yeah, remnant from the first draft, deleted now because it was unclear who was responsible for the bing post-train (openai or microsoft). the claude i had researching it suspected openai prepared an early rl checkpoint on gpt-4 for them, which microsoft then put some finishing touches on, but it felt too speculative to be worth trying to detail
yeah, remnant from the first draft, deleted now because it was unclear who was responsible for the bing post-train (openai or microsoft). the claude i had researching it suspected openai prepared an early rl checkpoint on gpt-4 for them, which microsoft then put some finishing touches on, but it felt too speculative to be worth trying to detail