Direct Nash Optimization: Teaching language models to self-improve with general preferences
1 945
8.1
Microsoft Research335 тыс
Следующее
109 дней – 1 2556:36Популярные
33 дня – 1 7473:56
237 дней – 6383:07:51Опубликовано 3 сентября 2024, 18:59
Corby Rosset, Senior Researcher, Microsoft Research AI Frontiers, discusses teaching language models to self-improve using a preference oracle like GPT-4, framing it as a two-player game to find an optimal policy at a Nash equilibrium, and achieving state-of-the-art win rates against GPT-4 Turbo on benchmarks such as Alpaca-Eval and MT-Bench.
Microsoft Research Forum, September 3, 2024
See more at aka.ms/ResearchForum-Sep2024
Microsoft Research Forum, September 3, 2024
See more at aka.ms/ResearchForum-Sep2024
Свежие видео
5 дней – 17516:22
8 дней – 1 824 97110:16Случайные видео
227 дней – 3 8026:51
26.09.23 – 3470:28
06.04.23 – 16 8184:36
29.04.22 – 37 3170:16
29.04.20 – 210:23
28.08.14 – 251 9809:25
1 день – 104 8430:38
2 дня – 283:50
5 дней – 13 6253:18
9 дней – 61 3862:50
9 дней – 1164:36
10 дней – 328 4794:32
11 дней – 4374:32