4 ms·
Yeah, I figure this is also why it often says “Ah, I found the problem! Let me check the …”. It hasn’t found the problem, but it’s more likely to continue with
by nojs 1y ago
Yeah, I figure this is also why it often says “Ah, I found the problem! Let me check the …”. It hasn’t found the problem, but it’s more likely to continue with the solution if you jam that string in there.
- adastra22 1y agoWe don’t know how Claude code is internally implemented. I would not be surprised at all if they literally inject that string as an alternative context and then go with the higher probability output, or if RLHF was structured in that way and so it always generates the same text.
- data-ottawa 1y agoVery likely RLHF, based only on how strongly aligned open models repeatedly reference a "policy" despite there being none in the system prompt. I would assume that priming the model to add these tokens ends up with better autocomplete as mentioned above.
- steveklabnik 1y agoClaude Code is a big pile of minified Typescript, and some people have effectively de-compiled it.
- sejje 1y agoSo how does it do it?
- steveklabnik 1y agoI haven't read this particular code, I did some analysis of various prompts it uses, I didn't hear about anything specific like this. Mostly wanted to say "it's at least possible to dig into it if you'd like," not that I had the answer directly.
- Aeolun 1y agoCouldn’t you have claude itself de-minify it?
- steveklabnik 1y agoMaybe. It’s not something I have enough of an interest in to out the time into trying it out.