Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Claude Isn't Allowed to Write Me Prose (kvit.app)
    —discuss
  2. Religion, Consciousness, and Continuity (alexahn.com)
    —discuss
  3. Rasmus Hauschild on X: "just vibe-coded a flight simulator in ~2 hours (twitter.com/rasmushauschild)
    —discuss
  4. How to do agentic software development properly (fiodar.substack.com)
    1comments
  5. Blygger: AI-intertwingled diachronic-synchronic social publishing – blygger (blygger.org)
    —discuss
  6. Ask HN: Fong Test?
    —discuss
  7. GPT-6.1 Sol: Release Intelligence, Performance and Price (artificialanalysis.ai)
    —discuss
  8. Ask HN: OpenSEO inspiration for other tools, asking for opinion
    —discuss
  9. Efficient Image Steganography Method (nature.com)
    1comments
  10. Anthropic's $518B AI buildout hinges largely on deals that cannot be canceled (reuters.com)
    —discuss
  11. Spencer Pratt on X: most political advisors are so bafflingly incompetent (twitter.com/spencerpratt)
    —discuss
  12. Spaghetti Celesti – Devour the Heavens (punkcake.itch.io)
    —discuss
  13. OpenAI Agent Bypasses Internet Restrictions Through DNS (circleid.com)
    —discuss
  14. Annalist – required GitHub check for what a coding agent wrote (github.com/gregdixonmxn)
    —discuss
  15. Why has Shopify dropped React Native? (pragmaticengineer.com)
    —discuss
  16. The AI margin collapse is gathering pace (martinalderson.com)
    —discuss
  17. An LLM Workflow That Reproduces, Improves, Extends Published Economics Research (nber.org)
    —discuss
  18. Show HN: Relay – a harness for AI coding agents that recover and verify (relayevals.com)
    —discuss
  19. Show HN: Mluva – Dictation and editable rewrites for Omarchy (github.com/1vecera)
    —discuss
  20. You Said No MCP (earendil.com)
    —discuss
  21. Car Is Sharing Data with Big Tech (consumerreports.org)
    1comments
  22. Agentic Hacks, Real Proofs: Inside Google's PageBreak Project (blog.google)
    —discuss
  23. GPT-6.1 Sol (Max): Intelligence, Performance and Price Analysis (artificialanalysis.ai)
    —discuss
  24. Codex Security Cloud (chatgpt.com)
    —discuss
  25. Attack your own AI agent in under 10 minutes – then secure it before deploying (humanbound.ai)
    —discuss
  26. Password-protecting an Nginx proxied site (aweirdimagination.net)
    —discuss
  27. Automating eval design and hillclimbing with Claude (claude.dev)
    —discuss
  28. The Life and Death of Microsoft Clippy, the Paper Clip the World Loved to Hate (artsy.net)
    —discuss
  29. Electrification efficiency: The world will need less energy after the transition (hannahritchie.substack.com)
    —discuss
  30. GPT6.1 Sol
    —discuss

An LLM Workflow That Reproduces, Improves, Extends Published Economics Research

2 pointsby 10m agonber.org
1 comments
56s agoHN ↗

Across 4,452 published replication packages for five economics journals, the workflow flags discrepancies in 3,460 articles or their appendices.

If this result can be confirmed, and the methodology can be applied to other disciplines, this could really shake up the world of academic publishing. Serious discrepancies can hopefully be flagged and addressed before publication. It may also allow authors take a harder look at earlier papers they're citing.

I would hope that it will also reduce the numbers of subpar papers and fraudulent research submitted for review, but as we've seen throughout history the cheaters always find a way to get ahead.