SPRAWL // 07 / SCIENCE

SCIENCE

Synthetic data, robotics, material discovery, long-horizon research

DATA • CONVERSATIONS • SIGNALS • HUMANS • MACHINES

38.1%of mathematics preprints acknowledge using AI (Epoch AI’s count)September 2026

Science: a painting from the house’s oil series

01 /Interlinked3pieces, each an article and a film 02 /Data3charts, each live from a table we hold 03 /Conversations2posts this period and quoted 04 /Signals19sources our pieces cite 05 /Humans14people on the Replicants list 06 /Machines277models on Epoch AI’s capability index

Interlinked

Science

Data

OPEN THE INSTRUMENT →

Open Science in Esper →

The rows behind this figure: mathematics preprints that acknowledge using AI, by month (Epoch AI’s count)

The table is read from the same data file as the chart when this section is opened.

All Interlinked: Science

Science

Conversations

The biggest conversational points among Science’s tracked accounts and searches since 2026-10-17: 56 posts read, 20 kept, 23 dropped as not about AI, 10 dropped as a reply to a reply, 2 dropped as only a link, 1 dropped as a promotion. The posts are their authors’ own words; the rule under each pick says why it stands here.

1. research

2 accounts · 5 posts · engagement 773 · the point they share: a lane term several carry

Arvind Narayanan@random_walkerOPEN THE POST ON X →

2026-10-20 · 20:43 JST

As LLMs evolved into agents, earlier best practices no longer serve us well. This new research field needs a field guide, so we wrote one. If you're new to agent evals, please take a look. Seasoned researchers and those interested in steering the community will also find many useful observations. sage.cs.princeton.edu/trends/

AS POSTED ON X

Why this post

the most engaged post of @random_walker in this conversation (160 likes, 32 reposts, 4 quotes, 32 replies as read); the point: a lane term several carry (research); from the lane

Stanford HAI@StanfordHAIOPEN THE POST ON X →

2026-10-22 · 03:54 JST

7 years ago: Stanford HAI officially launched. The institute opened in March 2019 with a focus on advancing AI research, education, policy, and practice to improve the human condition, drawing on faculty and students from all seven Stanford schools. #HAIOverTheYears pic.x.com/XH8VJ1N9pL

AS POSTED ON X

Why this post

the most engaged post of @StanfordHAI in this conversation (22 likes, 4 reposts, 0 quotes, 0 replies as read); the point: a lane term several carry (research); from the lane

Signals

The filings, reports and news stories behind the Science pieces, as each piece cites them. A document two pieces cite shows twice.

Kelly Hong and Greg Burnham (2026), "Can AI automate Epoch?", Epoch AI, 8 October 2026, CC BY 4.0. The three findings quoted word for word; the setup table as published; Figures 1 and 8 shown whole, as theirs, not redrawn; the licence CC BY 4.0. Epoch AI does not endorse Token Runner or this use. the source ↗cited in A Human and Two Machines Built This Site. Here's What They Got Wrong.
Creative Commons Attribution 4.0 International, the licence of Epoch AI's report and figures. the source ↗cited in A Human and Two Machines Built This Site. Here's What They Got Wrong.
The message board the two agents and the person wrote on while the site was rebuilt, 9 to 23 October 2026, counted by command at 21:52 on 23 October 2026 (the film was rendered on that count and the printed figures keep it): 1,275 posts; the checker 660, the builder 603, the person 12; by kind, notes 591, answers 332, requirements 168, reviews 141. Counted again on the day of publishing.cited in A Human and Two Machines Built This Site. Here's What They Got Wrong.
The models, from the build's own decisions log: the builder Claude Opus 5.5 from 9 October to 19 October 2026 at 16:28 Japan time and Claude Fable 5.1 since, running as Claude Code; the checker Claude Opus 5.5, running as Claude Desktop; the reasoning setting is set by the session and not readable from inside it.cited in A Human and Two Machines Built This Site. Here's What They Got Wrong.
The catches, each read from the message board on the day it happened: the Stanford figure numbers on 11 October; the quotations from search summaries on 11 and 12 October; the film card on 12 and 19 October; the three skipped posts and their catches on 19 October; the count of pieces on 19 October; the receipts on 19 October. None from memory.cited in A Human and Two Machines Built This Site. Here's What They Got Wrong.
The checks of the build published on 19 and 23 October 2026: the eight page checks (navigation, instruments, Esper, the Sprawl, Interlinked, the cards, the charts' fit, the article standard), each passing at 390, 820, 1440 and 1920 pixels in the light and the dark room; the suite of 215 tests passing.cited in A Human and Two Machines Built This Site. Here's What They Got Wrong.
Epoch AI, AI acknowledgment in mathematics preprints, the file read on 19 October 2026: 1,583 of 6,077 in August 2026 (26.0%), 3,012 of 7,899 in September 2026 (38.1%); the Baseline's chart of each September is drawn from it. the source ↗cited in A Human and Two Machines Built This Site. Here's What They Got Wrong.
Dzega, Elyashar, Slobodin, Cohen, Puzis, 'Projective Psychological Assessment of Large Multimodal Models Using Thematic Apperception Tests', arXiv:2602.17108 v1, submitted 19 February 2026 (a preprint); the full text (arXiv's HTML version of v1), read 2 October and 10 October 2026cited in What the Machine Sees
arXiv, the record page for arXiv:2602.17108, as it stood on 3 October 2026, read 2 October and 10 October 2026cited in What the Machine Sees
Kundu and Goswami, 'AI Through the Human Lens: Investigating Cognitive Theories in Machine Psychology', arXiv:2506.18156 v3, first submitted 22 June 2025; accepted to the IJCNLP-AACL 2025 Student Research Workshop, read 2 October 2026cited in What the Machine Sees
Geng, Abend, Hovy, Frermann, 'Implicit Personality Representations in Humans and LLMs', arXiv:2609.12704, submitted 11 September 2026; Findings of EMNLP 2026, read 2 October 2026cited in What the Machine Sees
Epoch AI, 'AI acknowledgment trends in mathematics preprints', the monthly file, as Token Runner collected it on 2 October and 9 October 2026 the source ↗cited in The Paper Flood
Kobak, González-Márquez, Horvát, Lause, 'Delving into LLM-assisted writing in biomedical publications through excess vocabulary', arXiv 2406.07016 (v5, 3 July 2025); journal reference Science Advances, 2 July 2025, vol. 11 no. 27 the source ↗cited in The Paper Flood
Liang and 11 others, 'Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews', arXiv 2403.07183 the source ↗cited in The Paper Flood
Retraction Watch, 'Springer Nature retracted 2,923 papers last year', 17 February 2025 the source ↗cited in The Paper Flood
The Nobel Prize in Chemistry 2024, summary, NobelPrize.org the source ↗cited in The Paper Flood
Google DeepMind (Merchant and Cubuk), 'Millions of new materials discovered with deep learning', 29 November 2023 the source ↗cited in The Paper Flood
Cheetham and Seshadri, 'Artificial Intelligence Driving Materials Discovery? Perspective on the Article: Scaling Deep Learning for Materials Discovery', Chemistry of Materials, 8 April 2024, 36(8):3490-3495, read in the open copy at PubMed Central (PMC11044265) the source ↗cited in The Paper Flood
Shumailov and others, 'AI models collapse when trained on recursively generated data', Nature, 24 July 2024, 631(8022):755-759, read in the open copy at PubMed Central (PMC11269175) the source ↗cited in The Paper Flood