No. 001

Infrastructure

• arXiv declares independence from Cornell after 35 years

  Science, March 2026

  arXiv is becoming an independent nonprofit on July 1, 2026, with Simons Foundation backing, a $6M budget, and 27 staff. The CEO hire will signal the direction: a tech-forward leader means arXiv might finally move beyond PDF, while a governance-focused one means stability but slow evolution. Researchers worried about an OpenAI-style drift should note that arXiv's incentive structure is fundamentally different, but worth watching.

• Introducing the Elicit API

  Elicit, March 2026

  Elicit opened programmatic access to its 138M-paper index. Researchers can now run literature searches and generate reports from their own scripts, not just the web UI.

AI Scientists and the Governance Gap

• AI scientists are changing research: institutions, funders and publishers must respond

  Nature, March 25 2026

  Nature's editorial argues that once AI systems autonomously generate hypotheses, design experiments, and interpret results, the questions of credit, accountability, and governance stop being hypothetical.

• How to build an AI scientist: first peer-reviewed paper spills the secrets

  Nature, March 26 2026

  Nature covers the first peer-reviewed account of an end-to-end AI scientist system, focusing less on what it found and more on how the work was structured.

Peer Review Under Pressure

• Gemini provides automated feedback for theoretical computer scientists at STOC 2026

  Google Research, March 2026

  Google tested Gemini as a pre-submission proof checker for STOC. 97% of participating researchers found it helpful; one said it caught a bug that invalidated an entire proof.

Integrity and the Scientific Record

• Librarian finds "preposterous number" of fake references in paper from Springer Nature journal

  Retraction Watch, March 6 2026

  A hospital librarian discovered that 12 of 14 references in a published paper did not exist. Springer Nature said checking for hallucinated citations at scale remains an unsolved problem.

• "Claude Scholar" automates 90% of the academic research lifecycle

  LinkedIn, March 29 2026

  An open-source tool claims to automate literature review, statistical analysis, paper writing, citation verification, and rebuttal drafting, targeting top venues like NeurIPS and Nature. The feature list includes "language refinement to sound human," which tells you everything about the tool's relationship to transparency. Marketed with "professors are going to hate this." They should.

• The perils of using generative AI to perform research tasks: editors' and publishers' viewpoints

  The Scholarly Kitchen, March 9 2026

  Editors and publishers agree AI cannot be banned, but flag a growing divide: publishers optimize for throughput, editors worry about deskilling junior researchers who never learn to engage deeply with literature.