One of the best zuck interview! Must recommend!

o1-preview = early checkpoint of o1 o1-mini = latest checkpoint of o1-mini this is why its better. and especially since most people test it on math and code

What xjdr has been preaching on TL

Amazing that after few hours of prompting, that o1-preview has gotten quite close to the design choices that we have basically made in the past month to implement this new feature

I had to tell it what it was doing wrong, but i didn't tell what needs to be done to fix it. the fix were not simple, since each time they basically required reimagining almost the whole design. but it works

The pre-trainers use their divine skills to create and elevate the mind of the models but the post-trainers infuse life in them using a part of their souls this is why Gemini models do good on benchmarks but simply can't rizz the end user

Witnessing summarised CoT of o1 when I give it unclear instructions have made me realise how much my juniors have suffered when I simply added dyslexic ass PR comments

I don't know why people talk like this just work on harder problems bro instead of having to give up on AI coder cause it writes everything for you

I owe it to infra engineers at OpenAI to exhaust the rate limits of o1 their blood, sweat and tears in setting up H200s will not be wasted on processing salesforce tokens on 4o-mini