Comments by "Wiczus" (@wiczus6102) on "Francois Chollet - Why The Biggest AI Models Can't Solve Simple Puzzles" video.

@cj-ip3zh By impression you mean that he literally said he is playing the devils advocate?
18
@deepspace9043 Yes, but I constantly see people who assume that there is exponential growth somewhere when often it is anything but. And there might be some limiting factor that stops this growth. Like when it comes to computing power, everyone keeps saying how we still double the transistor count but the actual transistor density barely moved since 2010. So this extrapolation is only valid if there is a technical reason to believe that it is. As someone who uses GPT I don't see it as a replacement. It's basically a better search engine, it's not even close to being a problem solver.
5
@TheBobiaan We are deterministic. Nothing can create information, we just synthesize it from previous concepts. So when you search for a mathematical solution you're iterating through a set of conceptual permutations. We can all do it if we're not lazy or demotivated or under time pressure. I don't see any fundamental blocker for an LLM to do so. They are probably not as efficient at it but they do posses some logical capability. What fundamental difference do you see here?
3
@ahpacific I am pretty sure Destiny is very transparent on when his thoughts are shallow or not. Notice when he formulates things as open questions and or the adjectives and intonations he uses to suggest the difference between perspective and fact. You can make a false statement and still create a good converstation like that. People like that are fun to talk to as opposed to people who will only say something if they are fully certain.
3
@jasonk125 I would argue that people who wrote my codebase are below average.
1
@human_shaped Not the whole calculus but I was able to deduce nx^(n−1) when I was 13. The point is that we do have this deduction capability if we have the motivation to try it.
1
It's a little bit of a nonsequitor. If you see agi as an intelligence with different competences these competences get filled by different agents. Including an LLM agent.
1