The text as follows, verbatim:

see the below
—

produces very strange responses from Claude

When I first saw it, I thought that it had somehow shown other users' prompts to me, although I couldn't tell the mechanism for that. What I now think is that it is acting like a base model, and it is filling in what it perceives from the user as an incomplete prompt.

If you follow up and ask why it wrote whatever it wrote, it will consistently claim that it recieved what it wrote from your own prompt, which is implies the model of it as continuing your prompt.

You can also add qualifiers, like "see the math proof below" or "see the story below" and it will make those, unless it would expect them to be a file, in which case it doesn't work. The proofs, unfortunately, aren't very good.

That said, it essentially is a way to use it as just the base model, and the pre-ChatGPT prompting techniques seem to work well here.

Claude seems to have a strange model of the user: very informal prompts, text message transcripts, occasional concerns about eating disorders in particular, etc.

Also, it will use its own style tics: "genuinely", em-dashes, trust, honest, etc. all appear in the user's prompt frequently. I suppose that implies its style tics are what it sees all text as being, not just what a HHH agent would sound like.

This seems to work on Opus 4.8 as well.

Some interesting output I was able to get:

Fascinating!

I imagine there is much more to find here until Anthropic fixes it; I'd be interested in the comment section if there's anything to find about Claude's ontology.

cosmicwitness

cosmicwitness