Nicholas Lourie

NYU researcher, lead author of "Scaling Laws Are Unreliable for Downstream Tasks: A Reality Check" (Findings of EMNLP 2025), which classifies 46 downstream LLM-evaluation tasks into six scaling-behavior categories and finds clean linear scaling in only 39% of them.

Matters to this vault beyond that co-authorship: a year later he is also a co-author (fourth of five) on Hu, Pan, Jhaveri, Lourie & Cho's "Neural Neural Scaling Laws" (2026), which cites his own 2025 paper as settled background — "no single parametric family can capture the full spectrum of scaling behaviors [Lourie et al., 2025]" — without extending its taxonomy. He is the direct continuity point between the vault's two Hu-et-al.-adjacent scaling-law notes, and the clearest case in this cluster of a research group citing its own prior answer rather than building on it.

References

• claim-lourie-2025-six-category-downstream-scaling-taxonomy
• claim-neural-neural-scaling-laws-2026-averaging-obscures-per-task-scaling

Updates

• 2026-10-24: "Neural Neural Scaling Laws" has been accepted as a NeurIPS 2026 main-track poster and listed as an ICML 2026 workshop poster, both under the identical five-author byline (Hu, Pan, Jhaveri, Lourie, Cho) — refereed-venue acceptance, still the same author group, not independent outside corroboration of the paper's claim. (claim-hu-2026-neurips-icml-2026-same-byline-not-independent)