From OpenAI's RLHF paper[1]: "By default, when we train a PPO model on our API distribution, it suffers from an “alignment tax”, as its performance on several public NLP datasets decreases." On the HELM[2] site, you can see accuracy benchmarks for InstructGPT <OpenAI model> vs baseline models. The InstructGPT models perform worse on a lot of benchmarks.
It's the "alignment tax". From OpenAI's RLHF paper[1]: "By default, when we train a PPO model on our API distribution, it suffers from an “alignment tax”, as its performance on several public NLP datasets decreases." On the HELM[2] site, you can see accuracy benchmarks for InstructGPT <OpenAI model> vs baseline models. The InstructGPT models perform worse on a lot of benchmarks.
You don't have to care about your job. Most people have about four good hours of work in them. Just do your four hours, and take the rest of the day off. No job is worth burnout.
You should find a 100% remote job. If other people are there full time, but you're only there 0-2 days a week, you're never going to be that close with them. You said you're happy when you're at home, but going into the office makes you anxious. So you shouldn't have an issue working at home full time.
David Heinemeier Hansson gave a great talk on air quality and touched upon various HVAC systems. One thing that really stuck out, was his claim that you need 80 CFM per person per room.
https://www.youtube.com/watch?v=MRqh8oLY7Ik