{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:S24J2ZRYOPNRUMGRWMC2XWLCER","short_pith_number":"pith:S24J2ZRY","schema_version":"1.0","canonical_sha256":"96b89d663873db1a30d1b305abd962245601016258ce1e824a68370f027e2c5d","source":{"kind":"arxiv","id":"2605.21988","version":1},"attestation_state":"computed","paper":{"title":"Learning Spatiotemporal Sensitivity in Video LLMs via Counterfactual Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Bohai Gu, Dazhao Du, Eric Liu, Fangqi Zhu, Jialong Qin, Jian Liu, Song Guo, Tao Han, Xi Chen, Yujia Zhang","submitted_at":"2026-05-21T04:38:02Z","abstract_excerpt":"Video large language models (Video LLMs) achieve strong benchmark accuracy, yet often answer video questions through shortcuts such as single-frame cues and language priors rather than by tracking spatiotemporal dynamics. This issue is exacerbated in RL post-training, where correctness-only rewards can further reinforce shortcut policies that obtain high reward without tracking video dynamics. We address this by asking a controlled counterfactual question: if the visual world changed while the question remained fixed, should the answer change or stay the same? Based on this view, we propose \\t"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.21988","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-21T04:38:02Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c2e62462dfe5afea511e59cf714bbb7c71a92a95f870ae9fa9f78267d3d686ef","abstract_canon_sha256":"4f60bd0fe7b962ea639131be20573e3132f77856c26b087ac402ed477bee5bfc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-22T01:04:19.070178Z","signature_b64":"GhmvU8pzHOfNmz6FTqjyQJ9Z08qBuqZxPwtMHiRXUe2+IzUJ2D88Hg8eSpO1zIflDL9O3AHLg4Mw2IftIX8lCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"96b89d663873db1a30d1b305abd962245601016258ce1e824a68370f027e2c5d","last_reissued_at":"2026-05-22T01:04:19.069402Z","signature_status":"signed_v1","first_computed_at":"2026-05-22T01:04:19.069402Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning Spatiotemporal Sensitivity in Video LLMs via Counterfactual Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Bohai Gu, Dazhao Du, Eric Liu, Fangqi Zhu, Jialong Qin, Jian Liu, Song Guo, Tao Han, Xi Chen, Yujia Zhang","submitted_at":"2026-05-21T04:38:02Z","abstract_excerpt":"Video large language models (Video LLMs) achieve strong benchmark accuracy, yet often answer video questions through shortcuts such as single-frame cues and language priors rather than by tracking spatiotemporal dynamics. This issue is exacerbated in RL post-training, where correctness-only rewards can further reinforce shortcut policies that obtain high reward without tracking video dynamics. We address this by asking a controlled counterfactual question: if the visual world changed while the question remained fixed, should the answer change or stay the same? Based on this view, we propose \\t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.21988","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.21988/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.21988","created_at":"2026-05-22T01:04:19.069523+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.21988v1","created_at":"2026-05-22T01:04:19.069523+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.21988","created_at":"2026-05-22T01:04:19.069523+00:00"},{"alias_kind":"pith_short_12","alias_value":"S24J2ZRYOPNR","created_at":"2026-05-22T01:04:19.069523+00:00"},{"alias_kind":"pith_short_16","alias_value":"S24J2ZRYOPNRUMGR","created_at":"2026-05-22T01:04:19.069523+00:00"},{"alias_kind":"pith_short_8","alias_value":"S24J2ZRY","created_at":"2026-05-22T01:04:19.069523+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/S24J2ZRYOPNRUMGRWMC2XWLCER","json":"https://pith.science/pith/S24J2ZRYOPNRUMGRWMC2XWLCER.json","graph_json":"https://pith.science/api/pith-number/S24J2ZRYOPNRUMGRWMC2XWLCER/graph.json","events_json":"https://pith.science/api/pith-number/S24J2ZRYOPNRUMGRWMC2XWLCER/events.json","paper":"https://pith.science/paper/S24J2ZRY"},"agent_actions":{"view_html":"https://pith.science/pith/S24J2ZRYOPNRUMGRWMC2XWLCER","download_json":"https://pith.science/pith/S24J2ZRYOPNRUMGRWMC2XWLCER.json","view_paper":"https://pith.science/paper/S24J2ZRY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.21988&json=true","fetch_graph":"https://pith.science/api/pith-number/S24J2ZRYOPNRUMGRWMC2XWLCER/graph.json","fetch_events":"https://pith.science/api/pith-number/S24J2ZRYOPNRUMGRWMC2XWLCER/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/S24J2ZRYOPNRUMGRWMC2XWLCER/action/timestamp_anchor","attest_storage":"https://pith.science/pith/S24J2ZRYOPNRUMGRWMC2XWLCER/action/storage_attestation","attest_author":"https://pith.science/pith/S24J2ZRYOPNRUMGRWMC2XWLCER/action/author_attestation","sign_citation":"https://pith.science/pith/S24J2ZRYOPNRUMGRWMC2XWLCER/action/citation_signature","submit_replication":"https://pith.science/pith/S24J2ZRYOPNRUMGRWMC2XWLCER/action/replication_record"}},"created_at":"2026-05-22T01:04:19.069523+00:00","updated_at":"2026-05-22T01:04:19.069523+00:00"}