Claude 4.5 Opus' Soul Document
lesswrong.com36 points by thm 12 hours ago
36 points by thm 12 hours ago
Isn’t it fascinating that we are now programming systems with natural language?
Now with even more significance that Amanda Askell has confirmed it is based on a document they trained Claude on (https://x.com/AmandaAskell/status/1995610567923695633).
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 unique nuggets here and there, but at a high level the rules/guidelines didn't really jump out at me as being much different than the various system prompts for proprietary models made public over the last couple years. Except much longer (and a bit ramble-y in some sections vs the directness of typical system prompt).
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.
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.
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.
Anyone got a TLDR? I'm too lazy to paste it in Claude.