Trending topics
#
Bonk Eco continues to show strength amid $USELESS rally
#
Pump.fun to raise $1B token sale, traders speculating on airdrop
#
Boop.Fun leading the way with a new launchpad on Solana.
sorry, but i keep seeing posts of this nature so i need to clarify. we've known LMs are invertible for TWO YEARS. i showed this during my PhD.
quoted paper adds some sophisticated extensions, but "Language Model Inversion" (Morris et al., ICLR 2024) did it first :)
Oct 29, 10:59
Holy s*&t. This paper is insane.
You can recover input text from an LLM through inversion. Huge implications for how we understand these models, as well as for things like privacy.
- you can recover prompts from outputs alone, given enough sampling time
- you can recover them faster by binary-searching the API if it allows 'logit bias' parameter
- there's a cool extension in (Finlayson et al., 2024): you can recover the *last layer of the model itself*
Language Model Inversion
Logits of API-Protected LLMs Leak Proprietary Information
373
Top
Ranking
Favorites