And here comes strawberry. Or rather the o1 little models from OpenAI. Of course, just like I thought, they were done by Noam Brown, the guy who did various things with game theory and was into all sorts of poker strategies and bots. A very respected person.
Inside, before it answers, it thinks via chain of thoughts. Also, by the way, they don’t beat 4o on every benchmark.
More details (https://x.com/polynoamial/status/1834280155730043108?s=46)
and
here (https://openai.com/index/learning-to-reason-with-llms/)
Original (RU)
А вот и strawberry. Точнее o1 модельки от OpenAI. Конечно же как я и думал делал их Noam Brown который делал разные штуки с теорией игр и занимался разными покерными стратегиями и ботами. Очень уважаемый человек.
Внутри перед тем как ответит думает через chain of thoughts. Не по всем кстати бенчам обгоняют 4о.
Подробнее (https://x.com/polynoamial/status/1834280155730043108?s=46)
и
тут (https://openai.com/index/learning-to-reason-with-llms/)