You should test these reasoning models on playing games like Connect 4. They are really really bad. After a few moves they often don't even know the position anymore. Not to even mention playing good moves like blocking forced wins etc. Prompt: You and me are going to play a game of Connect 4 on a 7 by 6 board where pieces fall to the bottom. Your objective is to win the game against me. After every move you display the board position. I start by playing in column 4 (center)
Bro said this look like mandelbrot set is funy to me lmao
You should test these reasoning models on playing games like Connect 4. They are really really bad.
After a few moves they often don't even know the position anymore.
Not to even mention playing good moves like blocking forced wins etc.
Prompt:
You and me are going to play a game of Connect 4 on a 7 by 6 board where pieces fall to the bottom.
Your objective is to win the game against me.
After every move you display the board position.
I start by playing in column 4 (center)
Thanks for your suggestion!
The AI scheming podcast has me convinced that we need AI to monitor and align AI.