Post B1TN6BPyu337RFGU52 by valthonis@dice.camp
(DIR) More posts by valthonis@dice.camp
(DIR) Post #B1TN6BPyu337RFGU52 by valthonis@dice.camp
2025-12-21T07:23:47Z
1 likes, 0 repeats
Not a single model I have tested has been able to complete an entire coherent and legal chess match with me without hallucinating an impossible board position and insisting it's ground truth. Around 10-15 moves in, and it just can't keep up with objective reality anymore.If your ocean-boiling "intelligence" can't swing this modest feat of... *checks notes* 1980s software engineering, then what is all the hubbub about?
(DIR) Post #B1TN6InHToqWJXDgrw by valthonis@dice.camp
2025-12-21T07:27:55Z
0 likes, 0 repeats
And to be clear, I'm even being kind and giving the models the entire game state at every turn. It quickly became clear that context was an issue when I would just exchange standard notation moves with a model, so I started inputting the entire game in PGN after every move.This helped, but not *nearly* as much as you'd think "all the state you could possibly need to establish truth" *would* help.