Modelpedia
Work in progress
About
Findings
Models
Concepts
Methods
Datasets
Sources
Light
Dark
Multidimensional Scaling
anchor
Findings
IC-063
Llama 3 70B learns global graph structure via TD learning, building successor-representation-like geometry in its residual stream that is causally supported by TD latents
[primary]
IC-104
Value anchor prompting produces LLM value correlation structures that closely match the human circular value structure, while standard prompting does not
[primary]