8 ms·

Succinctly stated and something that resonates strongly with me. In the last internet revolution (web search), results started high quality because the inputs

• celdon25 3y ago
  The comparison is only valid if AI ends up being monopolized. A continuously evolving ecosystem on the other hand has a better chance of adapting to those pressures. I am sure there are search engines that don’t index the SEO crap, but I don’t remember the names, and they have other flaws. Open source AI needs to get a lot of investment for this to be mitigated. Relying on market incentives to drive development without the possibility of forking is too dangerous given the high stakes.
  • piva00 3y ago
    Given enough time, everything seems to go the way of consolidation. I don't have much hope that AI will be different...
• Solvency 3y ago
  This comment/sentiment/idea has already been thoroughly discussed, shared, retweeted, posted, and echoed a couple thousand times online already. Humans like you have been redundantly saying the same things in varying degrees of novelty, like this, for centuries.
  • alkonaut 3y ago
    That's what the topic of the post we are discussing is. He is basically reiterating the point. Which is a good point. And there have been zero good arguments made (as far as I'm aware) to why this fear would be unfounded.
    • Solvency 3y ago
      I'm pointing out the irony. The Internet is already saturated with us continually metaphorically JPG compressing original ideas into textual redundancy, with many people having only been exposed to those copies of copies.
      • hermitcrab 3y ago
        I'm not surprised this isn't an original idea. But I haven't seen it discussed.
  • lm28469 3y ago
    And "humans like you" have been saying what you just said for the same amount of time, what's your point? If we can only dicuss ideas which have never bee shared we can close this website right now
• TaylorAlexander 3y ago
  I think it’s clear that LLMs cannot be the end state of this technology, and we will need systems that can reason and develop hypotheses and test them internally. These systems may benefit from more curated datasets (such as those collected before the bullshit wave began) along with real world interaction data from YouTube and robotics. Such systems could eventually be used to rank web pages for their bullshit level, which of course would present a risk of censorship but when used properly could lead to insightful data handling. It’s just really clear that a giant text averaging machine can only go so far, and while we do see some higher level emergent properties, we’re going to have to move beyond the current state of the art in the next decade, and such future systems may be much less affected by the internet’s bullshit. Even without the bullshit singularity I think such measures would be a necessity and we would see them developed soon.
  • pton_xd 3y ago
    > It’s just really clear that a giant text averaging machine can only go so far It's not really a text averaging machine, it's a pattern matching machine. Right now the "depth" of the patterns it can match can only go so far, but in a few years with more advances in chips and memory the depth is going to increase and the patterns it can match will fan out accordingly.
    • vrighter 3y ago
      it is a statistical model. If everyone is saying X and is wrong, and one guy says Y and is right, the llm will bit out X. Because that is the most probable thing in the dataset. It literally is a text averaging machine
  • visarga 3y ago
    Not just real world data from videos. AI models need feedback from many sources: humans, code execution, web search, simulations, games, robotics, math verification, or from actual experiments in the real world. All of these are environments that can take the output of a model and do some processing and return feedback. The model can learn and search for solutions, creating its own training data as a RL agent. Since all deployed models produce some kind of effect and feedback from the world there is an opportunity there to collect data targeted on the current level of the model, the most useful kind of data. That's why I think AI will be ok even with the proliferation of bots online. It's not 100% pure synthetic data in a loop, it is a agent-environment loop. tl;dr Models learn better from their own experiences, not ours.
• __loam 3y ago
  We may enter into a highly ironic regime where ai companies subsidize writers and artists at scale to produce better data.
  • aj7 3y ago
    I like this idea very much. Is this similar to human moderation of training data?
    • __loam 3y ago
      Sure. I suspect it might be hard to actually recruit people like this considering how the people who got sampled are hugely pissed off at AI firms rn, but money will go a long way here.
      • spaceguillotine 3y ago
        Royalty agreement could make this work, every time their data gets used, they get a cut, it would be the only way i could see people agreeing to this.
• Turing_Machine 3y ago
  I guess some of it might depend on how good the AI-generated content gets, and also how good the AI gets at detecting AI-generated content. If the AI was good at detecting it, it wouldn't matter if the AI-generated content sucked, yes? Even a low probability of detection would help. Let's say our algorithm is 50% likely to detect AI junk. That means that half the junk data won't make it in to model. Even 20% would probably be worthwhile, especially if it also threw away human-generated junk (and let's be realistic here: there is, and always has been, no shortage of terrible and/or wrong human-generated content). Let's say those crappy filler paragraphs that get stuck between pictures on meme clickbait pages... I suspect most of those people have already been replaced, but that prose was horrible long before the current LLM boom. I suspect there is a lot of effort being expended right now on ways to ensure the training data (whether human or AI generated) isn't shite.
  • vrighter 3y ago
    ai detectors can be used to train ais to avoid triggering them. It's a race that cannot be won. ai detection can't, and already doesn't, work.
    • Turing_Machine 3y ago
      > ai detection can't, and already doesn't, work. Then how can humans often tell that something is generated by AI? Besides, I wasn't suggesting AI detection as such. I was suggesting crap detection, regardless of how the crap was produced. You probably don't want crappy human-generated content going into your training set, either.
      • asadotzler 3y ago
        Humans are smarter than machines. We "sense" bullshit in patterns that machines cannot yet intuit. I can tell when something's AI generated. It has a smell or a flavor that's different from human work and it's a variable pattern because, like a southern accent, I can tell if it's Georgia or Alabama, Gemini or Claud, even if very capable machines cannot.
• rocqua 3y ago
  Unlike the low background radiation steel, high quality content will still continue to be created. Arguably even at a faster rate with these powerful tools. The proportion will drop ofcourse but that only means curation will become king. With search this curation was difficult, because the value was so low you could only do it profitably if it was completely automatic, and that was difficult. An optimistic take would be that LLMs will make curation so much more valuable, that it will be done much better. If the wider world gets to use this curation to limit spam, and highlight good work, it would be amazing for the world.
  • hermitcrab 3y ago
    >LLMs will make curation so much more valuable, that it will be done much better. Won't that also mean a return to a world of gatekeepers?
    • slowmovintarget 3y ago
      That gatekeepers will be the ones deploying the ML systems.
    • rocqua 3y ago
      Yes, hopefully they will be good.
  • ofrzeta 3y ago
    Let the crawler AI discard a sample if it detects it as AI generated? That's at least quite expensive I think.
  • slowmovintarget 3y ago
    You're forgetting that we have mechanisms in place to curate curation and farm attention toward the crap, not toward quality. It will be harder and harder to break the surface tension of this gooey gelatin wrapper we've placed over creative activity.
    • MichaelZuo 3y ago
• daxvena 3y ago
  I doubt it's going to go the same way as search. You can't run Google on consumer hardware, but you can run LLMs locally. At worse, newer models will get worse and you can just stick to older models. You could also argue that proprietary models gated by an API are better than anything you can run locally, and yeah maybe those will get worse with time. They're not going to get any worse than what you can run locally though. If they do, open models will overtake them, and then we'd be in a better position overall.
  • Jensson 3y ago
    > You can't run Google on consumer hardware, but you can run LLMs locally. You can't run an up to date model locally. When I ask Googles models they have knowledge from stuff just a week ago, without using search. You wont get that from a giant local model.
• hackerlight 3y ago
  I don't think it impacts AGI timelines much. Worst case scenario, we just cut off training at 2024 data. But it's not like another ~5 years of internet data is the magic that will get us over the finish line to AGI. We should have enough data already. We will get to AGI with synthetic data (e.g. AlphaGeometry for code, or simulators e.g. RL inside unreal engine), world model from video, algorithm improvements, and 100x more compute. I'd add that majority of LLM-generated stuff put onto the internet isn't pure junk. It's endowed with human-created context and curation. If someone submits some LLM-created code to Github, it's because it works. A small fraction is pure noise (e.g. state-actors spamming Twitter) but that should remain a minority.