{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:C7MNKQ5AUIFL2PD4F47VHNALFG","short_pith_number":"pith:C7MNKQ5A","schema_version":"1.0","canonical_sha256":"17d8d543a0a20abd3c7c2f3f53b40b2987908ef04d477c9f3e684b81c89a811e","source":{"kind":"arxiv","id":"2306.09479","version":2},"attestation_state":"computed","paper":{"title":"Inverse Scaling: When Bigger Isn't Better","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CY"],"primary_cat":"cs.CL","authors_text":"Aaron Kirtland, Aaron Mueller, Alexander Lyzhov, Alexis Ross, Alicia Parrish, Alisa Liu, Ameya Prabhu, Andrew Gritsevskiy, Daniel Wurgaft, Derik Kauffman, Ethan Perez, Euan McLean, Gabriel Recchia, Ian R. McKenzie, Jiacheng Liu, Joe Cavanagh, Max Weiss, Michael Pieler, Najoung Kim, Samuel R. Bowman, Sicong Huang, The Floating Droid, Tomasz Korbak, Tom Tseng, Xudong Shen, Yuhui Zhang, Zhengping Zhou","submitted_at":"2023-06-15T20:11:23Z","abstract_excerpt":"Work on scaling laws has found that large language models (LMs) show predictable improvements to overall loss with increased scale (model size, training data, and compute). Here, we present evidence for the claim that LMs may show inverse scaling, or worse task performance with increased scale, e.g., due to flaws in the training objective and data. We present empirical evidence of inverse scaling on 11 datasets collected by running a public contest, the Inverse Scaling Prize, with a substantial prize pool. Through analysis of the datasets, along with other examples found in the literature, we "},"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":"2306.09479","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-06-15T20:11:23Z","cross_cats_sorted":["cs.AI","cs.CY"],"title_canon_sha256":"06bb9528c26592512741923b2adc864fd2eab591c9fbc8af82888e71b8fbf50a","abstract_canon_sha256":"57d70b7012b70b2fed0a07509099793360c89915d83ce9a8cd4f5d14da53bf93"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:17:57.747963Z","signature_b64":"s7G+6CzIL7DL1BKi4yxKVtWHBTzkrBpkNu3YEH4C5x0DSexUjbWFV+qFKNbLLtrygESiPVm5G28YzhmJSoypBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"17d8d543a0a20abd3c7c2f3f53b40b2987908ef04d477c9f3e684b81c89a811e","last_reissued_at":"2026-07-05T08:17:57.747474Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:17:57.747474Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Inverse Scaling: When Bigger Isn't Better","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CY"],"primary_cat":"cs.CL","authors_text":"Aaron Kirtland, Aaron Mueller, Alexander Lyzhov, Alexis Ross, Alicia Parrish, Alisa Liu, Ameya Prabhu, Andrew Gritsevskiy, Daniel Wurgaft, Derik Kauffman, Ethan Perez, Euan McLean, Gabriel Recchia, Ian R. McKenzie, Jiacheng Liu, Joe Cavanagh, Max Weiss, Michael Pieler, Najoung Kim, Samuel R. Bowman, Sicong Huang, The Floating Droid, Tomasz Korbak, Tom Tseng, Xudong Shen, Yuhui Zhang, Zhengping Zhou","submitted_at":"2023-06-15T20:11:23Z","abstract_excerpt":"Work on scaling laws has found that large language models (LMs) show predictable improvements to overall loss with increased scale (model size, training data, and compute). Here, we present evidence for the claim that LMs may show inverse scaling, or worse task performance with increased scale, e.g., due to flaws in the training objective and data. We present empirical evidence of inverse scaling on 11 datasets collected by running a public contest, the Inverse Scaling Prize, with a substantial prize pool. Through analysis of the datasets, along with other examples found in the literature, we "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.09479","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/2306.09479/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":"2306.09479","created_at":"2026-07-05T08:17:57.747527+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.09479v2","created_at":"2026-07-05T08:17:57.747527+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.09479","created_at":"2026-07-05T08:17:57.747527+00:00"},{"alias_kind":"pith_short_12","alias_value":"C7MNKQ5AUIFL","created_at":"2026-07-05T08:17:57.747527+00:00"},{"alias_kind":"pith_short_16","alias_value":"C7MNKQ5AUIFL2PD4","created_at":"2026-07-05T08:17:57.747527+00:00"},{"alias_kind":"pith_short_8","alias_value":"C7MNKQ5A","created_at":"2026-07-05T08:17:57.747527+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2606.05183","citing_title":"The Granularity Gap: A Multi-Dimensional Longitudinal Audit of Sycophancy in Gemini Models","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2606.25476","citing_title":"A Red Teaming Framework for Large Language Models: A Case Study on Faithfulness Evaluation","ref_index":77,"is_internal_anchor":false},{"citing_arxiv_id":"2606.13262","citing_title":"From Verdict to Process: Agentic Reinforcement Learning for Multi-Stage Fact Verification","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04421","citing_title":"Trivium: Temporal Regret as a First-Class Objective for Causal-Memory Controllers","ref_index":76,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05497","citing_title":"LEVANTE-bench: Multi-Scale Comparison of VLMs to Children Using Cognitive Tasks (or, \"Is Your VLM Smarter Than a 5th Grader?\")","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01923","citing_title":"Resonant Context Anchoring: Decoupling Attention Routing and Signal Gain at Inference Time","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29424","citing_title":"EntroRouter: Learning Efficient Model Routing via Entropy Regulation","ref_index":78,"is_internal_anchor":false},{"citing_arxiv_id":"2605.31031","citing_title":"GraphARC: A Comprehensive Benchmark for Graph-Based Abstract Reasoning","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04010","citing_title":"The Variance Brain Foundation Models Forgot: Third-Order Statistics Predict Cognition Where Billion-Parameter Models Fail","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2503.11926","citing_title":"Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation","ref_index":77,"is_internal_anchor":false},{"citing_arxiv_id":"2512.14917","citing_title":"Evaluating Code Reasoning Abilities of Large Language Models Under Real-World Settings","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13825","citing_title":"History Anchors: How Prior Behavior Steers LLM Decisions Toward Unsafe Actions","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2406.13352","citing_title":"AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09678","citing_title":"Absurd World: A Simple Yet Powerful Method to Absurdify the Real-world for Probing LLM Reasoning Capabilities","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02285","citing_title":"Complexity Horizons of Compressed Models in Analog Circuit Analysis","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/C7MNKQ5AUIFL2PD4F47VHNALFG","json":"https://pith.science/pith/C7MNKQ5AUIFL2PD4F47VHNALFG.json","graph_json":"https://pith.science/api/pith-number/C7MNKQ5AUIFL2PD4F47VHNALFG/graph.json","events_json":"https://pith.science/api/pith-number/C7MNKQ5AUIFL2PD4F47VHNALFG/events.json","paper":"https://pith.science/paper/C7MNKQ5A"},"agent_actions":{"view_html":"https://pith.science/pith/C7MNKQ5AUIFL2PD4F47VHNALFG","download_json":"https://pith.science/pith/C7MNKQ5AUIFL2PD4F47VHNALFG.json","view_paper":"https://pith.science/paper/C7MNKQ5A","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.09479&json=true","fetch_graph":"https://pith.science/api/pith-number/C7MNKQ5AUIFL2PD4F47VHNALFG/graph.json","fetch_events":"https://pith.science/api/pith-number/C7MNKQ5AUIFL2PD4F47VHNALFG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/C7MNKQ5AUIFL2PD4F47VHNALFG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/C7MNKQ5AUIFL2PD4F47VHNALFG/action/storage_attestation","attest_author":"https://pith.science/pith/C7MNKQ5AUIFL2PD4F47VHNALFG/action/author_attestation","sign_citation":"https://pith.science/pith/C7MNKQ5AUIFL2PD4F47VHNALFG/action/citation_signature","submit_replication":"https://pith.science/pith/C7MNKQ5AUIFL2PD4F47VHNALFG/action/replication_record"}},"created_at":"2026-07-05T08:17:57.747527+00:00","updated_at":"2026-07-05T08:17:57.747527+00:00"}