AI for Academic Research & Publishing¶

Vibe research, feedback machines, AI-assisted academic writing, and the future of scholarly work.

Summaries¶

AI for Business & Economic Research: From Chatbots to Agents — Velikov's Penn State Smeal Community-of-Practice deck synthesizing the wiki for a faculty audience
EDHEC Master Class: Using AI for Research — Velikov's 6-hour EDHEC class: two Beamer decks + six runnable Claude Code demos (skills, FRED agent, referee, strategic-revision, CAPM replication pipeline)
Vibe Research — Gregoire wrote a finance paper in 4 days with AI
Vibe Research II: Three Days with Fable 5 — Grégoire on Fable 5: closed-form proofs, autonomous video walkthroughs, and access revoked by US government order
Reflections on Vibe Researching — Gans's year-long AI-first experiment and its failures
Feedback Machines — Bäckman on structured paper feedback with Claude/Cursor
Prompts-to-Paper — Chen uses AI co-authors (o1, Sonnet, Deep Research) to write a theory paper on hedging AI risk
Stress-Test Research Pipeline — Lopez-Lira's pipeline for evaluating PhD research ideas
Generative AI for Research (UCT Course) — Shock's 12-week NQF Level 9 postgraduate course: LLM foundations, ethics, literature review, writing, data analysis, agents, and Claude Code advanced track
Claude Code for Academics: APSA 2026 (Spina) — Comprehensive practical guide: CLAUDE.md/session-log/DATA.md scaffold, Cunningham Conjecture, skills vs commands, Plan mode, safety, and the "bottleneck is you" honest assessment

AI for Economic Theorists & Mathematicians (Markus Academy 166 overview) — Ortoleva & Sandomirskiy's four-episode series: theory as a fixed point, creativity evidence from math, model choice, agent swarms
Core Uses of AI for Economic Theory (166-1) — Ortoleva's seven use cases; sketching as most underused; proof attack/repair/inspiration; the two-weekend rabbit hole
Can AI Be Creative? (166-2) — Sandomirskiy on the Erdős retraction, poor attribution vs. superb aggregation, and three conjectures that fell
Theorist Toolbox (Koren) — Claude Code skills for proofs: single-pass math-proof, adversarial codex-math, and a six-agent co-math team with reviewer gates, Lean verification, and strict-mode hooks

AI One-Shot Papers — Goldsmith-Pinkham on AI-generated papers' over-reliance on diff-in-diff
Can AI Replace Researchers? — Karpf arguing the real crisis is broken incentives
Research in the Time of AI — Goldsmith-Pinkham's research pipeline ontology
OpenAI Automated Researcher — OpenAI's plan for fully automated research by 2028
When Will the Research Paper Disappear? — Cowen on the obsolescence of static papers
Generative AI for Economic Research — Korinek on use cases and implications for economists

How I Use Claude Code for Empirical Research — Cunningham's MixtapeTools, Referee 2, cross-language replication
Multiple Agents Auditing Your DiD Code — Hallucination as measurement error, cross-language verification
Can Claude Code Find Facts? — Automated DiD paper in 3.5 hours, epistemological questions
Integration & Collaboration (Markus Academy 162-8) — Goldsmith-Pinkham on verification debt, commits as review units, GitHub Issues as feedback, Overleaf sync; live IPO replication that surfaces two real agent errors
Integration and Collaboration in AI Research Work (Substack companion) — The written protocol: verification debt, DECISIONS.md/LOG.md/attrition tables, issues as feedback, "provenance ≠ correctness"
Verifying LLM Output (Bäckman) — Practical verification protocol: git diffs, fresh-agent review, second model, /quiz-me comprehension skill, spec-only reimplementation in another language, and decisions.md
Read the Paper, Write the Code — Kohler et al. (ETH): 48-paper benchmark; OpenCode GPT-5.4 hits 91% sign, >80% within 95% CI without access to original code
Introducing Living Science — Tan (UChicago): agent-driven replication + extension of influential econ papers on new data (min. wage, Shimer ins/outs, Blanchard–Katz), published as a public living record

Research and Publishing Are Now Two Different Things — Supply-demand analysis of manuscript production costs
A Modest Proposal for Editors — Little's Law applied to journals, LLM desk screening
Four Criteria for Using Agents (Referee Reports) — Cunningham's split-pdf → deck → referee2 → blindspot → tikz pipeline
Reviewer: Multi-Agent Reviewer for Econ Papers — Haaland's 10-step Codex pipeline with schema-validated reviewers, parser repair, privacy-first repo hygiene
Point by Point: R&R Response Letters — Nagaraj's web app: per-concern workspaces, completeness tracking, LaTeX export

Teaching AI Your Voice — Blattman's method for preserving authorial style
Document Collection — Blattman's "email as database" pattern for document retrieval
Writing & Thinking (Markus Academy 162-5) — Goldsmith-Pinkham: style-guide skill, strategic-revision skill, editor-not-rewriter pattern
Writing & Thinking with AI Assistance (Substack) — Goldsmith-Pinkham companion post: writing-is-thinking / homogenization / banal-writing framing