{"paper":{"title":"NodeSynth: Socially Aligned Synthetic Data for AI Evaluation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"NodeSynth generates synthetic queries via a fine-tuned taxonomy generator that cause mainstream LLMs to fail at rates up to five times higher than human benchmarks.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Darlene Neal, Erin van Liemt, Jamila Smith-Loud, Kshitij Pancholi, Qazi Mamunur Rashid, Xuan Yang, Yanzhou Pan, Zhengzhe Yang","submitted_at":"2026-05-14T05:06:50Z","abstract_excerpt":"Recent advancements in generative AI facilitate large-scale synthetic data generation for model evaluation. However, without targeted approaches, these datasets often lack the sociotechnical nuance required for sensitive domains. We introduce NodeSynth, an evidence-grounded methodology that generates socially relevant synthetic queries by leveraging a fine-tuned taxonomy generator (TaG) anchored in real-world evidence. Evaluated against four mainstream LLMs (e.g., Claude 4.5 Haiku), NodeSynth elicited failure rates up to five times higher than human-authored benchmarks. Ablation studies confir"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Evaluated against four mainstream LLMs, NodeSynth elicited failure rates up to five times higher than human-authored benchmarks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The synthetic queries produced by the fine-tuned TaG are representative of genuine sociotechnical risks without introducing new biases or artifacts that inflate failure rates.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"NodeSynth generates evidence-anchored synthetic queries that trigger up to five times higher failure rates in mainstream LLMs than human-authored benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"NodeSynth generates synthetic queries via a fine-tuned taxonomy generator that cause mainstream LLMs to fail at rates up to five times higher than human benchmarks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"df512049aca8f432ed0f1018b7418ce36aeec7e48bf7331b6a38d7484e7c5fea"},"source":{"id":"2605.14381","kind":"arxiv","version":1},"verdict":{"id":"518124b9-cf1e-4531-88a2-96abe160579a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:35:10.015065Z","strongest_claim":"Evaluated against four mainstream LLMs, NodeSynth elicited failure rates up to five times higher than human-authored benchmarks.","one_line_summary":"NodeSynth generates evidence-anchored synthetic queries that trigger up to five times higher failure rates in mainstream LLMs than human-authored benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The synthetic queries produced by the fine-tuned TaG are representative of genuine sociotechnical risks without introducing new biases or artifacts that inflate failure rates.","pith_extraction_headline":"NodeSynth generates synthetic queries via a fine-tuned taxonomy generator that cause mainstream LLMs to fail at rates up to five times higher than human benchmarks."},"references":{"count":53,"sample":[{"doi":"","year":2024,"title":"On llms-driven synthetic data generation, curation, and evaluation: A survey","work_id":"d8efa201-fecf-4ca6-90f0-02640ce6452b","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Synthetic data in ai: Challenges, applications, and ethical implications.arXiv preprint arXiv:2401.01629, 2024","work_id":"3a0c03f2-d0f9-44ff-b974-336150c54e9c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing","work_id":"4df6cdce-69a0-402e-ab53-59e5c10b8cae","ref_index":3,"cited_arxiv_id":"2406.08464","is_internal_anchor":true},{"doi":"","year":2023,"title":"Self-instruct: Aligning language models with self-generated instruc- tions","work_id":"40d20eeb-4469-4548-b78d-d83166af6146","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Examining the expanding role of synthetic data throughout the ai development pipeline","work_id":"61800835-745c-4c41-b94c-42dda3b576ec","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":53,"snapshot_sha256":"080025d427f0b1bddbba18deb351013e2cae84fbd3aa4900a0534a031d69a915","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"12ac72d2387adf47e73ac9813713fe539819fcecb01b2bccb590a11e487d1545"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}