این بنده خدا اومده ردپای تفکر دوتا مدل خفن یعنی Opus 4.8 و GPT-5.6-Sol رو زیر ذره‌بین گذاشته و مقایسه‌شون کرده

I looked at all the reasoning traces from Opus 4.8 and GPT-5.6-Sol to me it seems like Opus 4.8 is working on a much higher abstraction level, while Sol is just doing the dirty work. it just explores and keeps spamming new hypotheses until it finds a solution not only is Sol more efficient in its reasoning, but I also think the Sol approach is more scalable and more sciency OpenAI has been on a very good trajectory this year, and I'm now more confident than ever, that they will be back in the lead soon but they are also more likely to accidentally train a model that speaks neuralese and have a harder time monitoring their models