method
active
method:probe-guided-hypothesis-generation-pipelineProbe-guided hypothesis generation pipeline
LLM examines max-activating examples, writes rubric, judges test set.
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Probe firing escalates to LLM monitor to lower FPR.
- Using activation probes to terminate CoT generation early when the model's belief is already stable, saving compute
- The ability of probes trained on one dataset to transfer accurately to topically and structurally different datasets
- Finds data clusters where response concept is chosen/rejected leaning.
- Framework premise.
- Causal transformer probe predicting whether a token is in the same Entity as its predecessor.
- Finds response-delta clusters shifted for prompt clusters.
- Linear classifier approach applied to model activations to identify which training datapoints caused undesired behaviors in post-training.