{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:CW7IWZ46YM7VWQNJOBNGPBBPUG","short_pith_number":"pith:CW7IWZ46","canonical_record":{"source":{"id":"2601.15197","kind":"arxiv","version":6},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-01-21T17:15:22Z","cross_cats_sorted":["cs.CL","cs.CV","cs.RO"],"title_canon_sha256":"a2f940e430f04da099d006def40320a211a8eaabe42cdfdf8b048208a6a7d653","abstract_canon_sha256":"5d8f3b3c3e63e88ea90b00a8b588f3289b724b72ca3ba6584ba3dfe04fad01cd"},"schema_version":"1.0"},"canonical_sha256":"15be8b679ec33f5b41a9705a67842fa1981380046c4fd4fb2995a6d2eef2fecf","source":{"kind":"arxiv","id":"2601.15197","version":6},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2601.15197","created_at":"2026-05-18T03:09:24Z"},{"alias_kind":"arxiv_version","alias_value":"2601.15197v6","created_at":"2026-05-18T03:09:24Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.15197","created_at":"2026-05-18T03:09:24Z"},{"alias_kind":"pith_short_12","alias_value":"CW7IWZ46YM7V","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"CW7IWZ46YM7VWQNJ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"CW7IWZ46","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:CW7IWZ46YM7VWQNJOBNGPBBPUG","target":"record","payload":{"canonical_record":{"source":{"id":"2601.15197","kind":"arxiv","version":6},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-01-21T17:15:22Z","cross_cats_sorted":["cs.CL","cs.CV","cs.RO"],"title_canon_sha256":"a2f940e430f04da099d006def40320a211a8eaabe42cdfdf8b048208a6a7d653","abstract_canon_sha256":"5d8f3b3c3e63e88ea90b00a8b588f3289b724b72ca3ba6584ba3dfe04fad01cd"},"schema_version":"1.0"},"canonical_sha256":"15be8b679ec33f5b41a9705a67842fa1981380046c4fd4fb2995a6d2eef2fecf","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:09:24.337849Z","signature_b64":"1vTHYmnF5OOzfPdgqTe2nFLSyahdghFqHuSMtcixZCuLaMeb2sywIJ4zo1JDSYVWRhD8rhZvEqCmrfGUfvJDBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"15be8b679ec33f5b41a9705a67842fa1981380046c4fd4fb2995a6d2eef2fecf","last_reissued_at":"2026-05-18T03:09:24.337228Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:09:24.337228Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2601.15197","source_version":6,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:24Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"fZIZo0aAVc0WCWr65tFkKcPFbA2G/NEBxIMHvPp4Er+a4fw4ABIlrwehKURAaCyMPkOd1lE0Ub525YydKysoAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-10T09:54:04.236875Z"},"content_sha256":"478e3c2912f09d6f290d35383dbfbb64baeba022920266f2ebfcc4d4ee9acc35","schema_version":"1.0","event_id":"sha256:478e3c2912f09d6f290d35383dbfbb64baeba022920266f2ebfcc4d4ee9acc35"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:CW7IWZ46YM7VWQNJOBNGPBBPUG","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"LangForce: Bayesian Decomposition of Vision Language Action Models via Latent Action Queries","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"LangForce restores language grounding in vision-language-action models by maximizing conditional pointwise mutual information between instructions and actions.","cross_cats":["cs.CL","cs.CV","cs.RO"],"primary_cat":"cs.AI","authors_text":"Bin Yu, Changti Wu, Cong Huang, Kai Chen, Laurence T. Yang, Shijie Lian, Xiaopeng Lin, Yuzhuo Miao, Zhaolong Shen","submitted_at":"2026-01-21T17:15:22Z","abstract_excerpt":"Vision-Language-Action (VLA) models have shown promise in robot manipulation but often struggle to generalize to new instructions or complex multi-task scenarios. We identify a critical pathology in current training paradigms where goal-driven data collection creates a dataset bias. In such datasets, language instructions are highly predictable from visual observations alone, causing the conditional mutual information between instructions and actions to vanish, a phenomenon we term Information Collapse. Consequently, models degenerate into vision-only policies that ignore language constraints "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Without requiring new data, LangForce significantly improves generalization... including an 11.3% improvement on the challenging OOD SimplerEnv benchmark.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the identified information collapse is the dominant cause of poor OOD performance and that maximizing conditional PMI via the dual-branch architecture will reliably correct it without introducing new failure modes or requiring careful hyperparameter tuning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LangForce prevents vision-language-action models from ignoring language by maximizing conditional pointwise mutual information via Bayesian decomposition with latent action queries.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"LangForce restores language grounding in vision-language-action models by maximizing conditional pointwise mutual information between instructions and actions.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9355ce0d1517b1d25726560d9f452f287cbbeb6630412fe28969e01a72e53486"},"source":{"id":"2601.15197","kind":"arxiv","version":6},"verdict":{"id":"44b4db63-88ca-4edb-a2c5-f47d5236710c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T11:56:02.585654Z","strongest_claim":"Without requiring new data, LangForce significantly improves generalization... including an 11.3% improvement on the challenging OOD SimplerEnv benchmark.","one_line_summary":"LangForce prevents vision-language-action models from ignoring language by maximizing conditional pointwise mutual information via Bayesian decomposition with latent action queries.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the identified information collapse is the dominant cause of poor OOD performance and that maximizing conditional PMI via the dual-branch architecture will reliably correct it without introducing new failure modes or requiring careful hyperparameter tuning.","pith_extraction_headline":"LangForce restores language grounding in vision-language-action models by maximizing conditional pointwise mutual information between instructions and actions."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"cb93c14ce138b235765f5cadb0d8806e4d2ef8f0c09bbee112fbb4c6887b7190"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"44b4db63-88ca-4edb-a2c5-f47d5236710c"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:24Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"lmecnY+9/gTcx1UuU1xz3+lk1NUS/XibXBEZU8zlYAjNCuAXXoI+wuWXEhtzJEjxYwa1nXYa0neCt8OZQy5+Aw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-10T09:54:04.237942Z"},"content_sha256":"7ef8a7483e55fc46f412385c8ef2c3786f91fb23126b9184fba503f373445750","schema_version":"1.0","event_id":"sha256:7ef8a7483e55fc46f412385c8ef2c3786f91fb23126b9184fba503f373445750"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/CW7IWZ46YM7VWQNJOBNGPBBPUG/bundle.json","state_url":"https://pith.science/pith/CW7IWZ46YM7VWQNJOBNGPBBPUG/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/CW7IWZ46YM7VWQNJOBNGPBBPUG/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-10T09:54:04Z","links":{"resolver":"https://pith.science/pith/CW7IWZ46YM7VWQNJOBNGPBBPUG","bundle":"https://pith.science/pith/CW7IWZ46YM7VWQNJOBNGPBBPUG/bundle.json","state":"https://pith.science/pith/CW7IWZ46YM7VWQNJOBNGPBBPUG/state.json","well_known_bundle":"https://pith.science/.well-known/pith/CW7IWZ46YM7VWQNJOBNGPBBPUG/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:CW7IWZ46YM7VWQNJOBNGPBBPUG","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"5d8f3b3c3e63e88ea90b00a8b588f3289b724b72ca3ba6584ba3dfe04fad01cd","cross_cats_sorted":["cs.CL","cs.CV","cs.RO"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-01-21T17:15:22Z","title_canon_sha256":"a2f940e430f04da099d006def40320a211a8eaabe42cdfdf8b048208a6a7d653"},"schema_version":"1.0","source":{"id":"2601.15197","kind":"arxiv","version":6}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2601.15197","created_at":"2026-05-18T03:09:24Z"},{"alias_kind":"arxiv_version","alias_value":"2601.15197v6","created_at":"2026-05-18T03:09:24Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.15197","created_at":"2026-05-18T03:09:24Z"},{"alias_kind":"pith_short_12","alias_value":"CW7IWZ46YM7V","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"CW7IWZ46YM7VWQNJ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"CW7IWZ46","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:7ef8a7483e55fc46f412385c8ef2c3786f91fb23126b9184fba503f373445750","target":"graph","created_at":"2026-05-18T03:09:24Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Without requiring new data, LangForce significantly improves generalization... including an 11.3% improvement on the challenging OOD SimplerEnv benchmark."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the identified information collapse is the dominant cause of poor OOD performance and that maximizing conditional PMI via the dual-branch architecture will reliably correct it without introducing new failure modes or requiring careful hyperparameter tuning."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"LangForce prevents vision-language-action models from ignoring language by maximizing conditional pointwise mutual information via Bayesian decomposition with latent action queries."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"LangForce restores language grounding in vision-language-action models by maximizing conditional pointwise mutual information between instructions and actions."}],"snapshot_sha256":"9355ce0d1517b1d25726560d9f452f287cbbeb6630412fe28969e01a72e53486"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"cb93c14ce138b235765f5cadb0d8806e4d2ef8f0c09bbee112fbb4c6887b7190"},"paper":{"abstract_excerpt":"Vision-Language-Action (VLA) models have shown promise in robot manipulation but often struggle to generalize to new instructions or complex multi-task scenarios. We identify a critical pathology in current training paradigms where goal-driven data collection creates a dataset bias. In such datasets, language instructions are highly predictable from visual observations alone, causing the conditional mutual information between instructions and actions to vanish, a phenomenon we term Information Collapse. Consequently, models degenerate into vision-only policies that ignore language constraints ","authors_text":"Bin Yu, Changti Wu, Cong Huang, Kai Chen, Laurence T. Yang, Shijie Lian, Xiaopeng Lin, Yuzhuo Miao, Zhaolong Shen","cross_cats":["cs.CL","cs.CV","cs.RO"],"headline":"LangForce restores language grounding in vision-language-action models by maximizing conditional pointwise mutual information between instructions and actions.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-01-21T17:15:22Z","title":"LangForce: Bayesian Decomposition of Vision Language Action Models via Latent Action Queries"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.15197","kind":"arxiv","version":6},"verdict":{"created_at":"2026-05-16T11:56:02.585654Z","id":"44b4db63-88ca-4edb-a2c5-f47d5236710c","model_set":{"reader":"grok-4.3"},"one_line_summary":"LangForce prevents vision-language-action models from ignoring language by maximizing conditional pointwise mutual information via Bayesian decomposition with latent action queries.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"LangForce restores language grounding in vision-language-action models by maximizing conditional pointwise mutual information between instructions and actions.","strongest_claim":"Without requiring new data, LangForce significantly improves generalization... including an 11.3% improvement on the challenging OOD SimplerEnv benchmark.","weakest_assumption":"That the identified information collapse is the dominant cause of poor OOD performance and that maximizing conditional PMI via the dual-branch architecture will reliably correct it without introducing new failure modes or requiring careful hyperparameter tuning."}},"verdict_id":"44b4db63-88ca-4edb-a2c5-f47d5236710c"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:478e3c2912f09d6f290d35383dbfbb64baeba022920266f2ebfcc4d4ee9acc35","target":"record","created_at":"2026-05-18T03:09:24Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"5d8f3b3c3e63e88ea90b00a8b588f3289b724b72ca3ba6584ba3dfe04fad01cd","cross_cats_sorted":["cs.CL","cs.CV","cs.RO"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-01-21T17:15:22Z","title_canon_sha256":"a2f940e430f04da099d006def40320a211a8eaabe42cdfdf8b048208a6a7d653"},"schema_version":"1.0","source":{"id":"2601.15197","kind":"arxiv","version":6}},"canonical_sha256":"15be8b679ec33f5b41a9705a67842fa1981380046c4fd4fb2995a6d2eef2fecf","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"15be8b679ec33f5b41a9705a67842fa1981380046c4fd4fb2995a6d2eef2fecf","first_computed_at":"2026-05-18T03:09:24.337228Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:09:24.337228Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"1vTHYmnF5OOzfPdgqTe2nFLSyahdghFqHuSMtcixZCuLaMeb2sywIJ4zo1JDSYVWRhD8rhZvEqCmrfGUfvJDBw==","signature_status":"signed_v1","signed_at":"2026-05-18T03:09:24.337849Z","signed_message":"canonical_sha256_bytes"},"source_id":"2601.15197","source_kind":"arxiv","source_version":6}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:478e3c2912f09d6f290d35383dbfbb64baeba022920266f2ebfcc4d4ee9acc35","sha256:7ef8a7483e55fc46f412385c8ef2c3786f91fb23126b9184fba503f373445750"],"state_sha256":"ed60f6c789de8d6726d229332115392c279092d3a7d0ea154b82fd7b760815af"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"s1MFRgw1hTIF0MWGSWZdGO/1ZXkLioB01/uPQVkiyfnKQA5EcM7LjtDw3E7EzYdgNfHL7NorFw6EWkEcnIL4BQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-10T09:54:04.242360Z","bundle_sha256":"7b7bd84cdcc43b164c1a61e902d1c7347a269021ea913db48e3e12cccb29fc0a"}}