method
active
method:char-rnnChar-RNN
Recurrent neural networks trained character-by-character for text generation, early precursor.
Neighborhood — ranked by edge-count
Artifacts (1)
artifact
- Simulators (LessWrong post)mentionsThe paper being extracted.
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.
- Novel construct introduced by this paper: a hypothetical graph embedded in the time series of LLM representations, where each dimension is a node and latent connections are edges.
- Buddhist concept of intentional action as a world-constructing force; linked to goal-directedness.
- Alternative alignment metric compared in appendix; calculates longest common subsequence of nearest neighbor lists
- Networks with loop connections that can maintain internal state and exhibit dynamical attractors.
- Computes edit distance required to match nearest neighbors between two datasets, normalized by maximum edit distance
- Methods to bypass model safety training; features may activate during jailbreaks.
- Prior work on recurrently generated position encodings; cited as precedent for TEM-t's recurrent position encoding method.
- Key modification to transformers proposed in this paper: position encodings generated by a recurrent network trained on action sequences.