{"paper":{"title":"HumorGen: Cognitive Synergy for Humor Generation in Large Language Models via Persona-Based Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Cognitive personas synthesizing humor data let a 7B model match or beat much larger LLMs at comedy.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Edward Ajayi, Prasenjit Mitra","submitted_at":"2026-03-19T13:12:53Z","abstract_excerpt":"Humor generation poses a significant challenge for Large Language Models (LLMs), because their standard training objective (next-token prediction) inherently conflicts with the surprise and incongruity required for comedy. To bridge this gap, we introduce the Cognitive Synergy Framework, a methodology for generating highquality humor data inspired by psychological theories of humor. Utilizing a Mixtureof-Thought (MoT) approach, we deploy six cognitive personas (e.g., The Absurdist, The Cynic) to synthesize diverse comedic perspectives for a given prompt. This framework produces a theory-ground"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"our 7B model significantly outperforms larger instruction-tuned baselines and achieves performance competitive with state-of-the-art proprietary models. We find that cognitive-driven data curation is far more critical than alignment algorithms or model scale for humor generation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The humor data synthesized using the six cognitive personas through the Mixture-of-Thought approach provides a high-quality, diverse training signal that effectively improves the model's humor generation capabilities beyond what standard methods achieve.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A 7B LLM fine-tuned on humor data generated via six cognitive personas and Mixture-of-Thought outperforms larger instruction-tuned baselines and competes with proprietary models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Cognitive personas synthesizing humor data let a 7B model match or beat much larger LLMs at comedy.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f44079b6296e81c9b17ad67431e10819f748eb4d75dee7f3f6fb3bcc5bf04a7a"},"source":{"id":"2604.09629","kind":"arxiv","version":2},"verdict":{"id":"3a2418ce-7faa-4f0b-bd9d-4399329aa1b1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T08:38:56.673007Z","strongest_claim":"our 7B model significantly outperforms larger instruction-tuned baselines and achieves performance competitive with state-of-the-art proprietary models. We find that cognitive-driven data curation is far more critical than alignment algorithms or model scale for humor generation.","one_line_summary":"A 7B LLM fine-tuned on humor data generated via six cognitive personas and Mixture-of-Thought outperforms larger instruction-tuned baselines and competes with proprietary models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The humor data synthesized using the six cognitive personas through the Mixture-of-Thought approach provides a high-quality, diverse training signal that effectively improves the model's humor generation capabilities beyond what standard methods achieve.","pith_extraction_headline":"Cognitive personas synthesizing humor data let a 7B model match or beat much larger LLMs at comedy."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.09629/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":1,"snapshot_sha256":"19a1d8a4f38ade4a25b145f7d88b6dbb2a1bb7dc3d663fa3c258fd7bf72dd01a"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}