Death by AI
davebarry.substack.com
davebarry.substack.com
Vibrant watering hole with drinks & po' boys, as well as a jukebox, pool & electronic darts.
It doesn't serve po' boys, have a jukebox (though the playlists are impeccable), have pool, or have electronic darts. (It also doesn't really have drinks in the way this implies. It's got beer and a few canned options. No cocktails or mixed drinks.)
They got a catty one-star review a month ago for having a misleading description by someone who really wanted to play pool or darts.
I'm sure the owner reported it. I reported it. I imagine other visitors have as well. At least a month on, it's still there.
This happens all the time on automotive forums/FB groups and it's a huge problem.
I should probably admire the AI for showing a lot of restraint on its first steps to global domination and/or wiping out humanity.
I had to fight a similar battle with Google Maps, which most people believe to be a source of truth, and it took years until incorrect information was changed. I'm not even sure if it was because of all the feedback I provided.
I see Google as a firehose of information that they spit at me ("feed"), they are too big to be concerned about any inconsistencies, as these don't hurt their business model.
Now, frequently, the AI summaries are on top. The AI summary LLM is clearly a very fast, very dumb LLM that’s cheap enough to run on webpage text for every search result.
That was a product decision, and a very bad one. Currently a search for "Suicide Squad" yields
> The phrase "suide side squad" appears to be a misspelling of "Suicide Squad"
I don't know that it's a bad decision, time will judge it. Also, we can expect the quality of the results to improve over time. I think Google saw a real threat to their search business and had to respond.
Millions, probably tens of millions of people have jobs trying to manipulate search results - with billions of dollars of resources available to them. With no internal information, it's safe to say no more than thousands of Googlers (probably fewer) are working to combat them.
If every one of them is a 10x engineer they're still outnumbered by more than 2 orders of magnitude.
https://www.wheresyoured.at/the-men-who-killed-google/
> The key event in the piece is a “Code Yellow” crisis declared in 2019 by Google’s ads and finance teams, which had forecast a disappointing quarter. In response, Raghavan pushed Ben Gomes — the erstwhile head of Google Search, and a genuine pioneer in search technology — to increase the number of queries people made by any means necessary.
(Quoting from this follow-up post: https://www.wheresyoured.at/requiem-for-raghavan/)
I don't think they are; they have realised (quite accurately, IMO) that users would still use them even if they boosted their customers' rankings in the results.
They could, right now, switch to a model that penalises pages for each ad. They don't. They could, right now, penalise highly monetised "content" like courses and crap. They don't do that either.[1]
If Kagi can get better results with a fraction of the resources, there is no argument to be made that Google is playing a losing game.
--------------------------------------
[1] All the SEO stuff is damn easy to pick out; any page that is heavily monetised (by ads, or similar commercial offering) is very very easy to bin. A simple "don't show courses unless search query contains the word courses" type of rule is nowhere near computationally expensive. Recording the number of ads on a page when crawling is equally cheap.
It’s nowhere near good either. What about the searches for cuorses or classes or training?
Why would they drop that? It's not as if they have to throw away all the preprocessing they do on the search query.
They can continue preprocessing exactly like they do it now.
Google's algorithm is the target for every SEO firm in the world. No one is targeting Kagi. Therefore, Kagi can use techniques that would not work at Google.
Meanwhile, every single person I know has come to trust Google less. That will catch up with them eventually.
Surely there is a way to correct it: getting the issue on the front page of HN.
I have tried reporting these fake paths in the past but it didn’t get them removed.
https://www.waze.com/live-map/directions/us/va/arlington/alc...
Well, in this case the inaccurate information is shown because the AI overview is combining information about two different people, rather than the sources being wrong. With traditional search, any webpages would be talking about one of the two people and contain only information about them. Thus, I'd say that this problem is specific to the AI overview.
> Dave Barry, the humorist, experienced a brief "death" in an AI overview, which was later corrected. According to Dave Barry's Substack, the AI initially reported him as deceased, then alive, then dead again, and finally alive once more. This incident highlights the unreliability of AI for factual information.
That wouldn't solve the problem of mixing up multiple people. But the first problem most people have is probably actually that it pulls up a person that is more famous than who they were actually looking for.
I think Google does have some type of knowledge graph. I wonder how much AI model uses it.
Maybe it hits the graph, but also some kind of Google search, and then the LLM is like Gemini Flash Lite and is not smart enough to realize which search result goes with the famous person from the graph versus just random info from search results.
I imagine for a lot of names, there are different levels of fame and especially in different categories.
It makes me realize that my knowledge graph application may eventually have an issue with using first and last name as entity IDs. Although it is supposed to be for just an individual's personal info so I can probably mostly get away with it. But I already see a different issue when analyzing emails where my different screen names are not easily recognized as being the same person.
That is such a classic problem with Google (from long before AI).
I am not optimistic about anything being changed from this, but hope springs eternal.
Also, I think the trilobite is cute. I have a [real fossilized] one on my desk. My friend stuck a pair of glasses on it, because I'm an old dinosaur, but he wanted to go back even further.
The site structure is also fairly prehistoric!
"a measurement of the extent to which we realize that we are trapped in a world almost totally devoid of reason. Laughter is how we express the anxiety we feel at this knowledge"
https://www.politico.eu/article/dutch-scandal-serves-as-a-wa...
> In 2019 it was revealed that the Dutch tax authorities had used a self-learning algorithm to create risk profiles in an effort to spot child care benefits fraud.
This was a pre-LLM AI, but expected "hilarity" ensues: broken families, foster homes, bankruptcies, suicides.
> In addition to the penalty announced April 12, the Dutch data protection agency also fined the Dutch tax administration €2.75 million in December 2021.
The government fining itself is always such a boss move. Heads I win, tails you lose.
I also hope that the AI and Google duders understand that this is most people's experience with their products these days. They don't work, and they twist reality in ways that older methods didn't (couldn't, because of the procedural guardrails and direct human input and such). And no amount of spin is going to change this perception - of the stochastic parrots being fundamentally flawed - until they're... you know... not. The sentiment management campaigns aren't that strong just yet.
So did I, except I'm probably from an earlier generation. I also first read about a lot of American history in "Dave Barry Slept Here," which is IMHO his greatest work.
Too much to ask, surely.
I know it's the HN darling and is probably talked about too much already but it doesn't have this problem. The only AI stuff is if you specifically ask for it which in your case would be never. And unlike Google where you are at the whims of the algorithm you can punish (or just block) AI garbage sites that SEO their way into the organic results. And a global toggle to block AI images.
I can see a bright future in blaming things on AI that have nothing to do with AI, at least on here.
Many things have changed since then.
How does the police force distinguish between a map route and people randomly bumbling there? Were there signs that were ignored?
I got a ticket that way once when I was visiting because I only knew how to get back to my hotel from the airport so I drove to the airport then to the hotel-- and I guess the police watch for people looping through the airport to avoid the tolls. In my case I wasn't aware of the weird toll/no-toll thing-- I was just lost and more concerned with finding my hotel than the posted 'no through traffic' signs.
Later, after moving to VA, I noticed google maps was explicitly routing trips from near the airport to other places to take a loop through the airport to minimize toll costs which would have been quite clever if it weren't prohibited.
Whatever method they use to update their data is broken, or they do not care about countries our size enough to make sure it is reasonably correct and up-to-date.
Sounds like you need to report it at your municipality or whatever local gov is responsible for keeping their GIS up to date.
- The "Monthly" option is selected by default.
- If you click "Yearly", it tells you the actual full yearly price without dividing it by 12.
That's so rare and refreshing that I'm tempted to sign up just out of respect.
The other concern I saw is that they might deliver pro-Russia propaganda. If that happens, I'll trust Kagi to firewall them appropriately. Google also intentionally delivers geopolitical propaganda.
https://medium.com/luminasticity/argument-ai-and-defamation-...
But I suppose that is better than outright making stuff up.
Be ideal if it did disambiguate a la Wikipedia.
That's why this Dave Barry has a right. It's a subsection.
It'd be like opening Dave Barry (comedian) on Wikipedia and halfway through the article in a subsection it starts detailing the death of a different Dave Barry.
The versions with "Dave Barry, the humorist and Pulitzer Price winner, passed away last November 20…" and "Dave Barry, a Bostonian … died on November 20th…" are also rather unambiguous regarding who this might be about. The point being, even if the meaning of the particular identity of the subject is moved outside to an embedding context, it is still crucial for the meaning of these utterances.
That's deeply concerning, especially when these two companies control almost all the content we access through their search engines, browsers and LLMs.
This needs to be regulated. These companies should be held accountable for spreading false information or rumours, as it can have unexpected consequences.
Regulated how? Held accountable how? If we start fining LLM operators for pieces of incorrect information you might as well stop serving the LLM to that country.
> since it can have unexpected consequences
Generally you hold the person who takes action accountable. Claiming an LLM told you bad information isn’t any more of a defense than claiming you saw the bad information on a Tweet or Reddit comment. The person taking action and causing the consequences has ownership of their actions.
I recall the same hand-wringing over early search engines: There was a debate about search engines indexing bad information and calls for holding them accountable for indexing incorrect results. Same reasoning: There could be consequences. The outrage died out as people realize they were tools to be used with caution, not fact-checked and carefully curated encyclopedias.
> I'm worried about this. Companies like Wikipedia spent years trying to get things right,
Would you also endorse the same regulations against Wikipedia? Wikipedia gets fined every time incorrect information is found on the website?
EDIT: Parent comment was edited while I was replying to add the comment about outside of the US. I welcome some country to try regulating LLMs to hold them accountable for inaccurate results so we have some precedent for how bad of an idea that would be and how much the citizens would switch to using VPNs to access the LLM providers that are turned off for their country in response.
Other companies have been fined for misleading customers [0] after a product launch. So why make an exception for Big Tech outside the US?
And why is the EU the only bloc actively fining US Big Tech? We need China, Asia and South America to follow their lead.
[0] https://en.m.wikipedia.org/wiki/Volkswagen_emissions_scandal
But how do we know they're telling the truth? How do we know it wasn't intentional? And more importantly, who's held accountable?
While Google's AI made the mistake, Google deployed it, branded it, and controls it. If this kind of error causes harm (like defamation, reputational damage, or interference in public opinion), intent doesn't necessarily matter in terms of accountability.
So while it's not illegal to be wrong, the scale and influence of Big Tech means they can't hide behind "it was the AI, not us."
sounds good to me?
Fines, when backed by strong regulation, can lead to more control and better quality information, but only if companies are actually held to account.
The organization that runs the website, the Wikimedia Foundation, is also not a company. It's a nonprofit.
And the Wikimedia Foundation have not “spent years trying to get things right”, assuming you're referring to facts posted on Wikipedia. That was in fact a bunch of unpaid volunteer contributors, many of whom anonymous and almost all of whom unaffiliated with the Wikimedia Foundation.
You can't really argue with those facts.
Did they ? Lots of people, and some research verify this, think it has a major left leaning bias, so while usually not making up any facts editors still cherry pick whatever facts fit the narrative and leave all else aside.
It's a bigger problem than AI errors imo, there are so many Wikipedia articles that are heavily biased. A.I makes up silly nonsense maybe once in 200 queries, not 20% of the time. Also, people perhaps are more careful and skeptical with A.I results but take Wikipedia as a source of truth.
A German 90s/2000s rapper (Textor, MC of Kinderzimmer Productions) produced a radio feature about facts and how hard it can be to prove them.
One personal example he added was about his Wikipedia Article that stated that his mother used to be a famous jazz singer in her birth country Sweden. Except she never was. The story had been added to an Album recension in a rap magazine years before the article was written. Textor explains that this is part of 'realness' in rap, which has little to do with facts and more with attitude.
When they approached Wikipedia Germany, it was very difficult to change this 'fact' about the biography of his mother. There was published information about her in a newspaper and she could not immediately prove who she was. Unfortunately, Textor didn't finish the story and moved on to the next topic in the radio feature.
https://en.wikipedia.org/wiki/Meg_Tilly is my sister. It claims that she is of Irish descent. She is not. The Irish was her stepfather (my father), and some reporter confusing information about a stepparent with information about a parent.
Now some school in Seattle is claiming that she is an alumnus. That's also false. After moving from Texada, she went to https://en.wikipedia.org/wiki/Belmont_Secondary_School and then https://esquimalt.sd61.bc.ca/.
But for all that, Wikipedia reporting does average out to more accurate than most newspaper articles...
Yes, that's exactly what AI is.
100% with you.
LLM is good enough i believe. No need to invent anything new.
I don’t know if this is a fundamental problem with the llm architecture or a problem with proper prompts.
For it to be a social network there should be a way for me to indicate that I want to hear either more or less of you specifically, and yet HN is specifically designed to be more about ideas than about people.
Discovery is comparatively harder - search has been dominated by noise. Word of mouth still works however, and is better than before - there are more people actively engaged in curating catalogues, like "awesome X" or <https://kagi.com/smallweb/>.
Most of it is also at little risk of being "eaten", because the infrastructure on which it is built is still a lot like the "old" Internet - very few single points of failure[1]. Even Kagi's "Small Web" is a Github repository (and being such, you can easily mirror it).
[1]: Two such PoFs are DNS, and cloudflarization (no thanks to the aggressive bots). Unfortunately, CloudFlare also requires you to host your DNS there, so switching away is double-tricky.
It's interesting that LLMs produce each output token as probabilities but it appears that in order to generate the next token (which is itself expressed as a probability), it has to pick a specific word as the last token. It can't just build more probabilities on top of previous probabilities. It has to collapse the previous token probabilities as it goes?
You can also see decision paralysis in action if you implement CoT - it's common to see the model "pondering" about a bunch of possible options before picking one.
In the spirit of social activism, I will take it upon myself to name all of my children Google, even the ones that already have names.
I mean, yes, but it's worse than that - the machine has no idea what a "name" is, how they relate to singleton humans, what a human is, or that "Dave Barry" is one of them (name OR human). It's all just strings of tokens.
“So for now we probably should use it only for tasks where facts are not important, such as writing letters of recommendation and formulating government policy.”
:-)
Clearly the Mandela Effect needed nukes. Clearly.
There are already real teeth to the whims of what corporations do with you.
Seems to be another Dave Barry who was a political activist that passed away in 2016
This is how "talking to AI" feels like for anything mildly complex.
Or elect them President.
The whole point of regulation is for when the profit motive forces companies towards destructive ends for the majority of society. The companies are legally obligated to seek profit above all else, absent regulation.
What regulation? What enforcement?
These terms are useless without details. Are we going to fine LLM providers every time their output is wrong? That’s the kind of proposition that sounds good as a passing angry comment but obviously has zero chance of becoming a real regulation.
Any country who instituted a regulation like that would see all of the LLM advancements and research instantly leave and move to other countries. People who use LLMs would sign up for VPNs and carry on with their lives.
Enforcement ensures accountability.
Fines don't do much in a fiat money-printing environment.
Enforcement is accountability, the kind that stakeholders pay attention to.
Something appropriate would be where if AI was used in a safety-critical or life-sustaining environment and harm or loss was caused; those who chose to use it are guilty until they prove they are innocent I think would be sufficient, not just civil but also criminal; where that person and decision must be documented ahead of time.
> Any country who instituted a regulation like that would see all of the LLM advances and research instantly leave and move to other countries.
This is fallacy. Its a spectrum, research would still occur, it would be tempered by the law and accountability, instead of the wild-west where its much more profitable to destroy everything through chaos. Chaos is quite profitable until it spread systemically and ends everything.
AI integration at a point where it can impact the operation of nuclear power plants through interference (perceptual or otherwise) is just asking for a short path to extinction.
Its quite reasonable that the needs for national security trump private business making profit in a destructive way.
Would this guilty-until-proven-innocent rule apply also to non-ML code and manual decisions? If not, I feel it's kind of arbitrarily deterring certain approaches potentially at the cost of safety ("sure this CNN blows traditional methods out of the water in terms of accuracy, but the legal risk isn't worth it").
In most cases I think it'd make more sense to have fines and incentives for above-average and below-average incident rates (and liability for negligence in the worse cases), then let methods win/fail on their own merit.
I would say yes because the person deciding must be the one making the entire decision but there are many examples where someone might be paid to just rubberstamp decisions already made. Letting the person who decided to implement the solution off scot-free.
The mere presence of AI (anything based on underlying work of perceptrons) being used accompanied by a loss should prompt a thorough review which corporations currently are incapable of performing for themselves due to lack of consequences/accountability. Lack of disclosure, and the limits of current standing, is another issue that really requires this approach.
The problem of fines is that they don't provide the needed incentives to large entities as a result of money-printing through debt-issuance, or indirectly through government contracts. Its also far easier to employ corruption to work around the fine later for these entities as market leaders. We've seen this a number of times in various markets/sectors like JPM and the 10+ year silver price fixing scandal.
Merit of subjective rates isn't something that can be enforced, because it is so easily manipulated. Gross negligence already exists and occurs frighteningly common but never makes it to court because proof often requires showing standing to get discovery which isn't generally granted absent a smoking gun or the whim of a judge.
Bad things happen certainly where no one is at fault, but most business structure today is given far too much lee-way and have promoted the 3Ds. Its all about: deny, defend, depose.
> I would say yes [...]
So if you're a doctor making manual decisions about how to treat a patient, and some harm/loss occurs, you'd be criminally guilty-until-proven-innocent? I feel it should require evidence of negligence (or malice), and be done under standard innocent-until-proven-guilty rules.
> The mere presence of AI (anything based on underlying work of perceptrons) [...]
Why single out based on underlying technology? If for instance we're choosing a tumor detector, I'd claim what's relevant is "Method A has been tested to achieve 95% AUROC, method B has been tested to achieve 90% AUROC" - there shouldn't be an extra burden in the way of choosing method A.
And it may well be that the perceptron-based method is the one with lower AUROC - just that it should then be discouraged because it's worse than the other methods, not because a special case puts it at a unique legal disadvantage even when safer.
> The problem of fines is that they don't provide the needed incentives to large entities as a result of money-printing through debt-issuance, or indirectly through government contracts.
Large enough fines/rewards should provide large enough incentive (and there would still be liability for criminal negligence where there is sufficient evidence of criminal negligence). Those government contracts can also be conditioned on meeting certain safety standards.
> Merit of subjective rates isn't something that can be enforced
We can/do measure things like incident rates, and have government agencies that perform/require safety testing and can block products from market. Not always perfect, but seems better to me than the company just picking a scape-goat.
Yes, that proof is called a professional license, without that you are presumed guilty even if nothing goes wrong.
If we have licenses for AI and then require proof that the AI isn't tampered with for requests then that should be enough, don't you think? But currently its the wild west.
A professional license is evidence against the offense of practicing without a license, and the burden of proof in such a case still rests on the prosecution to prove beyond reasonable doubt that you did practice without a license - you aren't presumed guilty.
Separately, what trod1234 was suggesting was being guilty-until-proven-innocent when harm occurs (with no indication that it'd only apply to licensed professions). I believe that's unjust, and that the suggestion stemmed mostly from animosity towards AI (maybe similar to "nurses administering vaccines should be liable for every side-effect") without consideration of impact.
> If we have licenses for AI and then require proof that the AI isn't tampered with for requests then that should be enough, don't you think?
Mandatory safety testing for safety-critical applications makes sense (and already occurs). It shouldn't be some rule specific to AI - I want to know that it performs adequately regardless of whether it's AI or a traditional algorithm or slime molds.
I don't think so. We read about the handful of failures while there are billions of successful queries every day, in fact I think AI Overviews is sticky and here to stay.
"...Semantic Errors / Hallucinations On factual queries—especially legal ones—models hallucininate roughly 58–88% of the time
A journalism‑focused study found LLM-based search tools (e.g., ChatGPT Search, Perplexity, Grok) were incorrect in 60%+ of news‑related queries
Specialized legal AI tools (e.g., Lexis+, Westlaw) still showed error rates between 17% and 34%, despite being domain‑tuned "
Reminds me of the toaster in Red Dwarf
I mean, when you give an LLM good input, it seems to have a good chance of creating a good result. However, when you ask an LLM to retrieve facts, it often fails. And when you look at the inner workings of an LLMs that should not surprise us. After all, they are designed to apply logical relationships between input nodes. However, this is more akin to applying broad concepts than recalling detailed facts.
So if you want LLMs to succeed with their task, provide them with the knowledge they need for their task (or at least the tools to obtain the knowledge themself).
it's neither, really.
> After all, they are designed to apply logical relationships between input nodes
They are absolutelly not. Unless you assert that logical === statistical (which it isn't)
For clarification: yes, when I wrote 'logical,' I did not mean Boolean logic, but rather something like probabilistic/statistical logic.
Yet another argument for switching to DuckDuckGo
The man is literally responding to what happens when you Google the name. It's displaying his picture, most of the information is about him. He didn't put it there or ask for it to be put there.
Personally, if I got a resurrection from it, I would accept the nudge and do the political activism in Dorchester.