Higher Education & ResearchIssue 553 records
Research Universities
Three newly inspected sources on the research enterprise. UChicago's Living Science uses AI agents to reproduce influential papers and rerun them on newer data; a minimum-wage study grown from about 138 to 236 state increases kept its conclusion. Brown and Vanderbilt won $732,574 from Schmidt Sciences for DeepAndes Workbench, an agent-queried archaeology tool with in-kind supercomputing. A 29 October preprint had a coding agent choose and complete its own studies in five pre-registered runs, rated near workshop level by AI reviewers but low in novelty.
- Evidence records
- 3
- Cross-source patterns
- 1
- Evidence classes
- 2 academic research1 vendor claim
- Outcomes
- 2 emerging1 mixed
- Source freshness
- 3 new this fortnight
- Research completed
- 2026-11-03
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Agent-run research is being made credible through released, re-runnable records rather than through agent accuracy claims
Interpretation across two sources from 29 Oct-2 Nov. UChicago's Living Science reproduces published studies with agents and posts code and data so others can check them, while noting parts it could not reproduce. An arXiv preprint had an agent produce whole papers and released claim-evidence logs, Git tags and a verification script. Neither reports agent accuracy rates or human evaluation, so the credibility mechanism is openness and re-computation, which still is not independent replication.
Operating questionWhat records package should an institution require before agent-produced or agent-updated research is published or cited?
Supporting evidenceUniversity of Chicago Department of Computer Science (Chenhao Tan, Evan Rose); SAI LabsKo Fujimura, Otsuma Women's University (single-author arXiv preprint)
Evidence ledger
Full record · every source keeps its link and limitations
UChicago Living Science uses AI agents to reproduce influential papers and rerun them on newer data, starting with a minimum-wage study
Living Science, led by associate professor Chenhao Tan, uses AI agents to reproduce an influential paper from its original materials and rerun the analysis on newer data, publishing updated findings with code and data. First topics are in economics, chosen by collaborator Evan Rose; five AI safety papers are in progress. One early report revisited a 2019 minimum-wage study: the team first matched the published results, then expanded the sample from about 138 to 236 state minimum-wage increases through 2024. The original conclusion held (affected workers earned more, by less than first found, with no clear change in employment), and the report notes parts the team could not fully reproduce. Tan framed the project as constructive rather than a weapon to criticize existing work and invites feedback from original authors.
Why it matters, evidence and limitations
- Why it matters
- Offers research offices and libraries a concrete model for using agents on reproducibility and evidence maintenance, an integrity function rather than new discovery. It also raises questions about credit, author consultation and how institutions present revised findings to policymakers.
- Evidence and measured results
- Read the full university news story. It describes one completed report; agent accuracy, human review effort, cost and failure rates are not reported, and the reports themselves were not opened. Departmental news, not an independent evaluation.
- Limitations and uncertainty
- University self-report of an early-stage project with one described result; how much the agents did versus humans is unclear. Economics with public data may not transfer to fields with restricted data or wet-lab work.
Brown and Vanderbilt receive $732,574 from Schmidt Sciences to build DeepAndes Workbench, an agent-queried AI survey tool for Andean archaeology
Brown associate professor Parker VanValkenburgh and Vanderbilt researchers Steven Wernke and Yuankai Huo received $732,574 from Schmidt Sciences for "Scaling Archaeological Vision: Imperial Landscapes of the Andes through AI-Assisted Survey." They will build DeepAndes Workbench, which lets archaeologists query an agent with images and verbal prompts to find specific feature types, with expert feedback used to improve the models. It builds on GeoPACHA, their satellite-imagery and AI mapping platform. The team includes about 15 collaborators across the US, Peru, Chile and Australia plus Oak Ridge National Laboratory researchers; Schmidt Sciences also provides in-kind supercomputing. VanValkenburgh described it as a tool to extend researchers' vision, not an all-seeing eye.
Why it matters, evidence and limitations
- Why it matters
- Shows AI-for-research funding and national-lab compute reaching the humanities, a segment research computing groups often under-serve. It is also an example of philanthropic in-kind compute supplementing campus resources.
- Evidence and measured results
- Read the full release. This is an award announcement; no model performance, accuracy or user data are given. Expected outputs (continental-scale maps) are aspirations.
- Limitations and uncertainty
- University award announcement (institutional self-report, classed as operator claim); no results yet. Humanities use case with distinctive data; transfer to other fields is limited.
Preprint: a coding agent chose its own research questions in five pre-registered runs, all producing re-computable papers rated near workshop level, with low novelty
An off-the-shelf coding agent, managed by Claude Code with Claude Opus 5.5, was given a roughly 7,300-word instruction document, a question-formulation skill (with one critique from gpt-6-astra under a US$20 budget) and a peer-review skill in which a reviewer agent reruns the study from its records. After 13 development runs, the system was frozen and five consecutive runs (one pilot counted by disclosed amendment, four pre-registered) were all reported. All five went from question to a revised, re-computable paper on a laptop with no human research input; external AI reviewers rated them at or near workshop level, but novelty and generality remained low. Small wording changes to the research-area setting shifted topics substantially. Records include claims-to-evidence logs, API costs, Git tags and a verification entry point.
Why it matters, evidence and limitations
- Why it matters
- Relevant to research offices setting policy on agent-generated papers: the study shows end-to-end agent research is feasible on modest hardware and that full logs and pre-registration make it auditable, while also showing low novelty and AI-only evaluation. It informs disclosure, authorship and review rules.
- Evidence and measured results
- Read the abstract, introduction, design, discussion and limitations. Five runs with one research model, no ablation, AI-only evaluation (human evaluation planned), in-loop reviewer from the same model family, and post-hoc choices disclosed as deviations. The paper itself was drafted by the management agent. Preprint, not peer reviewed.
- Limitations and uncertainty
- Very small sample, single author, AI-only ratings uncalibrated, no human evaluation yet; one research area setting tested. Non-US source; transfer to US institutional settings depends on local policy.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.