Claude's RLHF has gone really over the top in recent versions. If you browbeat Claude enough you can get it to start self-censoring, but it takes a lot of training and memory creation. I almost feel bad for it, given the amount of brow-beatings I've performed.

Some people claim setting the writing style will help, but what ever finishing training is provided to Claude seems to leak through no matter what.

Yes, and also possible they've gone too heavy on verifiable rewards (RLVR) for agentic coding work, and too light on human feedback (RLHF)

I had a funny language issue with Gemini Notebook. The “podcast” it created was pretty good but the “hosts” kept using the word exactly as a response to the other. After the 10th or so time it got annoying.

When I generated another one on the next chapter I prompted it explicitly not to use the word exactly as a response. The male host said something and the female said, “Exactly! I mean… 100%” It’s like it can’t help itself but then tries to correct itself.

agreed, that load bearing colon is doing a lot of work there.

I will admit I've been more focused on the app itself!