{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:TPZY2QZUHI4V7XACWNLURLHB5M","short_pith_number":"pith:TPZY2QZU","schema_version":"1.0","canonical_sha256":"9bf38d43343a395fdc02b35748ace1eb1c0689539bcaab69d0718ae62931faac","source":{"kind":"arxiv","id":"2211.02011","version":5},"attestation_state":"computed","paper":{"title":"Inverse scaling can become U-shaped","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jason Wei, Najoung Kim, Quoc V. Le, Yi Tay","submitted_at":"2022-11-03T17:26:44Z","abstract_excerpt":"Scaling up language models has been empirically shown to improve performance on a wide range of downstream tasks. However, if we were to observe worse performance as a function of scale (\"inverse scaling\") on certain tasks, this would indicate that scaling can also encourage behaviors that are misaligned with human preferences. The Inverse Scaling Prize (McKenzie et al. 2022) identified eleven such inverse scaling tasks, evaluated on models of up to 280B parameters and up to 500 zettaFLOPs of training compute. This paper takes a closer look at these inverse scaling tasks. We evaluate models of"},"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":"2211.02011","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-11-03T17:26:44Z","cross_cats_sorted":[],"title_canon_sha256":"3c6158c77caf784193b08789ce13468d4676cc02a767be996c65800b5084d53e","abstract_canon_sha256":"e2c39bd79a26187a2ab790884f2854a521d6ef5b5cdd4c0f2e8307120bfc89ef"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:13:17.132889Z","signature_b64":"/BKhIqUOP6dUqTG65lu6jhThpw3FD+1dlWSU8AgT0+W1HLW1wkW+ZhcLD82kLnyWu9s6y/3qgc9+5ht8LP3MDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9bf38d43343a395fdc02b35748ace1eb1c0689539bcaab69d0718ae62931faac","last_reissued_at":"2026-07-05T06:13:17.132430Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:13:17.132430Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Inverse scaling can become U-shaped","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jason Wei, Najoung Kim, Quoc V. Le, Yi Tay","submitted_at":"2022-11-03T17:26:44Z","abstract_excerpt":"Scaling up language models has been empirically shown to improve performance on a wide range of downstream tasks. However, if we were to observe worse performance as a function of scale (\"inverse scaling\") on certain tasks, this would indicate that scaling can also encourage behaviors that are misaligned with human preferences. The Inverse Scaling Prize (McKenzie et al. 2022) identified eleven such inverse scaling tasks, evaluated on models of up to 280B parameters and up to 500 zettaFLOPs of training compute. This paper takes a closer look at these inverse scaling tasks. We evaluate models of"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2211.02011","kind":"arxiv","version":5},"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/2211.02011/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":"2211.02011","created_at":"2026-07-05T06:13:17.132487+00:00"},{"alias_kind":"arxiv_version","alias_value":"2211.02011v5","created_at":"2026-07-05T06:13:17.132487+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2211.02011","created_at":"2026-07-05T06:13:17.132487+00:00"},{"alias_kind":"pith_short_12","alias_value":"TPZY2QZUHI4V","created_at":"2026-07-05T06:13:17.132487+00:00"},{"alias_kind":"pith_short_16","alias_value":"TPZY2QZUHI4V7XAC","created_at":"2026-07-05T06:13:17.132487+00:00"},{"alias_kind":"pith_short_8","alias_value":"TPZY2QZU","created_at":"2026-07-05T06:13:17.132487+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08403","citing_title":"Game Theory Driven Multi-Agent Framework Mitigates Language Model Hallucination","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2606.07568","citing_title":"A Systematic Study of Behavioral Cloning for Scientific Data Annotation","ref_index":138,"is_internal_anchor":false},{"citing_arxiv_id":"2601.19831","citing_title":"Neural Neural Scaling Laws","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TPZY2QZUHI4V7XACWNLURLHB5M","json":"https://pith.science/pith/TPZY2QZUHI4V7XACWNLURLHB5M.json","graph_json":"https://pith.science/api/pith-number/TPZY2QZUHI4V7XACWNLURLHB5M/graph.json","events_json":"https://pith.science/api/pith-number/TPZY2QZUHI4V7XACWNLURLHB5M/events.json","paper":"https://pith.science/paper/TPZY2QZU"},"agent_actions":{"view_html":"https://pith.science/pith/TPZY2QZUHI4V7XACWNLURLHB5M","download_json":"https://pith.science/pith/TPZY2QZUHI4V7XACWNLURLHB5M.json","view_paper":"https://pith.science/paper/TPZY2QZU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2211.02011&json=true","fetch_graph":"https://pith.science/api/pith-number/TPZY2QZUHI4V7XACWNLURLHB5M/graph.json","fetch_events":"https://pith.science/api/pith-number/TPZY2QZUHI4V7XACWNLURLHB5M/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TPZY2QZUHI4V7XACWNLURLHB5M/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TPZY2QZUHI4V7XACWNLURLHB5M/action/storage_attestation","attest_author":"https://pith.science/pith/TPZY2QZUHI4V7XACWNLURLHB5M/action/author_attestation","sign_citation":"https://pith.science/pith/TPZY2QZUHI4V7XACWNLURLHB5M/action/citation_signature","submit_replication":"https://pith.science/pith/TPZY2QZUHI4V7XACWNLURLHB5M/action/replication_record"}},"created_at":"2026-07-05T06:13:17.132487+00:00","updated_at":"2026-07-05T06:13:17.132487+00:00"}