3 ms·

You're unintentionally arguing against your point. If you sat someone down who had never played chess before, and taught them the rules, they also can't play c

You're unintentionally arguing against your point. If you sat someone down who had never played chess before, and taught them the rules, they also can't play chess. They wouldn't remember the allowed moves and would make tons of illegal moves. The only way a person reaches the state of being able to play chess is to usually have learned the rules and also watched a bunch games. Either games between others, or most frequently games of "almost chess" that they play with a helpful instructor that corrects them every time they make an illegal move. In that respect both humans and llms learn to play chess the same way - by watching games. Or for an analogy, think about how many times you've sat down to learn a new board game and after the person who had played it before explains all the rules you still don't really get it. What's next? "Let's just play a practice round and I'll pick it up as we go." Ie learning from seeing the game played. I think a lot of the skepticism in LLMs is people over estimate / inaccurately remember how people really learn / behave.

Jensson 3y ago
> If you sat someone down who had never played chess before, and taught them the rules, they also can't play chess. They wouldn't remember the allowed moves and would make tons of illegal moves. They can recite the rules perfectly, hence they remember them, I didn't say he just saw the rules once, both the LLM and the human has seen the rules enough to recite them perfectly. Then a typical human can start to play chess based on those rules without having seen anyone play before, humans learn to play games by reading manuals all the time. You trying to argue against it here is absurd. > think about how many times you've sat down to learn a new board game and after the person who had played it before explains all the rules you still don't really get it I read the manual before playing board games because I like reading manuals, it is me teaching my friends then. You don't need someone to show you how to play games, reading the manual is enough...
- peteradio 3y ago
  Did anyone try what you proposed with the LLM? -- Give them lots of general context including other games but withhold any actual chess games. Given them the manual now see how it responds to a prompt to play chess. --
  - Jensson 3y ago
    If you give the chess manual during runtime it could work, plenty of examples of that working. If you give the chess manual during training time it wont work, which means that the LLM can't understand what it is trained on. But you could argue that the LLM can understand the data you give it during runtime, it is harder to argue against that, but it is pretty easy to say that it doesn't understand itself ie the data encoded in it. Maybe you need two models to understand things, since the models wont understand itself but a model can understand the data encoded in the next model. This isn't an unfixable problem, but it is a problem that has to be solved.

Jensson 3y ago

> If you sat someone down who had never played chess before, and taught them the rules, they also can't play chess. They wouldn't remember the allowed moves and would make tons of illegal moves. They can recite the rules perfectly, hence they remember them, I didn't say he just saw the rules once, both the LLM and the human has seen the rules enough to recite them perfectly. Then a typical human can start to play chess based on those rules without having seen anyone play before, humans learn to play games by reading manuals all the time. You trying to argue against it here is absurd. > think about how many times you've sat down to learn a new board game and after the person who had played it before explains all the rules you still don't really get it I read the manual before playing board games because I like reading manuals, it is me teaching my friends then. You don't need someone to show you how to play games, reading the manual is enough...

peteradio 3y ago
Did anyone try what you proposed with the LLM? -- Give them lots of general context including other games but withhold any actual chess games. Given them the manual now see how it responds to a prompt to play chess. --
- Jensson 3y ago
  If you give the chess manual during runtime it could work, plenty of examples of that working. If you give the chess manual during training time it wont work, which means that the LLM can't understand what it is trained on. But you could argue that the LLM can understand the data you give it during runtime, it is harder to argue against that, but it is pretty easy to say that it doesn't understand itself ie the data encoded in it. Maybe you need two models to understand things, since the models wont understand itself but a model can understand the data encoded in the next model. This isn't an unfixable problem, but it is a problem that has to be solved.