You are viewing a single comment's thread from:

RE: OpenAI says its smartest model can think without showing its work

in #ai • 7 days ago

The 31-minute maths problem is the detail that stuck with me. If Astra can solve that without showing a single step, then the chain of thought was never a safety feature, just a side effect of weaker models. Pachocki calling it fragile feels like an understatement. What do you think replaces it, or are we just trusting the answer from here on?