Collin Burns at OpenAI

[Spencer131]

Weak-to-strong generalization

We present a new research direction for superalignment, together with promising initial results: can we leverage the generalization properties of deep learning to control strong models with weak supervisors?

openai.com

Collin Burns helped author this paper at OpenAI, and according to this tweet, was one of the main leaders of the paper:

Fun fact: Andrej Karpathy, known as badmephisto in the cubing world, also works at OpenAI. Clearly there is a cuber to OpenAI pipeline

 

[OtaMota]

Weak-to-strong generalization

We present a new research direction for superalignment, together with promising initial results: can we leverage the generalization properties of deep learning to control strong models with weak supervisors?

openai.com

Collin Burns helped author this paper at OpenAI, and according to this tweet, was one of the main leaders of the paper:

View attachment 28200

Fun fact: Andrej Karpathy, known as badmephisto in the cubing world, also works at OpenAI. Clearly there is a cuber to OpenAI pipeline

 

[OtaMota]

I know on the feliks Nr thread, I liked it when old cubers continue to stay in cubing and compete, but that does not need to be the only path they take. It is also good when cubing veterans find new hobbies or jobs they enjoy, and it's nice to see how successful they can be. There are many paths cubers can take.