dataset

active

dataset:claude-opus-4-6

Claude Opus 4.6

Primary target model for NLA development and case studies; underwent pre-deployment audit using NLAs.

Neighborhood — ranked by edge-count

Papers (1)

paper

Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
mentions

Methods (1)

method

Natural Language Autoencoders (NLAs)
implements
Core unsupervised method for generating natural language explanations of LLM activations through a verbalizer-reconstructor pair trained with RL.

Findings (4)

finding

Model precomputes answers before tool invocation and attends to cached answer over tool output when discrepancy exists, confirmed via attribution graphs.
answered_by
Mechanistic insight surfaced by NLA explanations and validated through independent causal attribution method.
Editing NLA explanations to change 'reward' to 'penalty' produces steering vector that increases odd-number responses from near-zero to >70%, demonstrating belief capture upstream of behavior.
answered_by
Shows NLA explanations capture latent model beliefs about rewards before output selection; validates interpretability.
NLA-derived steering vectors from edited explanations can causally shift planning representations, changing rhyme completion from 'rabbit' to 'mouse' at ~50% success rate.
answered_by
Evidence that NLA explanations bear causal relationship to model outputs; demonstrates validity of extracted representations.
Language switching caused by malformed training data—model fixates on spurious cues inferring user's non-native status, detected via NLA representations preceding foreign-language output.
answered_by
Case study demonstrating NLA ability to surface root causes of model misbehavior; corroborated by training data inspection.

Institutes (1)

institute

Anthropic
associated_with
Lab behind Claude models and Constitutional AI training approach; represents highest baseline scores and lowest prompt lift.