Feels like judge was looking for any excuse not to apply harsh penalty and since Google brought up AI as competitor, the judge accepted it as acceptable excuse for very minor penalty.
You might think "but ChatGPT isn't a search engine", and that's true. It can't handle all queries you might use a search engine for, e.g. if you want to find a particular website. But there are many many queries that it can handle. Here's just a few from my recent history:
* How do I load a shared library and call a function from it with VCS? [Kind of surprising it got the answer to this given how locked down the documentation is.]
* In a PAM config what do they keywords auth, account, password, session, and also required/sufficient mean?
* What do you call the thing that car roof bars attach to? The thing that goes front to back?
* How do I right-pad a string with spaces using printf?
These are all things I would have gone to Google for before, but ChatGPT gives a better overall experience now.
Yes, overall, because while it bullshits sometimes, it also cuts to the chase a lot more. And no ads for now! (Btw, someone gave me the hint to set its personality mode to "Robot", and that really helps make it less annoying!)
I swear in the past week alone things that would've taken me weeks to do are taking hours. Some examples: create a map with some callouts on it based on a pre-existing design (I literally would've needed several hours of professional or at least solid amateur design work to do this in the past; took 10 minutes with ChatGPT). Figure out how much a rooftop solar system's output would be compromised based on the shading of a roof at a specific address at different times of the day (a task I literally couldn't have completed on my own). Structural load calculations for a post in a house (another one I couldn't have completed on my own). Note some of these things can't be wrong so of course you can't blindly rely on ChatGPT, but every step of the way I'm actually taking any suspicious-sounding ChatGPT output and (ironically I guess) running keyword searches on Google to make sure I understand what exactly ChatGPT is saying. But we're talking orders of magnitude less time, less searching and less cost to do these things.
Edit: not to say that the judge's ruling in this case is right. Just saying that I have zero doubt that LLM's are an existential threat to Google Search regardless of what Google's numbers said during their past earnings call.
You're relying on ChatGPT for this? How do you check the result? That sounds kind of dangerous...
That said, the word "relying" is taking it too far. I'm relying on myself to be able to vet what ChatGPT is telling me. And the great thing about ChatGPT and Gemini, at least the way I prompt, is that it gives me the entire path it took to get to the answer. So when it presents a "fact," in this example a load calculation or the relative strength of a wood species, for instance, I take the details of that, look it up on Google and make sure that the info it presented is accurate. If you ask yourself "how's that saving you time?" The answer is, in the past, I would've had to hire an engineer to get me the answer because I wouldn't even quite be sure how to get the answer. It's like the LLM is a thought partner that fills the gap in my ability to properly think about a problem, and then helps me understand and eventually solve the problem.
Btw, I would not trust an LLM to tell me how to build a suspension bridge. First, I'm unfamiliar with that space. Second, even if I was familiar, the stakes are, as you say, so high that it would be insane to trust something so complex without expert sign off. The post I'm specifically talking about? Near-zero stakes and near-zero risk.
<stepping on the soapbox> I beg folks to always try and pierce the veil of complexity. Some things are complex and require very specialized training and guardrails. But other complexity is fabricated. There are entrenched interests who want you to feel like you can't do certain things. They're not malicious, but they sometimes exist to make or protect money. There are entire industries propped up by trade groups that are there to make it seem like some things are too complex to be done by laypeople, who have lobbied legislators for regulations that keep folks like you from tackling them. And if your knee-jerk reply is that I'm some kind of conspiracy theorist or anarchist all I'm saying is it's a spectrum. Suspension bridge with traffic driving over it --> should double, triple, quadruple check with professional(s); a post in a house supporting the entire house's load (exaggeration for effect) --> get a single professional to sign off; a post in a house that's supporting a single floor joist with minimal live and dead load (my case!) --> use an LLM to help you DIY the "engineering" to get to good enough (good enough = high margin for error); replace a light switch --> DIY YouTube video.
I am the king of long-winded HN posts. Obviously the time I took to write this (look, ma, no LLM!) is asymmetric with what you wrote, but I'm genuinely wondering if any of this makes you think differently. If not, that's cool of course (and great for the engineers and permit issuers!).
The reason you hire a structural engineer is because they do - and they are on the hook if it goes wrong. Which is also why they have to stamp drawings, etc.
Because the next person who owns the house should have some idea who was screwing with the structure of it.
You might be 100% on top of it - in which case that structural engineer should have no problem stamping your calcs eh?
The only other thing I'll add is the ideal vs. the reality. What percent of structural projects done to single-family construction, in particular, do you think is done by engineers? I would guess it's far less than 50%. That's based on my own experience working in the industry, which I know you won't trust (why would you? Random internet guy after all). But for conversation's sake suffice it to say that I believe every time you walk into a house that's several decades old or older you're likely walking into a place that has been manipulated structurally without an engineer's stamp. And the vast majority (99%+ of the time) it's perfectly safe to be in that space.
Everyone thinks they are the exception. Occasionally, one of them is even right, eh?
And just to clarify I don't think I'm the exception. I was actually making the opposite argument. Almost anyone can and should attempt to deconstruct complexity because doing things is not always as difficult as it would seem (or as difficult as we've been told).
Appreciate the dialogue, lazide!
It isn’t due to ‘complexity’ either - rather indifference, laziness, or just plain stupidity.
I’ve seen people almost burn down their places multiple times - and at least one family actually die from an electrical fire. Also, partial building collapses.
The reason you don’t see it more often is because people generally don’t actually try.
Vetting things is very likely harder than doing the thing correctly.
Especially the thing you are vetting is designed to look correct more than actually being correct.
You can picture a physics class where teacher gives a trick problem/solution and 95% of class doesn’t realize until the teacher walks back and explains it.
This might not quite be true, strictly speaking, but a very similar statement definitely is. LLMs are highly prone to hallucinations, a term you've probably heard a lot in this context. One reason for this is that they are trained to predict the next word in a sequence. In this game, it's almost always better to guess than to output 'I'm not sure,' when you might be wrong. LLMs therefore don't really build up a model of the limits of their own 'knowledge,' they just guess until their guesses get better.
These hallucinations are often hard to catch, in part because the LLM will sound confident regardless of whether it is hallucinating or not. It's this tendency that makes me nervous about your use case. I asked an LLM about world energy consumption recently, and when it couldn't find an answer online in the units I asked for, it just gave a number from a website and changed (not converted) the units. I almost missed it, because the source website had the number!
Stepping back, I actually agree that you can learn new things like this from LLMs, but you either need to be able to verify the output or the stakes need to be low enough that it doesn't matter if you can't. In this case, even if you can verify the math, can you be sure that it's doing the right calculation in the right way? Did it point out the common mistakes that beginners make? Did it notice that you're attaching the support beam incorrectly?
Chances are, you've built everything correctly and it will be fine. But the chances of a mistake are clearly much higher than if you talked to an experienced human (professional or otherwise).
...but ironically that chatbot is Gemini from ai studio, so still the same company but a different product. Google search will look very different in the next 5-10 years compared to the same period a decade ago.
Kimi K2's output style is something like a mix of Cynic and Robot as seen here https://help.openai.com/en/articles/11899719-customizing-you... and I absolutely love it. I think more people should give it a try (kimi.com).
It's going to be a real problem going forward, because if AI hadn't killed them something else would have, and now it's questionable whether that "something else" will ever emerge. The need for something like SO is never going to go away as long as new technologies, algorithms, languages and libraries continue to be created.
However, your point stands: as new technologies develop, StackOverflow will be the main platform where relevant questions gain visibility through upvotes.
Or closing a general question because in the opinion of Someone Important, it runs afoul of some poorly-defined rule regarding product recommendations.
A StackOverflow that wasn't run like a stereotypical HOA would be very useful. The goal should be to complement AI rather than compete with it.
Think of programming languages as you currently think of CPU ISAs. We only need so many of those. And at this point, machine-instruction architecture has diverged so far from traditional ISAs that it no longer gets called that. Instead of x86 and ARM and RISC-V we talk about PTX and SASS and RDNA. Or rather, hardly anyone talks about them, because the interesting stuff happens at a higher level of abstraction.
People say all the time that LLMs are so much better for finding information, but to me it's completely at odds with my own user experience.
I'm not interested in dissecting specific examples because never been productive, but I will say that most people's bullshit detectors are not nearly as sensitive as they think they are which leads them to accepting sloppy incorrect answers as high-quality factual answers.
Many of them fall into the category of "conventional wisdom that's absolutely wrong". Quick but sloppy answers are okay if you're okay with them, after all we didn't always have high-quality information at our fingertips.
The only thing that worries me is how really smart people can consume this slop and somehow believe it to be high-quality information, and present it as such to other impressionable people.
Your success will of course vary depending on the topic and difficulty of your questions, but if you "can't remember" the last time you had a BS answer then I feel extremely confident in saying that your BS detector isn't sensitive enough.
Do you have a few examples? I'm curious because I have a very sensitive BS detector. In fact, just about anyone asking for examples, like the GP, has a sensitive BS detector.
I want to compare the complexity of my questions to the complexity of yours. Here's my most recent one, the answer to which I am fully capable of determining the level of BS:
I want to parse markdown into a structure. Leaving aside the actual structure for now, give me a exhaustive list of markdown syntax that I would need to parse.
I responded with:
I am seeing some problems here with the parsing: 1. Newlines are significant in some places but not others. 2. There are some ambiguities (for example, nested lists which may result in more than four spaces at the deepest level can be interpreted as either nested lists or a code block) 3. Autolinks are also ambiguous - how can we know that the tag is an autolink and not HTML which must be passed through? There are more issues. Please expand on how they must be resolved. How do current parsers resolve the issues?
That's also what Grok says:
> Is there a constant time string comparison algorithm in the JRE?
No, the Java Runtime Environment (JRE) does not provide a built-in constant-time comparison method directly for String objects in the standard library. The String.equals() and String.equalsIgnoreCase() methods perform variable-time comparisons that short-circuit upon finding the first mismatch, which can be vulnerable to timing attacks in security-sensitive contexts like password or token verification. However, the JRE does include a constant-time comparison utility for byte arrays via java.security.MessageDigest.isEqual(byte[] digesta, byte[] digestb), introduced in Java 6 and documented as performing a comparison where "the calculation time depends only on the length of digesta [and] does not depend on the ... contents." This makes it suitable for secure comparisons against timing side-channel attacks. To use it for strings:
Convert both strings to byte arrays using a consistent encoding (e.g., UTF-8 via getBytes("UTF-8")). Ensure the byte arrays have the same length first (as the method requires equal-length inputs for meaningful results). Call MessageDigest.isEqual() on the arrays.
I still don’t think a company with at least one touch point on such a high percentage on web usage should be allowed to have one of 2 mobile OSs that control that market, the most popular browser, the most popular search engine, the top video site (that’s also a massive social network), and a huge business placing ads on 3rd party sites.
Any two of these should be cause for concern, but we are well beyond the point that Google’s continued existence as a single entity is hugely problematic.
Been researching about waterproofing techniques in my area. Asked chatgpt about products in my region. Gladly mentioned some, provided links to shop. Found out I need to prep foundation with product X. One shop had only Y available, from description felt similar.
Asked about differences between products. Provided me with summary table that was crystal clear that one is more of a finishing stuff and the other is more of a structural and can also be used as finishing. Provided me with links to datasheets that confirm the information.
I could ask about alternative products and it listed me some, etc. Great when I need to research unknown field and has links... that is the good part :)
But it only works for stuff that is already consolidated. For example, something like a new version of a language will certainly spark new questions that can only be discussed with other programmers.
I'm not sure this is true? Most languages have fairly open development processes, so discussions about the changes are likely indexed in the web search tools LLMs use, if not in the training data itself. And LLMs are very good at extrapolating.
How long is the rear seat room is the 2018 XX Yy car? What is the best hotel to stay at in this city? I’m interested in these things and not interested in these amenities. I have leftovers that I didn’t like much, here’s the recipe, what can I do with it? (it turned it into a lovely soup btw).
These are the types of questions many of us search and don’t want to wade through a small ocean of text to get the answer to. Many people just stick Reddit on the query for that reason
If that's how most people use search engines these days, then I guess the transition into "type a prompt" will be smoother than I would have thought.
yes, but was it a good answer? were any sources backing up the answer credible (or even present)?
I don't know why we are suddenly so confortable trading speed for accuracy. Rule 0 of optiization involves making sure the probalem is actually being solved.
> I don't know why we are suddenly so confortable trading speed for accuracy.
Googling something or looking it up on Wikipedia is already trading speed for accuracy because you'll be reading a summary/reinterpretation. If I really cared about accuracy, I'd be reading highway design and civil engineering journals.
Also if you type a few words on Google, it’ll “autocomplete” with the most common searches. Or you can just go to trends.google.com and explore search trends in real time.
it's a miracle it survived that long. and i think it saving grace was that nobody wanted to browse reddit at work, nothing else.
so tired of AI apologists exploiting this isolated case as if it is some proof AI is magic and a solution to anything. it's all so inane and expose how that side is grasping for straw.
It still usually has the standard quality of answers for most questions I google. I google fewer questions because modern languages have better documentation cultures.
They tend to provide answers that are at least as correct as StackOverflow (i.e. not perfect but good enough to be useful in most cases), generally more specific (the first/only answer is the one I want, I don't have to find the right one first), and the examples are tailored to my use case to the point where even if I know the exact command/syntax, it's often easier to have one of the chatbots "refactor" it for me.
You still want to only use them when you can verify the answer and verifying won't take more time. I recently asked a bot to explain a rsync command line, then finding myself verifying the answers against the man page anyways (i.e. I could have used the manpage instead from the start) - and while the first half of the answer was spot on, the second contained complete hallucinations about what the arguments meant.
They are both terrible in terms of correctness compared to duckduckgo->stackoverflow.
As an example deepsek makes stuff up if I as for what syscall to use for deleting directories. And it really misleads me in a convincing way. If I search then I end up in the man page and I can exentually figure it out after 2-3 minutes
Raw data here if you want an update: https://data.stackexchange.com/stackoverflow/query/1882532/q...
It hasn't got better - down from a peak of 300k/month to under 10k/month.
Stack Overflow isn’t dead because of AI. It’s dead because they spent years ignoring user feedback and then doubled down by going after respected, unpaid contributors like Monica.
Would they have survived AI? Hard to say. But the truth is, they were already busy burning down their own community long before AI showed up.
When AI arrived I'd already been waiting for years for an alternative that didn’t aggressively shut down real-world questions (sometimes with hundreds of upvotes) just because they didn’t fit some rigid format.
It is dead because of both of those things. Everyone hated Stackoverflow's moderation, but kept using it because they didn't have a good alternative until AI.
> When AI arrived I'd already been waiting for years for an alternative that didn’t aggressively shut down real-world questions
Exactly.
The reason why I resort to AI is to find out alternative solutions quickly. But quite honestly, it's more of a problem with SO moderation. People are willing to answer even stale, actual/mistaken duplicate or slightly/seemingly irrelevant questions with good quality solutions and alternatives. But I always felt that their moderation dissuaded the contributors from it.
Meanwhile, the first reason why I always double check the AI results is because they hallucinate way too much. They fake completely believable answers far too often. The second reason is that AI often neglects interesting/relevant extra information that humans always recognize as important. This is very evident if you read elaborate SO answers or official documentation like MDN, docs.rs or archwiki. One particular example for this is the XY-problem. People seem to make similar mistaken assumptions and SO answers are very good at catching those. Recipe-book/cookbook documentation also address these situations well. Human generated content (even static or archived ones) seem to anticipate/catch and address human misconceptions and confusions much better than AI.
if like me you didn't know what this was referring to, here's some context: https://judaism.meta.stackexchange.com/questions/5193/stack-...
They also devolved into a work friendly variant of 4Chan's /g/ board. "Work friendly" as in nothing obviously obscene, but the overall tone and hostility towards newcomers is still there (among other things).
AI isn't competition for Google, AI is technology. Not only is Google using AI themselves, they are pretty damn near the top of the AI game.
It's also questionable how this is relevant for past crimes of Google. It's completely hypothetical speculation about the future. Could an AI company rise and dethrone classic Google? Yeah. Could Google themselves be the AI company that does it? Probably, especially when they can continue due abuse their monopoly across multiple fields.
There is also the issue that current AI companies are still just bleeding money, none of them have figured out how to make money.
We all know that "for now" will age horribly, and faster than we might even expect. When they put ads in, that's when we all know the boom is truly over.
So I get not liking this answer, but I haven't heard a better one.
I mean but it appears to be being remedy'd by itself why would the court proscribe something for a problem that no longer exists?
Is this an evidence based claim? From the Q2 2025 numbers Google saw double digit revenue growth YoY for search.
https://www.theguardian.com/us-news/2025/jul/23/google-expec...