

Well yes, and I think it’s probably true, you can use one model to train another, effectively cloning it. It seems like it’s much more expensive to make the first round of good models. Part of the argument could be that openai/anthropic are actually still driving the frontier, developing the researchers and expertise to do that, and continuing to innovate the tech

Yeah sure you can do that but people won’t listen. What I mean by thought police is actually doing policing, like “no you can’t say that or we will arrest you” which goes against the idea of freedom of speech. It’s reminiscent of when ussr was jailing karate teachers