undefined | Better HN

0 pointsvintermann3y ago0 comments

The Bing LM, or rather the service, did have "inner monologue" in the sense of text that it would generate, but not show to the user, and treat as "thoughts" to guide the generation of an actual reply that the user would see.

We know this because it happily told us, including the json format it uses internally.

0 comments

4 comments · 2 top-level

beepbooptheory3y ago· 2 in thread

And you trust what it told you?

int_19h3y ago

No, but the reconstructed examples have "im_start" and "im_end", which strongly implies that it is, if not verbatim, then a close enough restatement of the real deal. Take a look:

https://www.make-safe-ai.com/is-bing-chat-safe/Prompts_Conve...

vintermannOP3y ago

Yup, for the same reason I trust e.g. jailbreaks exposing the prompt: it was consistent.

Really, just asking again is a fine way to expose all sorts of "hallucinations" in a LM.

Vespasian3y ago

Interesting. I didn't know that.

When using gpt-4 directly through the API we can emulate this behavior

j / k navigate · click thread line to collapse