{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:NKCOACMXCO4FEWWHIS4BY6UOYR","short_pith_number":"pith:NKCOACMX","schema_version":"1.0","canonical_sha256":"6a84e0099713b8525ac744b81c7a8ec479b5f4d778101b075fee11e82c674808","source":{"kind":"arxiv","id":"2404.10859","version":2},"attestation_state":"computed","paper":{"title":"Forcing Diffuse Distributions out of Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Avi Schwarzschild, Daphne Ippolito, Nicholas Carlini, Yiming Zhang, Zico Kolter","submitted_at":"2024-04-16T19:17:23Z","abstract_excerpt":"Despite being trained specifically to follow user instructions, today's instructiontuned language models perform poorly when instructed to produce random outputs. For example, when prompted to pick a number uniformly between one and ten Llama-2-13B-chat disproportionately favors the number five, and when tasked with picking a first name at random, Mistral-7B-Instruct chooses Avery 40 times more often than we would expect based on the U.S. population. When these language models are used for real-world tasks where diversity of outputs is crucial, such as language model assisted dataset construct"},"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":"2404.10859","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-04-16T19:17:23Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"76b3195bc701848b737612a77fb9749d74ec66432c7dee5d1dd52d8da949592d","abstract_canon_sha256":"376efe3c0fee580371528e2e1c507ec16cd01dcdb657396134db74de93b7f031"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:53:21.923518Z","signature_b64":"c8Uuul0qsmztENCpVWu1PNX3BuOttDjWL7OdDwAoX4/OY1d4D03FTr40ew3OWktVo0Z2oKhXsJyls6dXbYcXDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6a84e0099713b8525ac744b81c7a8ec479b5f4d778101b075fee11e82c674808","last_reissued_at":"2026-07-05T08:53:21.923030Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:53:21.923030Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Forcing Diffuse Distributions out of Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Avi Schwarzschild, Daphne Ippolito, Nicholas Carlini, Yiming Zhang, Zico Kolter","submitted_at":"2024-04-16T19:17:23Z","abstract_excerpt":"Despite being trained specifically to follow user instructions, today's instructiontuned language models perform poorly when instructed to produce random outputs. For example, when prompted to pick a number uniformly between one and ten Llama-2-13B-chat disproportionately favors the number five, and when tasked with picking a first name at random, Mistral-7B-Instruct chooses Avery 40 times more often than we would expect based on the U.S. population. When these language models are used for real-world tasks where diversity of outputs is crucial, such as language model assisted dataset construct"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.10859","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/2404.10859/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":"2404.10859","created_at":"2026-07-05T08:53:21.923090+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.10859v2","created_at":"2026-07-05T08:53:21.923090+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.10859","created_at":"2026-07-05T08:53:21.923090+00:00"},{"alias_kind":"pith_short_12","alias_value":"NKCOACMXCO4F","created_at":"2026-07-05T08:53:21.923090+00:00"},{"alias_kind":"pith_short_16","alias_value":"NKCOACMXCO4FEWWH","created_at":"2026-07-05T08:53:21.923090+00:00"},{"alias_kind":"pith_short_8","alias_value":"NKCOACMX","created_at":"2026-07-05T08:53:21.923090+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17508","citing_title":"When the Next Step Is Not One Step: Distribution-Aware Execution Modeling for Concurrent Go Programs","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2506.00079","citing_title":"Who Gets the Kidney? Human-AI Alignment, Indecision, and Moral Values","ref_index":72,"is_internal_anchor":false},{"citing_arxiv_id":"2506.00079","citing_title":"Who Gets the Kidney? Human-AI Alignment, Indecision, and Moral Values","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09995","citing_title":"Annotations Mitigate Post-Training Mode Collapse","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18724","citing_title":"Beyond One Output: Visualizing and Comparing Distributions of Language Model Generations","ref_index":60,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NKCOACMXCO4FEWWHIS4BY6UOYR","json":"https://pith.science/pith/NKCOACMXCO4FEWWHIS4BY6UOYR.json","graph_json":"https://pith.science/api/pith-number/NKCOACMXCO4FEWWHIS4BY6UOYR/graph.json","events_json":"https://pith.science/api/pith-number/NKCOACMXCO4FEWWHIS4BY6UOYR/events.json","paper":"https://pith.science/paper/NKCOACMX"},"agent_actions":{"view_html":"https://pith.science/pith/NKCOACMXCO4FEWWHIS4BY6UOYR","download_json":"https://pith.science/pith/NKCOACMXCO4FEWWHIS4BY6UOYR.json","view_paper":"https://pith.science/paper/NKCOACMX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.10859&json=true","fetch_graph":"https://pith.science/api/pith-number/NKCOACMXCO4FEWWHIS4BY6UOYR/graph.json","fetch_events":"https://pith.science/api/pith-number/NKCOACMXCO4FEWWHIS4BY6UOYR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NKCOACMXCO4FEWWHIS4BY6UOYR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NKCOACMXCO4FEWWHIS4BY6UOYR/action/storage_attestation","attest_author":"https://pith.science/pith/NKCOACMXCO4FEWWHIS4BY6UOYR/action/author_attestation","sign_citation":"https://pith.science/pith/NKCOACMXCO4FEWWHIS4BY6UOYR/action/citation_signature","submit_replication":"https://pith.science/pith/NKCOACMXCO4FEWWHIS4BY6UOYR/action/replication_record"}},"created_at":"2026-07-05T08:53:21.923090+00:00","updated_at":"2026-07-05T08:53:21.923090+00:00"}