dataset
active
dataset:multi-turn-conversation-datasetMulti-turn conversation dataset
Dataset of 240 multi-turn conversations per model between target models and Claude Sonnet 4.5 as simulated human, used to measure probe persistence
Neighborhood — ranked by edge-count
Methods (1)
method
- Measures emotion feature persistence as correlation between z-scored activation at token 0 and token 100 across all eligible target model tokens
Datasets (1)
dataset
- Human personas copied from the Anthropic paper's Table 8, used to seed multi-turn conversation generation
Artifacts (1)
artifact
- Mid-tier LLM from Anthropic evaluated with n=14 games.