top 100 comments

sorted by: hot top controversial new old
[–] 181 points 5 months ago (27 children)

The death of Stackoverflow is one of these events where the site has been completely killed by AI and yet its contents is completely necessary for AI to know about solving programming problems. Its death will mark the end of AIs ability to learn how to solve programming issues. Its cannibalizing itself in the process, as it destroys its sources it destroys its own ability to learn.

  • source
  • hideshow 27 child comments
  • [–] 91 points 5 months ago* (24 children)

    It's not just that, it's shitting where it eats. People are using it to fill the internet with disinformation, then it trains itself on it's own disinformation, and breeds even worse disinformation. This is why AI can never be smarter than it was in 2021.

    On top of that, due to the indiscriminate DDOSing of the entire internet by AI bots, websites have been blocking any web crawlers that are not Google, which just contributes to their monopoly.

  • source
  • parent
  • hideshow 24 child comments
  • [+] -23 points 5 months ago (11 children)

    I'm pretty sure AI is objectively smarter today than it was 5 years ago.

  • source
  • parent
  • hideshow 11 child comments
  • [–] 14 points 5 months ago (4 children)

    How can something with no intelligence be smarter?

  • source
  • parent
  • hideshow 4 child comments
  • [+] -8 points 5 months ago (3 children)

    It is evident that it has intelligence, it outputs intelligent responses usually adequate to its input, even if it’s badly phrased. What it doesn’t have is sentience, conscience, and a learning loop.

  • source
  • parent
  • hideshow 3 child comments
  • [–] 1 point 5 months ago (2 children)
  • [–] -1 points 5 months ago (1 child)

    To anyone who interacts with it? Would you deny that a program automate mental labor in the same way that a sawmill automates manual labor? Isn’t that some degree of intelligence?

    Now, we have very imperfect LLMs who nevertheless can be instructed not with program code, but actual natural language, and they react accordingly. Isn’t that also intelligence? Computers that understood natural language was the realm of science fiction just five years ago.

    I get it that people hate LLMs, both because how idiots use them, and how corporations push them everywhere; but not recognizing the intelligence in those programs is naive at best.

  • source
  • parent
  • hideshow 1 child comment
  • [+] -7 points 5 months ago* (4 children)

    This is why AI can never be smarter than it was in 2021.

    You're completely wrong. First of all, datasets are getting bigger and objectively better. Secondly, technologies and methods of model training become dramatically more complex. Yes, AI content can create echo for itself, but it's a solvable issue.

    Anyone who disagrees with this fact should do a reality check and learn a little more about the current AI state. In 2021 we didn't even have reasoning in models.

  • source
  • parent
  • hideshow 4 child comments
  • [–] 1 point 5 months ago* (3 children)

    I see someone's been suckling the OpenAI teet.

  • source
  • parent
  • hideshow 3 child comments
  • [–] 2 points 5 months ago* (last edited 5 months ago) (2 children)

    What it have to do with OpenAI? It’s not what I’m talking about.

    Nonetheless, it’s kind of fun to see so much downvotes on my comment. Some people are so blinded by their hate (well deserved, by the way) that they will deny something that obvious.

  • source
  • parent
  • hideshow 2 child comments
  • [+] -18 points 5 months ago (6 children)
  • [–] 29 points 5 months ago (2 children)

    Our key finding is that by injecting information through an external synthetic data verifier, whether a human or a better model, synthetic retraining will not cause model collapse.

    Yeah if you have a source of truth then your model is basically getting trained on that.

    It’s like already having the answer

  • source
  • parent
  • hideshow 2 child comments
  • [–] 21 points 5 months ago*

    Our key finding is that by injecting information through an external synthetic data verifier, whether a human or a better model, synthetic retraining will not cause model collapse.

    Lol, so to make a great model, they just need to have an even better one available first or a human who can verify every single thing it ingests.

    Hmm, call me skeptical on this claim.

  • source
  • parent
  • [–] 10 points 5 months ago

    This assumes everything is valid on the external. If one slop cluster feeds off another - a slopveyor? - then there is nothing external for the validation hall-monitor to compare against. They're trusting another model's output as if it were gospel.

  • source
  • parent
  • [–] 85 points 5 months ago (42 children)

    Friends don't let friends use Google.

  • source
  • hideshow 42 child comments
  • [–] 27 points 5 months ago* (last edited 5 months ago) (22 children)

    I can't remember the name, but when the internet was just starting and there were a lot of search engines with no dominate ones, there was an aggregator program that you could input many search engines into, then use it as the searching tool. It would query all the engines and combine, sort, rank, and remove duplicate finds.

    Edit: more specific - It was much like an FTP or torrent program but you'd load up what search engines to use and your search words, and it would actively pull the info then provide a single page with all results.

    The reason I mention it is because we're sort of back at that point. Google is failing, Bing never was great, and all the alternatives have their issues, usually with not having the same database to work with. So if you gathered all the best ones, the ones without ties to corporate or AI, then put their results together, maybe you'd have something like what Google was at its peak before "do no evil" got painted over.

    Incidentally, Google became what it was/is because it gobbled up a lot of those early search engines' databases. I miss you, Hotbot. You were a good one.

  • source
  • parent
  • hideshow 22 child comments
  • [–] 45 points 5 months ago (13 children)

    Search used to be so good. I had an old Honda civic that suddenly wouldn't start. It wasn't the starter, alternator, or battery. I managed to find a forum post with my exact issue, which was that a small rubber piece on the clutch pressed a button to "tell" the starter it was okay to start. Twenty minutes later I had zip tied a piece of plastic into place and had a working car again.

    If I tried to diagnose that same issue today, it'd be dozens of SEO garbage slop sites without any actual useful information.

  • source
  • parent
  • hideshow 13 child comments
  • [–] 14 points 5 months ago* (last edited 5 months ago) (1 child)

    I was thinking the same thing recently. It's not the place it once was. But in general the internet has changed a lot. And it's not just AI.

    1. All sorts of paywalls especially in news sites.
    2. Everything is getting centralized into a few sites and they're usually eithe poorly indexable or not at all (Discord, facebook, X, Instagram and so on)
    3. Fediverse (Lemmy, Mastodon) also struggles with search engines.
    4. People trying to sell you shit, create a brand even more than before. Because of this all sorts of SEO optimization crap is done like writing BS articles nobody cares about.
    5. AI slop.
    6. Search engines have gotten better of getting rid of "illegal stuff".
    7. A lot of sites are just presentational bloat with no substance. Very cool looking landing pages with all sorts of cool animations but when you need to actually find the information that you need... the same UI usually gets in the way.

    Oh and now we're getting into age verification crap also yay

  • source
  • parent
  • hideshow 1 child comment
  • [–] 7 points 5 months ago

    An example of number 4, there's a poster I've seen on reddit that's posting very relevant content, but then every post ends with "@xxxxxxxx on all socials". It just takes the whole thing from content I might want to engage with to the exact opposite.

  • source
  • parent
  • [–] 5 points 5 months ago

    They are literally walling off all this information that used to be easy to access and for the public. It’s our data that we the people decide to share with the world and these rent seeing corporations are hiding it away so they can start charging us "tokens" to access our own public information.

  • source
  • parent
  • [+] -6 points 5 months ago (9 children)

    I asked gpt5 and it told me to check the clutch safety switch. The thing you fixed.

  • source
  • parent
  • hideshow 9 child comments
  • [–] 17 points 5 months ago (3 children)

    5-10 years ago, you could be pretty sure this was a thing that actually needed checked, since the post about the clutch safety switch was posted by a real person who presumably had the same problem as you and fixed it with this method.

    Now, there's no way to know if that's actually the case, or if "clutch safety switch" is just a likely string of words to feed someone who is having car trouble. You might get lucky, or you might get sent on eight consecutive goose chases because an LLM fundamentally doesn't know what factual knowledge is, it only knows how to reorder and regurgitate things that other people have said in other contexts.

  • source
  • parent
  • hideshow 3 child comments
  • [–] -3 points 5 months ago (2 children)

    I agree with the larger point you’re making, but chatbots are getting better at referencing posts / websites from which they’re taking a solution.

    That’s if and only if of course they used a web search tool to answer, and if that website is still alive — made less likely due to AI.

    But for debugging something like this, it is actually helpful for now with citations enabled.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 3 points 5 months ago (1 child)

    The problem is that right now would be the peak of this information being available. What are you going to find in 20 years when everyone has abandoned forums in favor of asking ChatGPT for all the answers? There would be nothing left to train the models on.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 2 points 5 months ago

    Did it give a diagram and troubleshooting steps from the factory service manual too? This is all stuff you would typically find in forums. There's always some dealer tech around who can copy and paste from their service equpiment/library

  • source
  • parent
  • [–] 1 point 5 months ago (2 children)

    I tried giving minimal information and still got similar results.

    When I think about what got worse about the internet, it's mostly the life stories before recipes, the novel length pages to maybe answer a simple question, and pretty much anything else related to SEO.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 7 points 5 months ago (2 children)

    I think the old aggregator you were thinking of was dogpile.com

  • source
  • parent
  • hideshow 2 child comments
  • [–] 4 points 5 months ago (4 children)
  • [–] 2 points 5 months ago (3 children)

    No, I don't see mention of it being an application but like Dogpile is a web-based collector.

    I did a search myself, but (given how searching sucks now) couldn't find anything. Lots of hits for search engines themselves, but getting past that to other methods back then is difficult.

    It was much like an FTP or torrent program but you'd load up what search engines to use and your search words, and it would actively pull the info then provide a single page with all results.

  • source
  • parent
  • hideshow 3 child comments
  • [–] 1 point 5 months ago

    I thought it was Autonomy. You installed a program, instructed puppies agents, logged out, and while you were offline the puppies searched through several engines. Next time you logged in the findings waited for you. That was the time of 56k modems and metered connections.

  • source
  • parent
  • [–] 1 point 5 months ago (1 child)
  • [–] 13 points 5 months ago (18 children)
  • [–] 13 points 5 months ago (9 children)

    Ecosia have planted 250,000+ trees so far and publish their accounts every month. I can’t think of a better option, unless there is a niche requirement.

  • source
  • parent
  • hideshow 9 child comments
  • [–] 11 points 5 months ago (8 children)

    they burn all their efforts by pushing genAI tech on their platform

  • source
  • parent
  • hideshow 8 child comments
  • [–] 7 points 5 months ago (7 children)

    Isn’t that hyperbole rather than truth? They’re still carbon negative.

    They don’t provide AI by default (at least, I don’t get it). So people like us can continue to not use AI and the hundreds of million who use it every day can still support tree planting.

    I don’t like AI, but if they don’t add it they could risk limiting their reach and environmental goals.

  • source
  • parent
  • hideshow 7 child comments
  • [–] 8 points 5 months ago (2 children)

    AI companies do not release any numbers themselves for carbon emissions. Therefore, companies that use AI cannot in any certainty claim to be carbon negative or neutral, because they have to count the supply chain emissions as well.

    not adding AI does not stifle environmental goals, in fact you can only truthfully claim to strive for carbon goals if you do not use AI. After all, there is a reason that Microsoft abandoned their emission goals with AI as the cited reason first and foremost, which shows how incredibly dirty AI can be, even if no one releases any sensible metrics.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 0 points 5 months ago* (last edited 5 months ago) (1 child)

    AI companies do not release any numbers themselves for carbon emissions.

    I think the EU is even helping them keep that "operationally sensitive information" private, which is a shame.

    not adding AI does not stifle environmental goals.

    I can't agree or disagree here. I know the demand for AI is huge: hundreds of millions of users per day and at least a billion per week. If Ecosia is seen not to have this feature, I would consider it possible that it hurts their adoption and therefore their goals.

    which shows how incredibly dirty AI can be, even if no one releases any sensible metrics.

    Yes. I'm not sure how much solace the "world greenest AI" slogan can really offer in that context. https://blog.ecosia.org/ecosia-ai/ - but when I'm recommending search to someone, I recommend Ecosia over Google, Bing, DDG, Qwant, Mojeek, etc. simply because I think they are more of a net positive than the other options.

    Who knows, maybe in a year or twos time I'll look back and regret it when more information surfaces. But they've been sensible enough until now with their operational choices to reach tree-planting goals.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 7 points 5 months ago* (3 children)

    Go to https://ecosia.org in a private browser window. It says "AI that answers to the planet". Search something and the AI Overview on top is enabled by default.

    I use them with AI disabled, but it should be the default setting.

    Edit: I just did a couple test searches and didn't get the AI overview. Don't know what triggers it.

  • source
  • parent
  • hideshow 3 child comments
  • [–] 1 point 5 months ago (1 child)

    Go to https://ecosia.org/ in a private browser window. It says “AI that answers to the planet”. Search something and the AI Overview on top is enabled by default.

    For what it's worth i've been using them for like a year and I clear my cookies often and never got this AI overview thingy you're talking about. I actually have no clue how it even looks like.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 1 point 5 months ago* (last edited 5 months ago)

    That's A-B testing. Another evil gas lighting dystopian thing that has got people questioning their sanity. Even something as basic as how a website's UI behaves has to be hidden behind some secret "innovative" algorithm that "optimizes our engagement."

  • source
  • parent
  • [–] 13 points 5 months ago (6 children)

    Startpage and Mojeek.

    DDG has contracts w Microslop.

  • source
  • parent
  • hideshow 6 child comments
  • [–] 1 point 5 months ago (2 children)

    Thanks for the warning; I'll leave DDG for Qwant!

  • source
  • parent
  • hideshow 2 child comments
  • [–] 3 points 5 months ago

    Ecosia and Qwant are building together an EU based index. Maybe then they will become actually independent alternatives to American giants. It will take a while.

  • source
  • parent
  • [–] 34 points 5 months ago

    More Perfect Union did a video on Google's descent into evil. I think it's this one

    TLDW: Once Google pivoted from being a search service to an advertising agency, it was motivated to keep users from hyperlinking away from Google, and so offered summaries and alternatives controlled by Alphabet that allowed it to keep offering you ads.

    So this AI service is just a natural iteration.

  • source
  • [–] 24 points 5 months ago* (10 children)

    he means reddit mostly. AI SLOP GENERATOR, TRAINING ON SLOP like reddit. with a little of plagiarizing from authors, and artists.

  • source
  • hideshow 10 child comments
  • [–] 13 points 5 months ago (4 children)

    I called to schedule a play date at my local dog daycare/boarding, it went to an AI answering service. I asked if it was AI since I could hear noice in the background (literal fake background chatter and noise), when she said yes, hung up. SO tired of AI everywhere. Fuck it all.

  • source
  • parent
  • hideshow 4 child comments
  • [–] 9 points 5 months ago (3 children)

    lmao, I bet they trained the stupid thing on recordings from massive call centers. It probably thinks that all the 'background' noise is just part of how humans communicate.

  • source
  • parent
  • hideshow 3 child comments
  • [–] 7 points 5 months ago (2 children)

    it probably thinks

    It doesn't think. People need to stop anthropomorphizing the statistical probability machine.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 2 points 5 months ago

    You're right. I'm usually more careful about that. In fact I've even left comments just like yours in the past.

    Just goes to show how hard the marketing engine is really working to paint these things in a certain (deceptive) way.

  • source
  • parent
  • [–] 5 points 5 months ago (4 children)

    ah yes, reddit, the most well-mannered and measured of social media platforms that never indulges itself in spreading hate in misinformation.

  • source
  • parent
  • hideshow 4 child comments
  • [–] 2 points 5 months ago (1 child)

    which are spread by the same actors as the other social media. Russia, ISRAEL, PALANTIR, AND THE US, India/sr lanki is also getting thier hands in, likely being paid as intermediary for the others(of the above mentioned), so its to ofuscate who the actual actors behind the misinformation.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 23 points 5 months ago

    It's the same arc every monopolistic corporation has taken before it, AI is just accelerating the pace of consuming your customer/product because profits must always increase.

    There will be no large scale shift from these experiences because most people are either ok, apathetic or blissfully ignorant to the situation, the best you can do is to remove yourself from the exploitation of the userbase. Linux instead of Windows or Android, Almost any search engine other than Google, fediverse instead of reddit, etc.

  • source
  • [–] 13 points 5 months ago (1 child)

    It's a self defeating strategy as more people turn to ai, less content gets produced so ai becomes static.

    I truly believe the token model will kill AI, it will become too expensive

  • source
  • hideshow 1 child comment
  • [–] 17 points 5 months ago

    It already is too expensive and adding more compute doesn't make it cheaper lol it just causes a race to the bottom among data center providers and an eventual crash there too.

  • source
  • parent
  • [–] 16 points 5 months ago (5 children)

    I don't understand how these companies want to seem and think they are so smart by choosing new niche data (scraped) to train AI in a bid to try and make it "smart"....

    Has any other living being become "smart" by only ingesting information directly from the Internet? You can train other animals to perform many tasks and can probably say they are smart when they perform them as expected. I doubt any of the training methods is to tape headphones, a screen and sometimes a microphone to their faces forever (I kinda don't wanna know if this false 😶).

    The best example we have, is ourselves, and even though we use the Internet, babies are not taught how to walk and talk by only interacting with the Internet.

    I feel like I might be saying too much, but I think the best AI we're gonna get is to unplug it from the Internet, and then fucking raise it for 20 years like a normal, super fast-thinking child prodigy. Then just make copies of that and train further by having it go to school for the things needed.

  • source
  • hideshow 5 child comments
  • [–] 4 points 5 months ago

    That's a very naive simplification of the AI training process. You start with that, then pay people pennies in a developing nation to produce hand crafted training data, resulting it using stupid words like delve and whimsical entirely too much.

    Merely training on internet content with no RLFH training results in probable gibberish like that of GPT-2

  • source
  • parent
  • [–] 8 points 5 months ago

    Only a minute away until Google starts a Soylent Green subsidiary company

  • source
  • [–] 8 points 5 months ago
    [–] 8 points 5 months ago

    Google has been cannibalising the net for a decade or so now

  • source
  • [–] 5 points 5 months ago (1 child)

    Honestly I think Google is pretty fucked in the long term

    Nobody google anymore. They just ask chat

  • source
  • hideshow 1 child comment
  • [–] 11 points 5 months ago

    We put this real question to AI Mode, "Why does Google Search suck now?" And you won't believe the answer! (Actually, you won't be surprised.)

    The Chocolate Factory's AI search substitute offered this completely trustworthy reply:

    "You are definitely not imagining it. Users, tech critics, and researchers have documented a measurable decline in Google Search quality. The core issue is that the search engine no longer feels like a tool designed to find the best corner of the web; instead, it feels like a vehicle designed to keep you on Google-owned properties or clicking on monetized links.

    "The degradation of Google Search stems from a mix of aggressive monetization, an ongoing arms race with web spammers, and the disruptive introduction of AI features."

    AI is gonna be the same path. Once people trust it, it'll stop mattering if it give the best info, what matters is if people buy products and believe advertisements/propaganda.

    There just will never be as much money in providing nonbiased and accurate answers as there is in manipulating people. So as long as capitalism runs everything, the end goal will always be manipulating people for someone's profit.

    It's not an accident, it's not "enshitification", it's the natural and inevitable result of unregulated capitalism. The only way to avoid it, is to heavily regulate capitalism. Acting like it's a separate problem that can be avoided without regulating capitalism just makes people think we can make unregulated capitalism work, if we could it would be working already.

  • source
  • [–] 4 points 5 months ago (1 child)

    I... Is that baby eating a human?

  • source
  • hideshow 1 child comment
  • load more comments
    view more: next ›