Not Claude but ChatGPT - I asked it to pipe down on exactly that kind of response. And it did.
back
4 comments
GPT-5 ends every single response with something like.
> If you’d like, I can demonstrate…
or
> If you want…
and that's /after/ I put in instructions to not do it.
It's weird that it does that given that the leaked system prompt explicitly told it not to.
I've done this in my Claude settings, but it still doesn't seem that keen on following it:
> Please be measured and critical in your response. I appreciate the enthusiasm, but I highly doubt everything I say is “brilliant” or “astute”, etc.! I prefer objectivity to sycophancy.
Remove everything after .... 'in your response' and you will likely get better results.
> I appreciate the enthusiasm, but I highly doubt everything I say is “brilliant” or “astute”, etc.!
Is this part useful as instruction for a model? Seems targeted to a human. And even then I'm not sure how useful it would be.
The first and last sentence should suffice, no?
I wonder if asking it to respond in the style of Linus Torvalds would be an improvement.
Too much context tokens.
Yes, ChatGPT can do this, more or less.
Yet I'll tell it 100 times to stop using em dashes and it refuses.
What kind of monster would tell a LLM to avoid correct typography?
one that wants to hide he's using LLMs