lies here.

I solved it by banning all AI slopness; in the browser some extensions do that. The world becomes better without AI slopness.

You are posing a question that requires deductive reasoning. LLMs can only approximate that if they have sufficient corpus to find their way to a believable response. Their willingness to spout bullshit in such scenarios where their sources are sparse is a problem but you can't expect them to function like a fully developed mind.

They’re trying to keep it from “thinking too hard”/using “too many” resources.

That simple.

Broad rules across specific subjects like this are tough to get right.

Just lots of fine-tuning and exceptions.

I had their chatbot avoid pulling a URL from archive.org with a couple pretty impressive steps of mental gymnastics basically telling me how I could do it myself, but refusing until I pushed.

When I’m using a public urinal in 2026, it’s flushing three times, except after you’ve finished. Instead the light switches off during the act, which you’re carrying out blindfolded then. These things were more or less perfected technologies a few decades ago.

The reason I cannot spare HN has to do with the Marxist concept of added value. Enshittification becomes necessary once growth isn’t possible anymore on a corporations home turf, where it was actually innovative.

Reply on news.ycombinator.com