# NeurIPS and EMNLP 2025 and 2026 on one embedding map: how much do they overlap? > Source: > Published: 2026-10-29 16:34:04+00:00 I embedded the title and abstract of all 23,984 NeurIPS and EMNLP papers from 2025 and 2026 with Qwen3-Embedding-0.6B and put them on one map. I wanted to know if the two conferences are turning into one. They moved closer, but they are still far from the same. - In 2025, 23% of NeurIPS papers had their nearest neighbour at EMNLP. In 2026 it is 29%. The other way round it went from 22% to 25%. Both use equal-size samples (3,000 papers per side, 50 draws), because NeurIPS is much bigger. - A logistic regression that predicts the conference from the vector gets an AUC of 0.89 in 2026, down from 0.93. - Cutting the 2D map into hexagons, 23% of papers sat in hexagons both conferences share about equally in 2025, and 32% in 2026. Subsampling 2026 to the size of 2025 still gives 31%. - The biggest part of the rise is LLM agents: two thirds of those papers now sit in shared hexagons, up from a third. Benchmarks and LLM judges went from 8% to 40%. - The authors did not converge. At equal size, the share of people with a paper at both did not go up. Abstracts changed style a lot between the two years. “Novel” dropped from a third of NeurIPS abstracts to about one in nine, so I checked that this is not what drives it. A TF-IDF classifier on titles alone also drops, from 0.86 to 0.82, and the nearest-neighbour shares still rise after projecting out the average 2025 to 2026 shift. 30 or 60 topics instead of 40, and EMNLP’s main track only, give the same picture. The 2026 data are pre-release versions of both paper lists. Interactive map (switch years, tap a hexagon to see its papers): [Where NeurIPS and EMNLP overlap — Hadi Mohammadi](https://mohammadi.cv/ai-conferences/2026/overlap) Code and data: [GitHub - mohammadi-hadi/ai-conferences: NeurIPS and EMNLP 2025-2026 on one map: did the two conferences move closer? · GitHub](https://github.com/mohammadi-hadi/ai-conferences) One question for people who work with embeddings: is there a better way to measure how much two corpora overlap than nearest-neighbour mixing on equal-size samples? I would like to know what you would use.