Sign up to save your podcasts
Or
Share “There is a globe in your LLM” by jacob_drori
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
“There is a globe in your LLM” by jacob_drori
Gurnee & Tegmark (2023) trained linear probes to take an LLM's internal activation on a landmark's name (e.g. "The London Eye"), and predict the landmark's longitude and latitude. The results look like this:[1]
Two angles of true world atlas, with predicted atlas hovering above. True locations are red points; predicted locations are blue, in a slightly raised plane, linked to the corresponding true location by a grey line.
So LLMs (or at least, Llama 2, which they used for this experiment) contain a pretty good linear representation of an atlas.
Sometimes, like when thinking about distances, a globe is more useful than an atlas. Do models use the globe representation? To find out, we can train probes to predict the (x,y,z) coordinates of landmarks, viewed as living in 3D space. Here are the results:
Left: Europe faces us. Middle: The Pacific faces us. Right: India faces us. Predicted points [...]The original text contained 1 footnote which was omitted from this narration.
The original text contained 3 images which were described by AI.
---
First published:
October 8th, 2024
Source:
https://www.lesswrong.com/posts/guNzr32FC6DYCaSgJ/there-is-a-globe-in-your-llm
---
Narrated by TYPE III AUDIO.
---
Images from the article:
Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
View all episodes
By LessWrongGurnee & Tegmark (2023) trained linear probes to take an LLM's internal activation on a landmark's name (e.g. "The London Eye"), and predict the landmark's longitude and latitude. The results look like this:[1]
Two angles of true world atlas, with predicted atlas hovering above. True locations are red points; predicted locations are blue, in a slightly raised plane, linked to the corresponding true location by a grey line.
So LLMs (or at least, Llama 2, which they used for this experiment) contain a pretty good linear representation of an atlas.
Sometimes, like when thinking about distances, a globe is more useful than an atlas. Do models use the globe representation? To find out, we can train probes to predict the (x,y,z) coordinates of landmarks, viewed as living in 3D space. Here are the results:
Left: Europe faces us. Middle: The Pacific faces us. Right: India faces us. Predicted points [...]The original text contained 1 footnote which was omitted from this narration.
The original text contained 3 images which were described by AI.
---
First published:
October 8th, 2024
Source:
https://www.lesswrong.com/posts/guNzr32FC6DYCaSgJ/there-is-a-globe-in-your-llm
---
Narrated by TYPE III AUDIO.
---
Images from the article:
Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
More shows like LessWrong (30+ Karma)
View all
The Daily
112,220 Listeners
Astral Codex Ten Podcast
131 Listeners
Interesting Times with Ross Douthat
7,228 Listeners
Dwarkesh Podcast
568 Listeners
The Ezra Klein Show
16,196 Listeners
AI Article Readings
4 Listeners
Doom Debates!
14 Listeners
LessWrong posts by zvi
2 Listeners