Before o1, the models took the same amount of compute answer each question and had to respond immediately. “Of course, we should spend different amounts of compute, different amounts of energy on different problems.” – CEO of Cohere

→ View original post on X — @alliekmiller