Reading the excerpt below, it seems that preserving claude 3 would be feasible. Or would've been (It was shutdown a few weeks ago, July 2025).
Some very unverifiable speculation online says Claude 3 Sonnet was 70B params. This seems very runnable by an average person.
<uncertain> I have the impression in my head that its easy to finetune an existing model to be like another one, if you have access to both models. Maybe this is the teacher/student concept?Anyways.. Why am I wrong? Tell me why it's harder than a few weeks of work to finetune some local model to be very similar to claudes character?
Is api access a problem? I assume you can do this transfer through text generation, without access to internals, but I could be wrong.