On: Chess-GPT's Internal World Model

[jhrmnn]

They also say “I left one training for a few more days and it reached 1500 ELO.” I find it quite likely the observed performance is largely limited by the spent compute.