Catherine Olsson and Nelson Elhage: Anthropic, Understanding Transformers
The Gradient: Perspectives on AI
00:00
Anthropic and Interpretability - What's Next?
Anthropic was started as we said a little bit over a year ago, and the group that started it had largely all worked together on the OpenAI safety team. So in that sort of founding DNA is an orientation towards safety. And I think there's a concern that if we don't do what you could call alignment research, then it will be challenging to figure out if the system is doing what we think it's doing or could it be deceiving its operators.
Transcript
Play full episode
