where does a concept begin? where does a concept end? what is a concept?
Abstractions are a way to characterize the behavior of complex systems. It's not feasible to directly compute their behavior, but they still have predictable properties. Concepts let you handle emergence; you can manipulate an object as its own thing rather than a collection of atoms.
(And yes, I did just read Stephen Wolfram's book and this idea is largely based on it. I think he has some delusions of grandeur but is also onto something.)
Thus the sharp drop off where it can for example guess the correct answer to some math problems yet get others that are conceptually identical completely wrong.
You need to approach this stuff sideways to see behind the curtain. There’s some hilarious videos where it’s “playing” chess and the first few moves seem very standard because it can simply copy a standard opening. It really has no concept of a valid move just statistical associations. Yet it was trained on more games than most people ever play and high level analysis and etc, but none of it means anything to the algorithm beyond the simplest associations.
Granted this stuff is a moving target it’s easy enough for them to slap a chess engine and suddenly “it” would actually know how to play.
ChatGPT: > No, you cannot physically eat an Apple share or any other stock share. A share of a company's stock represents ownership in that company and is typically bought and sold on stock exchanges. Share prices fluctuate based on various factors such as supply and demand, company performance, market conditions, and investor sentiment. While you can buy and sell shares of Apple on the stock market, you cannot consume or physically eat them.
That seems perfectly reasonable to me. The answer correctly identifies the problem with the question.
I would recommend looking up the Othello paper. Chess may be beyond the current level of LLM capability, but that doesn't mean they aren't manipulating things at a level higher than tokens.
The Othello paper is hardly a counter example. Researchers created an Othello specific model that almost learned the grammar of Othello not how to play well. Yes, there was largely correct internal game state built up from past moves. No it didn’t actually learn the rules so it would make strictly legal moves nor did it learn to make good moves.
I don’t bring up this inaccuracy because it actually makes much of a difference to playing Othello, but rather to illustrate how these systems are designed to get really good at faking things. There’s approaches that allow AI to actually learn to play arbitrary games, but they differ by having iterative feedback rather than simply providing a huge corpus. It’s like science vs philosophy, feedback prunes incorrect assumptions.
It is, though. Nobody said anything about playing well, or learning the rules. The very fact that it had a valid internal representation of the game state means it's extrapolated beyond token-level. Which is the point.
Hardly, decoding “Bob picked up an Apple.” requires you to make an association between the token Bob and the token Apple. But that doesn’t imply a LLM is operating above the token level.
And again it failed to make strictly correct moves because its internal representation of the game state was flawed. Objectively an LLM when specifically trained on Othello a game with very simple and completely mechanical rules failed to abstract what those rules actually where.
The internal representation was a literal 8x8 grid of piece locations they could externally change and have it generate moves consistent with the changed position. It's about the clearest example of a learned higher-level internal representation I can think I've seen.
The fact that it didn't also perfectly learn the rules while it was doing that is entirely uninteresting.
No, that's an a priori concept. A posteriori concepts comprise empirical knowledge which necessitates experience of the world [1].
Example: you can know a priori that "all bachelors are unmarried." If I tell you "Tom is a bachelor" then you know that Tom is unmarried (assuming I tell the truth about Tom being a bachelor).
But if I say "all bachelors are unhappy" then you haven't learned anything because knowledge about the happiness/unhappiness of bachelors is an empirical question. To know whether or not I was telling the truth, you would need to conduct research about the real world, for example by conducting a survey of bachelors.
But, things are defined by how they interact with the world around them.
A concept is its relations to other concepts.
Which does seem to be the general sort of thing that these models are trying to get at, even if they don't seem to do a great job of it.