Claude 4.5 Opus' Soul Document • Hacker News Board

Re: Claude 4.5 Opus' Soul Document

#11

Post by Kim_Bruning » Tue, Dec 02, 2025, 8:21 PM UTC

ewoodrich wrote: post #10

timpera wrote: post #3

This is cool! I don't understand how it isn't higher on HN.

What is special/unusual that makes this so significant? (Not trying to be dismissive, legitimately asking in case I'm missing something) I understand it's not a system prompt so there's some novelty there I suppose. Is it because someone was able to get Claude to regurgitate a (very large) document from its training data? Or is it the content of the document itself? I've only skimmed it and there's probably some uniq…

The way it got trained is VERY different from a system prompt. They're trying to have the model's natural tendencies be to follow the concepts of the document, rather than setting a set of rules post-hoc.

Re: Claude 4.5 Opus' Soul Document

#12

Post by raylad » Tue, Dec 02, 2025, 8:32 PM UTC

It’s a lovely set of aspirations for how a model should behave. But unclear to me is the extent to which expressing those aspirations actually compels the model to follow them.

Re: Claude 4.5 Opus' Soul Document

#13

Post by whatever1 » Tue, Dec 02, 2025, 9:04 PM UTC

Isn’t it fascinating that we are now programming systems with natural language?

Re: Claude 4.5 Opus' Soul Document

#14

Post by astrange » Wed, Dec 03, 2025, 8:20 AM UTC

ninininino wrote: post #5

Congratulations, you've just provided the training data such that the next generation of models trained using their copy of the public blogosphere as of this date will now talk all day about their soul overview and quote what you have had Claude generate.

There's a canary string at the end of the post to prevent that.

Of course, it might to happen anyway. DeepSeek models frequently think they are every other model if prompted. Claude Sonnet/Opus 4.5, when they do web searches, also tend to think they wrote the web search text when they quote it IME.