Using the advanced Elo calculations implemented in the ordo program,
and using the provided nominal Stockfish strengths as anchors,
I determined the following ratings from the games in this repo (omitting the isolated text-davinci-003 group):
# PLAYER : RATING POINTS PLAYED (%)
1 Stockfish_Elo2035 : 2035.0 16.0 20 80
2 Stockfish_Elo1954 : 1954.0 16.5 19 87
3 Stockfish_Elo1871 : 1871.0 187.5 252 74
4 Stockfish_Elo1785 : 1785.0 104.0 153 68
5 Stockfish_Elo1694 : 1694.0 20.5 40 51
6 gpt-3.5-turbo-instruct : 1682.9 193.0 485 40
7 Stockfish_Elo1597 : 1597.0 8.5 22 39
8 gpt-3.5-turbo : 1401.2 6.0 8 75
9 gpt-4 : 1377.5 63.0 123 51
10 RANDOM chess engine : 760.6 1.0 110 1
Note that accurately determining Elo for chess engine is a non-trivial task, as has long been recognized in the computer chess community. In particular, a simplistic tournament performance (FIDE lookup table) approach is not a good one.
Using the advanced Elo calculations implemented in the ordo program,
and using the provided nominal Stockfish strengths as anchors,
I determined the following ratings from the games in this repo (omitting the isolated text-davinci-003 group):
Note that accurately determining Elo for chess engine is a non-trivial task, as has long been recognized in the computer chess community. In particular, a simplistic tournament performance (FIDE lookup table) approach is not a good one.