Evidence receipt / belief
Published · transcript-backedSholto Douglas: belief
28 Mar 2024 Dwarkesh Podcast Sholto Douglas & Trenton Bricken — How LLMs actually think
“Maybe not the user, but we should be able to understand and interpret what these values are doing and the information that is transmitting. I think that's a really important goal for the future.”
Source trail
Everything needed to verify it.
- Speaker
- Sholto Douglas
- Attribution
- Verified speaker
- Claim type
- belief
- Recorded
- 28 Mar 2024
- Publisher
- Dwarkesh Podcast
Transcript context
…About the sort of secret communication with the model to its forward inferences, how much steganography and secret communication do you expect there to be? We don't know. The honest answer is we don't know. I wouldn't even necessarily classify it as secret information. A lot of the work that Trenton's team is trying to do is to actually understand that these are fully visible from the model side. Maybe not the user, but we should be able to understand and interpret what these values are doing and the information that is transmitting. I think that's a really important goal for the future. There are some wild papers though where people have had the model do chain-of-thought and it is not at all representative of what the model actually decides its answer is. You can even go in and edit the chain-of-thought so that the reasoning is totally garbled and it will still output the true answer.…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.