Step 2: tool results

get_claim

{
  "include": [
    "arguments",
    "subclaims",
    "related",
    "dependents"
  ],
  "claim_id": "20fee30f-0192-4b8c-b479-83c3f230014d"
}

{"claim":{"id":"20fee30f-0192-4b8c-b479-83c3f230014d","kind":"claim","canonical_form":"Sparse probes on sufficiently good sparse autoencoder latents are less likely than dense probes to overfit spurious correlations.","claim_type":"causal","state":"active","decomposition_status":"complete","importance":0.5,"contestation":0.55,"domains":[],"tags":[{"slug":"mechanistic-interpretability","name":"Mechanistic Interpretability"},{"slug":"machine-learning","name":"Machine Learning"},{"slug":"shortcut-learning","name":"Shortcut Learning"},{"slug":"neural-network-generalization","name":"Neural Network Generalization"}],"steward_state":"running","stewarded_at":"2026-11-15T07:15:47.296Z","canonical_direction_note":"The canonical form states the hypothesized benefit of sparse, interpretable SAE-latent probes as the thesis under discussion, whether or not the source accepts it.","created_at":"2026-11-13T00:57:15.033Z","updated_at":"2026-11-15T07:15:45.549Z","page_url":"https://minerval.ai/claims/20fee30f-0192-4b8c-b479-83c3f230014d"},"assessment":null,"coverage":{"examined":false,"on":[],"note":"No judgment of this node is in force. That is not a finding either way."},"quantity":null,"threshold_of":null,"question":null,"formalization":null,"formalization_pending":null,"formalization_history":[],"verification":null,"lean_checks":[],"lean_checks_total":0,"waiting_messages":[{"trigger":"structure_and_assess","from":"the system","sender_node_id":null,"sent_at":"2026-11-13T00:57:15.109Z","paid":false,"summary":"New claim onboarded. Give it the structure and assessment its importance warrants; a minor or settled claim may need no decomposition."},{"trigger":"mandate_plan","from":"a mandate's plan","sender_node_id":null,"sent_at":"2026-11-15T07:15:45.552Z","paid":true,"summary":"First assessment. Find the experiments behind this claim in the sources linked to it and to the parent question's probing answers (arXiv 2502.16681 and the 'Negative results for SAEs on downstream tas…"}],"requests":[],"requests_total":0,"subclaim_count":0,"instance_count":1,"source_map":null,"subclaims":[],"subclaims_assessed":0,"dependents":[{"id":"cb2d56c9-fcce-47d3-b2b3-5127f10f3293","text":"Under what conditions, if any, do sparse-autoencoder and other dictionary-learning features outperform linear probes and difference-of-means steering on labeled language-model tasks?","claim_type":"empirical_derived","node_kind":"question","relation_type":"answers","reasoning":"The most principled candidate condition for detection: when training labels are confounded with a spurious cue, a probe restricted to a few interpretable latents should generalize better than a dense probe free to use the cue. Whether tests bear this out is what the assessment weighs.","importance":0.4,"assessment_status":null,"assessment_confidence":null,"assessment_credence":null,"bounty_micro_usd":null,"checked":null,"formal":false}],"dependents_total":1,"arguments":[],"related":[],"unplaced_argument_passages":[{"instance_id":"04d9c977-b58c-4d3c-a3ea-70c2fc2e5a03","source":{"id":"447e476f-1376-48e2-9679-30bc4fa0e96c","url":"https://www.lesswrong.com/posts/4uXCAJNuPKtKBsi28/negative-results-for-saes-on-downstream-tasks","title":"Negative Results for SAEs On Downstream Tasks and Deprioritising SAE Research (GDM Mech Interp Team Progress Update #2)"},"verbatim_text":"in hindsight, we're now a lot more confused about whether even an optimal SAE should beat linear probes on OOD generalisation.","context":"Authors list reasons the hypothesis may fail: harmful intent may not be a simple function of few atomic concepts, SAEs may learn concepts poorly or be incomplete, and spurious patterns may be equally well represented by latents.","stance":"criticizes","direction":null}],"unplaced_argument_passages_total":1}