News for hackers. Stuff that matters.
Re:Geoffrey Hinton leaves Google and can now speak freely ab... by jstanley on Monday May 01, 2023 @11:19AM (#35774028)
I played chess against ChatGPT just yesterday, and it got into a winning position against me. After 24 moves it tried to play an illegal move and then when I told it that's illegal it played a bad move, and after that it didn't manage to find any more legal moves (I gave up after asking it to try again about 10 times).

But it's very close to being able to play chess.

My prompt was:

> we're playing chess. only send chess moves.

>

> 1. e4

Re:Geoffrey Hinton leaves Google and can now speak freely ab... by zimpenfish on Monday May 01, 2023 @11:40AM (#35774382)
What does it do if you play illegal moves?
Re:Geoffrey Hinton leaves Google and can now speak freely ab... by jstanley on Monday May 01, 2023 @11:46AM (#35774500)
Good idea, I tried that just now. It just accepts it and carries on.
Re:Geoffrey Hinton leaves Google and can now speak freely ab... by TedDoesntTalk on Monday May 01, 2023 @12:30PM (#35775185)
It does the same when you ask it to be DM in a D&D game. It allows the players to do many, many things outside the rules. I don't remember any examples but a general idea was, "The character Frodo now has a the ability to breath fire. He breathes fire on the orcs." Although IIRC that was ChatGPT 3.5.
Re:Geoffrey Hinton leaves Google and can now speak freely ab... by sebzim4500 on Tuesday May 02, 2023 @05:47AM (#35785249)
I wonder if this would happen if you used the GPT-4 api with a system message. In my experience that can help but does necessarily solve the issue.
Re:Geoffrey Hinton leaves Google and can now speak freely ab... by Closi on Tuesday May 02, 2023 @05:50AM (#35785270)
You can overcome this with prompt engineering.

“If a player tries to do something not strictly within the rules of <insert game>, then you must inform me that it is an invalid move and not accept it”

GPT appears to be slightly tuned to default to “yes and” in these ‘creative’ situations by default rather than “block/deny”.

IMO lots of things where people don’t think GPT can do something end up being possible with basic prompt engineering. Usually people go too-short and too-non-specific on the prompt.

Re:Geoffrey Hinton leaves Google and can now speak freely ab... by alwaysbeconsing on Monday May 01, 2023 @11:58AM (#35774697)
I'd also be curious how it reacts if you try to say that a legal move it made is illegal.
Re:Geoffrey Hinton leaves Google and can now speak freely ab... by pmoriarty on Monday May 01, 2023 @11:53AM (#35774608)
Did you repeat the board position back to it after each move? LLMs have a limited context, so they might forget the board position after a while, unless they're reminded.
Re:Geoffrey Hinton leaves Google and can now speak freely ab... by jstanley on Monday May 01, 2023 @12:26PM (#35775114)
Nope, I just let it send me moves and I sent moves back.