artifact
active
artifact:janus-transformer-introspection-postJanus' transformer introspection post
Twitter thread with infographics explaining information flow and recurrence in transformers, arguing LLMs can introspect.
Neighborhood — ranked by edge-count
Concepts (1)
concept
- Transformers are recurrent through autoregression because the K/V stream provides horizontal information flow across positions, even though each forward pass is feedforward.
Artifacts (1)
artifact
- The primary source paper, an interview article with Anima Labs members about language model phenomenology, published on smoothbrains.net and linked on LessWrong.