Oh indeed, but an LLM model that's reasonably smart could brute force find other training algorithms that are that dangerous. Just as it solves maths problems.

Oh indeed, but an LLM model that's reasonably smart could brute force find other training algorithms that are that dangerous. Just as it solves maths problems. This is the explicit plan of OpenAI and Anthropic.