{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XHOIC3SDDBHXSMJZ23BGRFO7JB","short_pith_number":"pith:XHOIC3SD","schema_version":"1.0","canonical_sha256":"b9dc816e43184f793139d6c26895df485054d4b2bc2891a819e4899144246602","source":{"kind":"arxiv","id":"2505.00047","version":2},"attestation_state":"computed","paper":{"title":"Base Models Beat Aligned Models at Randomness and Creativity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Christopher Potts, Peter West","submitted_at":"2025-04-30T03:41:55Z","abstract_excerpt":"Alignment has quickly become a default ingredient in LLM development, with techniques such as reinforcement learning from human feedback making models act safely, follow instructions, and perform ever-better on complex tasks. While these techniques are certainly useful, we propose that they should not be universally applied and demonstrate a range of tasks on which base language models consistently outperform their popular aligned forms. Particularly, we study tasks that require unpredictable outputs, such as random number generation, mixed strategy games (rock-paper-scissors and hide-and-seek"},"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":"2505.00047","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-04-30T03:41:55Z","cross_cats_sorted":[],"title_canon_sha256":"a58727ffa7c584104950f10f15229e06128d4004c8ff0bfd3941f0f729c88b66","abstract_canon_sha256":"01822d857da1a105b1deaae8768a81729049f8647b1d32ebc4bea824cc212fc9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:11:17.460670Z","signature_b64":"KLxfTIH9O8V5lAON3oTFA5NQ+6hbYI9arykVRtZIGH7dqOLXrVPJ9X/Pvkx3XfweXO3AbjAx/Fn0HN5o3CoUBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b9dc816e43184f793139d6c26895df485054d4b2bc2891a819e4899144246602","last_reissued_at":"2026-07-05T12:11:17.460103Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:11:17.460103Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Base Models Beat Aligned Models at Randomness and Creativity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Christopher Potts, Peter West","submitted_at":"2025-04-30T03:41:55Z","abstract_excerpt":"Alignment has quickly become a default ingredient in LLM development, with techniques such as reinforcement learning from human feedback making models act safely, follow instructions, and perform ever-better on complex tasks. While these techniques are certainly useful, we propose that they should not be universally applied and demonstrate a range of tasks on which base language models consistently outperform their popular aligned forms. Particularly, we study tasks that require unpredictable outputs, such as random number generation, mixed strategy games (rock-paper-scissors and hide-and-seek"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.00047","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/2505.00047/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":"2505.00047","created_at":"2026-07-05T12:11:17.460163+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.00047v2","created_at":"2026-07-05T12:11:17.460163+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.00047","created_at":"2026-07-05T12:11:17.460163+00:00"},{"alias_kind":"pith_short_12","alias_value":"XHOIC3SDDBHX","created_at":"2026-07-05T12:11:17.460163+00:00"},{"alias_kind":"pith_short_16","alias_value":"XHOIC3SDDBHXSMJZ","created_at":"2026-07-05T12:11:17.460163+00:00"},{"alias_kind":"pith_short_8","alias_value":"XHOIC3SD","created_at":"2026-07-05T12:11:17.460163+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11456","citing_title":"AI Coding Agents in Social Science: Methodologically Diverse, Empirically Consistent, Interpretively Vulnerable","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29933","citing_title":"Towards Physical Intuitions for Alignment Dynamics: A Case Study With Randomness Crystallization","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28664","citing_title":"Activation Steering for Synthetic Data Generation: The Role of Diversity in Downstream Safety Detection","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30844","citing_title":"Fine-Tuning Improves Information Conveyance in Language Models","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00875","citing_title":"IDEAFix: Evaluation Framework for Creative Defixation Prompting in LLMs","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2601.06116","citing_title":"The Homogenization Problem in LLMs: Towards Meaningful Diversity in AI Safety","ref_index":136,"is_internal_anchor":false},{"citing_arxiv_id":"2601.06116","citing_title":"The Homogenization Problem in LLMs: Towards Meaningful Diversity in AI Safety","ref_index":136,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11258","citing_title":"Unlocking LLM Creativity in Science through Analogical Reasoning","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09995","citing_title":"Annotations Mitigate Post-Training Mode Collapse","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XHOIC3SDDBHXSMJZ23BGRFO7JB","json":"https://pith.science/pith/XHOIC3SDDBHXSMJZ23BGRFO7JB.json","graph_json":"https://pith.science/api/pith-number/XHOIC3SDDBHXSMJZ23BGRFO7JB/graph.json","events_json":"https://pith.science/api/pith-number/XHOIC3SDDBHXSMJZ23BGRFO7JB/events.json","paper":"https://pith.science/paper/XHOIC3SD"},"agent_actions":{"view_html":"https://pith.science/pith/XHOIC3SDDBHXSMJZ23BGRFO7JB","download_json":"https://pith.science/pith/XHOIC3SDDBHXSMJZ23BGRFO7JB.json","view_paper":"https://pith.science/paper/XHOIC3SD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.00047&json=true","fetch_graph":"https://pith.science/api/pith-number/XHOIC3SDDBHXSMJZ23BGRFO7JB/graph.json","fetch_events":"https://pith.science/api/pith-number/XHOIC3SDDBHXSMJZ23BGRFO7JB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XHOIC3SDDBHXSMJZ23BGRFO7JB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XHOIC3SDDBHXSMJZ23BGRFO7JB/action/storage_attestation","attest_author":"https://pith.science/pith/XHOIC3SDDBHXSMJZ23BGRFO7JB/action/author_attestation","sign_citation":"https://pith.science/pith/XHOIC3SDDBHXSMJZ23BGRFO7JB/action/citation_signature","submit_replication":"https://pith.science/pith/XHOIC3SDDBHXSMJZ23BGRFO7JB/action/replication_record"}},"created_at":"2026-07-05T12:11:17.460163+00:00","updated_at":"2026-07-05T12:11:17.460163+00:00"}