6 ms·

LLMs are not emulating the human brain. Somebody may well be able to do that someday, but right now nobody is even trying to.

• josephg 24d ago
  Why would you need brain emulation to get superhuman intelligence?
  • ksenzee 24d ago
    Are you making a serious argument that superhuman intelligence is a plausible outcome of training LLMs on everything humanity knows so far? Or are you making the generic assertion that AGI is theoretically possible via means other than emulating the human brain? Because the latter is a strawman (nobody has asserted anything to the contrary), and I have seen no evidence at all to support the former.
    • josephg 24d ago
      I think we can compare the human brain and LLMs on a bunch of capabilities today, and see how we compare. By my reckoning: - LLMs have better long term memory (they know more than any human) and more working memory (LLMs have fast, uniform access to their whole context window). - LLMs are faster than we are. - Humans have online learning (we can do simultaneous learning and inference), giving us advantages in many novel tasks. - We can learn concepts from far less data. And we can manage our mental context more smoothly. - We seem to have better world models than current models. AI video just doesn't look right, somehow. I expect that these remaining weaknesses can be overcome without resorting to human brain emulation. I see no reason to think that current LLMs are at the limit of what technology is capable of.
      • shawnb576 24d ago
        Because LLMs don't understand anything. That's the tech. They can only predict what they have been trained with and fail daily at the most basic tasks. Granted they can do amazing things, no question there. But they are not "smart". For example, it seems that even at Fable scale, simple concepts like the passage of time or (gasp) timezones elude them. I live in UTC+10 and with any RFC8339 data LLMs are constantly confused - is it Sunday the 10th or Sunday the 9th, etc. I have tried many solutions for this and every time it finds a way to get it wrong.
        • penteract 24d ago
          To me it sounds like you're repeating what gp said about the lack of online learning. Do you think that's insurmountble? Getting confused about timezones does not place LLMs behind that many humans. (But doing so repeatedly does highlight the lack of online learning).
          • dboreham 23d ago
            I see this thread as progress because now there's three people saying this (seemed like it was just me for a year or two).
        • josephg 23d ago
          > Because LLMs don't understand anything. How do you square that then? They can do amazing things, but they're also not smart? Do you think its possible to solve Erdos problems without any "smarts"? Can you do it without even understanding mathematics? I find it very hard to hold the idea that LLMs don't understand anything. They can explain concepts, translate them, simplify them and implement them in code. From the outside, LLMs seem to understands most concepts better than most humans do. Do you understand anything? Couldn't I make the same argument? How would you prove that you understand what a for loop is, or that you know what calculus is? I assume you'd demonstrate your knowledge by using a for loop in a program, or explain calculus back to me. But LLMs can do that too. > For example, it seems that even at Fable scale, simple concepts like the passage of time or (gasp) timezones elude them. Funny example, because lots of human struggle with this too. The number of meetings I've had with people in the US! "Lets meet on thursday morning australia time!". Only, they actually meant thursday night US time, which is friday morning australia time. "Oooh that's so weird! Its the next day for you!". ...... Yes, I know. I think LLMs are just a different kind of intelligence than humans. They're better at some things than us, and worse than others. They can find latent security vulnerabilities in the linux kernel, but struggle to count the Rs in strawberry. They're not as smart as humans in many ways. But we're not as smart as LLMs in plenty of ways too. I didn't find those linux bugs.
    • CamperBob2 24d ago
      Are you making a serious argument that superhuman intelligence is a plausible outcome of training LLMs on everything humanity knows so far? Are you making a serious argument that it's not? Because you'll need to explain leading-edge mathematics advances that have come from LLMs, among other things.
      • ksenzee 23d ago
        Superhuman intelligence? Really? These leading-edge advances indicate intelligence beyond the level of humanity?
        • CamperBob2 23d ago
          Obviously. Otherwise humans would have made them. Also, you may have noticed in passing that humans aren't getting any smarter, while AI models are.
    • OJFord 24d ago
      What is meant by 'superhuman intelligence'? Certainly it seems to be the case that LLMs are capable of a sort of 'polyhuman' intelligence, in that the same LLM that advances mathematics with a novel proof can add unit testing for a new software feature, design a recipe, and create an SVG of a pelican riding a bicycle. As generalists I'd say they're already 'superhuman'.
    • scotty79 24d ago
      > training LLMs on everything humanity knows so far? That's not all of what we are doing for at least a year, possibly few. LLMs are trained increasingly on generated inputs. Soon human sourced material is going to be rounding error in the process of training.
      • ksenzee 23d ago
        To clarify: We are not training LLMs on any information that humanity does not already have access to.
        • scotty79 22d ago
          To clarify, information can be created at will, in automated fashion. If you add two random numbers and calculate the result and those happened to be numbers noone else ever had idea to add you created a new piece of information. Template is not new, but the piece of information is. And sure, this template might be very simple, too simple, but you can come up with more complex one. And metadata is data. You can create templates in similar manner to how you create new pieces of information using them. And labs training AI are doing it for years at this point. And it is ever increasing fraction of all training.