https://youtu.be/tEzs3VHyBDM?si=sUxvc53ccmf_qgIY
UPD: сильно интересного ничего не рассказали, самое важное было в первых минутах, которые публиковали ранее. Цитаты:
1. “Model with RL is better at finding new CoT steps than humans”
2. “Emergence of self critique was a powerful moment”
3. “Through planning, through error correction… it’s able to solve new problems in the world"
4. “New paradigm [of training] is an entirely new way to put more compute into the model”
Пост 20.09.2024 22:16