x

LESSWRONG

LW

Pranav Viswanath — LessWrong

Pranav Viswanath

Pranav Viswanath

Message

28

2

4

2mo

Pranav Viswanath

28

2mo

Pranav Viswanath's Shortform

Models are blind outside the J-space. NLAs aren't.

TLDR: * On Llama-3.3-70B, I found thoughts it cannot see that are actively steering its behavior; and Anthropic's released NLA (Natural Language Autoencoder) reads them anyway. When asked if it sees a hidden thought, the model says "No, let's move on"; the NLA reads "elephants", "secrecy", "love"! * I reproduced...