framework
active
framework:latentqaLatentQA
Prior non-mechanistic approach of training LLMs to answer questions about their own activations; AOs generalize it.
Neighborhood — ranked by edge-count
Papers (1)
paper
Concepts (1)
concept
- Non-mechanistic interpretabilityassociated_withExplaining activations without understanding mechanism.
Frameworks (1)
framework
- Activation OraclesextendsFramework training LLMs to answer questions about externally-provided activation vectors
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Reasoning approach using learnable hidden embeddings.
- Baseline method using a single orthogonal matrix trained to map source latents to target latents via CL auxiliary loss without behavioral objective.
- Learned in pretraining; belief over them is what interventions change.
- Statistical regularities stored in pretrained models.
- Methods that use latent reasoning; lack task generalization and are difficult to train with autoregressive parallelization.
- Hidden or underdeveloped structures existing 'between the lines' of a configuration that can be enhanced and developed through harmony-seeking computation.
- Configurational entities existing implicitly in a structure; guide perception and generation of next morphogenetic step; exemplified in St Mark's square cycles.
- Entities that become visible as centers in a configuration (e.g., rectangles of white space around a dot) that were not present before.