{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:2CNWIVJV6RBUEBUEUMSMVLTIW4","short_pith_number":"pith:2CNWIVJV","schema_version":"1.0","canonical_sha256":"d09b645535f443420684a324caae68b7025e3b6a7caf65fe49a77878834805c3","source":{"kind":"arxiv","id":"2104.06129","version":2},"attestation_state":"computed","paper":{"title":"What's in your Head? Emergent Behaviour in Multi-Task Transformer Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Aviv Ben-Arie, Jonathan Berant, Mor Geva, Uri Katz","submitted_at":"2021-04-13T12:04:30Z","abstract_excerpt":"The primary paradigm for multi-task training in natural language processing is to represent the input with a shared pre-trained language model, and add a small, thin network (head) per task. Given an input, a target head is the head that is selected for outputting the final prediction. In this work, we examine the behaviour of non-target heads, that is, the output of heads when given input that belongs to a different task than the one they were trained for. We find that non-target heads exhibit emergent behaviour, which may either explain the target task, or generalize beyond their original ta"},"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":"2104.06129","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-04-13T12:04:30Z","cross_cats_sorted":[],"title_canon_sha256":"b52fa12ea63774ef743c1cf4a7a0e05c5383b08f1a8d22d5349a54bd1f379462","abstract_canon_sha256":"efaf596fcc97f2b328b51bee91a451280597dc3b6dcde96f84418cbbd42b60b9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:11:28.914540Z","signature_b64":"ZRujS4g6jMyoV82KBPc/96W8Alw2Dv0UeNKtDCGy3vdCm7Qt+1+Cseimx0/6tf7aRy9/22L2agzdDnKWMFLrAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d09b645535f443420684a324caae68b7025e3b6a7caf65fe49a77878834805c3","last_reissued_at":"2026-07-05T03:11:28.914091Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:11:28.914091Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"What's in your Head? Emergent Behaviour in Multi-Task Transformer Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Aviv Ben-Arie, Jonathan Berant, Mor Geva, Uri Katz","submitted_at":"2021-04-13T12:04:30Z","abstract_excerpt":"The primary paradigm for multi-task training in natural language processing is to represent the input with a shared pre-trained language model, and add a small, thin network (head) per task. Given an input, a target head is the head that is selected for outputting the final prediction. In this work, we examine the behaviour of non-target heads, that is, the output of heads when given input that belongs to a different task than the one they were trained for. We find that non-target heads exhibit emergent behaviour, which may either explain the target task, or generalize beyond their original ta"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2104.06129","kind":"arxiv","version":2},"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/2104.06129/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":"2104.06129","created_at":"2026-07-05T03:11:28.914155+00:00"},{"alias_kind":"arxiv_version","alias_value":"2104.06129v2","created_at":"2026-07-05T03:11:28.914155+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2104.06129","created_at":"2026-07-05T03:11:28.914155+00:00"},{"alias_kind":"pith_short_12","alias_value":"2CNWIVJV6RBU","created_at":"2026-07-05T03:11:28.914155+00:00"},{"alias_kind":"pith_short_16","alias_value":"2CNWIVJV6RBUEBUE","created_at":"2026-07-05T03:11:28.914155+00:00"},{"alias_kind":"pith_short_8","alias_value":"2CNWIVJV","created_at":"2026-07-05T03:11:28.914155+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/2CNWIVJV6RBUEBUEUMSMVLTIW4","json":"https://pith.science/pith/2CNWIVJV6RBUEBUEUMSMVLTIW4.json","graph_json":"https://pith.science/api/pith-number/2CNWIVJV6RBUEBUEUMSMVLTIW4/graph.json","events_json":"https://pith.science/api/pith-number/2CNWIVJV6RBUEBUEUMSMVLTIW4/events.json","paper":"https://pith.science/paper/2CNWIVJV"},"agent_actions":{"view_html":"https://pith.science/pith/2CNWIVJV6RBUEBUEUMSMVLTIW4","download_json":"https://pith.science/pith/2CNWIVJV6RBUEBUEUMSMVLTIW4.json","view_paper":"https://pith.science/paper/2CNWIVJV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2104.06129&json=true","fetch_graph":"https://pith.science/api/pith-number/2CNWIVJV6RBUEBUEUMSMVLTIW4/graph.json","fetch_events":"https://pith.science/api/pith-number/2CNWIVJV6RBUEBUEUMSMVLTIW4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2CNWIVJV6RBUEBUEUMSMVLTIW4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2CNWIVJV6RBUEBUEUMSMVLTIW4/action/storage_attestation","attest_author":"https://pith.science/pith/2CNWIVJV6RBUEBUEUMSMVLTIW4/action/author_attestation","sign_citation":"https://pith.science/pith/2CNWIVJV6RBUEBUEUMSMVLTIW4/action/citation_signature","submit_replication":"https://pith.science/pith/2CNWIVJV6RBUEBUEUMSMVLTIW4/action/replication_record"}},"created_at":"2026-07-05T03:11:28.914155+00:00","updated_at":"2026-07-05T03:11:28.914155+00:00"}