method
active
method:chive-counterfactual-hypothesis-investigation-via-editsCHIVE (Counterfactual Hypothesis Investigation Via Edits)
Agentic pipeline that discovers unexpected model behaviors and investigates them with counterfactual prompt edits.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Counterfactual simulatabilityimplementsLens for evaluating explanations by whether they help predict behavior on counterfactual inputs.
Methods (4)
method
- Step 3 runs 5-15 counterfactual experiments.
- Step 1 of CHIVE.
- Step 2 flags unexpected behaviors with a 1-5 rubric and frozen classifier question.
- Step 4 judge scores support on 1-10 scale.
Artifacts (1)
artifact
- Released code and datasets for evaluation and training.
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.
- Ability to entertain competing hypotheses within one inference engine; proposed hallmark of mindful inference
- Encoding possible future or alternative states; bioelectric patterns can represent an alternative morphological target even in an intact animal.
- Pipeline retargetable to other behavior classes.
- The output value a model produces when an interchange intervention forces certain variables to take values from source inputs.
- The mental effort of holding models of how things could/should be different from actuality, contributing to compression stress.
- Pre-recorded latent vector encoding the expected causal variable values post-intervention; used as ground truth in the CLMAS auxiliary loss.
- True/false claims that a prompt edit shifts a behavior rate by at least 30pp; basis for all evaluation labels.
- The state a neural network is placed in when its activations are modified via intervention