{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:HFJFFBM2YL5DKB6UR3B7JZNQM3","short_pith_number":"pith:HFJFFBM2","schema_version":"1.0","canonical_sha256":"395252859ac2fa3507d48ec3f4e5b066d21d0e6b105f1dc760ca42664372c80b","source":{"kind":"arxiv","id":"2403.15796","version":3},"attestation_state":"computed","paper":{"title":"Understanding Emergent Abilities of Language Models from the Loss Perspective","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Aohan Zeng, Jie Tang, Yuxiao Dong, Zhengxiao Du","submitted_at":"2024-03-23T11:03:31Z","abstract_excerpt":"Recent studies have put into question the belief that emergent abilities in language models are exclusive to large models. This skepticism arises from two observations: 1) smaller models can also exhibit high performance on emergent abilities and 2) there is doubt on the discontinuous metrics used to measure these abilities. In this paper, we propose to study emergent abilities in the lens of pre-training loss, instead of model size or training compute. We demonstrate that the Transformer models with the same pre-training loss, but different model and data sizes, generate the same performance "},"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":"2403.15796","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-03-23T11:03:31Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"3473eb3ffe229d3410fbf90292f9c0ff4d7b882d76a727849038a4bd98ffc380","abstract_canon_sha256":"82b90b1926841b76fb2a450f8821f4459366acd0f56aa51c14d9b26e73b70093"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:01:12.332234Z","signature_b64":"Gyp7aRlinTEvU+afDY0YM8y158jiZByhJq8skEh2Ot0vVtuuHdfEq0O4FF0h0Gps9iB2awvufSBZXBMB9zVkAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"395252859ac2fa3507d48ec3f4e5b066d21d0e6b105f1dc760ca42664372c80b","last_reissued_at":"2026-07-05T10:01:12.331766Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:01:12.331766Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding Emergent Abilities of Language Models from the Loss Perspective","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Aohan Zeng, Jie Tang, Yuxiao Dong, Zhengxiao Du","submitted_at":"2024-03-23T11:03:31Z","abstract_excerpt":"Recent studies have put into question the belief that emergent abilities in language models are exclusive to large models. This skepticism arises from two observations: 1) smaller models can also exhibit high performance on emergent abilities and 2) there is doubt on the discontinuous metrics used to measure these abilities. In this paper, we propose to study emergent abilities in the lens of pre-training loss, instead of model size or training compute. We demonstrate that the Transformer models with the same pre-training loss, but different model and data sizes, generate the same performance "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.15796","kind":"arxiv","version":3},"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/2403.15796/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":"2403.15796","created_at":"2026-07-05T10:01:12.331820+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.15796v3","created_at":"2026-07-05T10:01:12.331820+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.15796","created_at":"2026-07-05T10:01:12.331820+00:00"},{"alias_kind":"pith_short_12","alias_value":"HFJFFBM2YL5D","created_at":"2026-07-05T10:01:12.331820+00:00"},{"alias_kind":"pith_short_16","alias_value":"HFJFFBM2YL5DKB6U","created_at":"2026-07-05T10:01:12.331820+00:00"},{"alias_kind":"pith_short_8","alias_value":"HFJFFBM2","created_at":"2026-07-05T10:01:12.331820+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22873","citing_title":"SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2606.22873","citing_title":"SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30815","citing_title":"When transformers learn \"impossible\" languages, what do they learn?","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17767","citing_title":"Feature Learning in Linear-Width Two-Layer Networks: Two vs. One Step of Gradient Descent","ref_index":213,"is_internal_anchor":false},{"citing_arxiv_id":"2502.12120","citing_title":"LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19195","citing_title":"The Thermodynamic Costs of Simple Linear Regression","ref_index":69,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17767","citing_title":"Feature Learning in Linear-Width Two-Layer Networks: Two vs. One Step of Gradient Descent","ref_index":213,"is_internal_anchor":false},{"citing_arxiv_id":"2404.06395","citing_title":"MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2502.02737","citing_title":"SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model","ref_index":168,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HFJFFBM2YL5DKB6UR3B7JZNQM3","json":"https://pith.science/pith/HFJFFBM2YL5DKB6UR3B7JZNQM3.json","graph_json":"https://pith.science/api/pith-number/HFJFFBM2YL5DKB6UR3B7JZNQM3/graph.json","events_json":"https://pith.science/api/pith-number/HFJFFBM2YL5DKB6UR3B7JZNQM3/events.json","paper":"https://pith.science/paper/HFJFFBM2"},"agent_actions":{"view_html":"https://pith.science/pith/HFJFFBM2YL5DKB6UR3B7JZNQM3","download_json":"https://pith.science/pith/HFJFFBM2YL5DKB6UR3B7JZNQM3.json","view_paper":"https://pith.science/paper/HFJFFBM2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.15796&json=true","fetch_graph":"https://pith.science/api/pith-number/HFJFFBM2YL5DKB6UR3B7JZNQM3/graph.json","fetch_events":"https://pith.science/api/pith-number/HFJFFBM2YL5DKB6UR3B7JZNQM3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HFJFFBM2YL5DKB6UR3B7JZNQM3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HFJFFBM2YL5DKB6UR3B7JZNQM3/action/storage_attestation","attest_author":"https://pith.science/pith/HFJFFBM2YL5DKB6UR3B7JZNQM3/action/author_attestation","sign_citation":"https://pith.science/pith/HFJFFBM2YL5DKB6UR3B7JZNQM3/action/citation_signature","submit_replication":"https://pith.science/pith/HFJFFBM2YL5DKB6UR3B7JZNQM3/action/replication_record"}},"created_at":"2026-07-05T10:01:12.331820+00:00","updated_at":"2026-07-05T10:01:12.331820+00:00"}