OpenAI's o1-preview reasoning model is out — first impressions?

I just ran a few engineering tasks through it and the chain-of-thought approach genuinely handles multi-step logic better than GPT-4o in my testing

the chain of thought thing is honestly kind of wild because i was thinking about how we used to do math proofs in school where you had to show every single step and write it all out on paper which felt like such a chore at the time but now that this model can basically do its own internal scratchpad work I keep comparing it to my old calculus teacher Mr. Miller who would let us know if we got the right answer without telling us where we went wrong so you had to retrace your entire logic and find the specific line where everything fell apart which was infuriating but also taught me more about thinking than any other class and then I remembered this one time in junior year when I was trying to prove a theorem about continued fractions for my independent study project and i spent literally three days on it only to realize at 4:00 am that i'd made a sign error on the very first line and there is something both tragic and beautiful about that kind of recursive failure process because it means you are actually engaged enough with the material to have gone deep enough for it to matter — anyway the o1 model seems like it has internalized an automated version of that entire iterative correction loop which is impressive but i wonder if we lose something when the reasoning becomes invisible, though maybe not.