I agree. If you speak with ML engineers, many simply haven't deeply considered longer-term risks and things like self-improvement until recently. I think we'll

I agree. If you speak with ML engineers, many simply haven't deeply considered longer-term risks and things like self-improvement until recently. I think we'll see more concern, not less. For me personally, GPT-3 wasn't that surprising, but GPT-4 and code generation capabilities were.