12.6k stars for a repo that just lists AI papers weekly

What it does

DAIR.AI maintains this repository as a running index of notable AI papers from each week. The README acts as a chronological table of contents, linking out to yearly markdown files—years/2024.md, years/2025.md, and years/2026.md—where the actual weekly roundups are stored. The same curation is also pushed to a Substack newsletter.

The interesting bit

The project has racked up over 12,600 stars despite shipping no code, models, or benchmarks. In a field flooded with daily preprints, a simple human-filtered list turns out to be a surprisingly valuable commodity.

Key highlights

• Weekly coverage running from 2024 through mid-2026, suggesting consistent upkeep
• Broad scope spanning AI, machine learning, deep learning, and NLP (per the repo’s stated topics)
• Pure curation: no code or dependencies, just markdown indexes and hyperlinks
• Tied to the nlpnews.substack.com newsletter for readers who prefer email delivery

Caveats

• The README is strictly an index; you must click through to the yearly files to see the actual paper lists
• Selection criteria for what qualifies as a “top” paper are not described in the README, so the filtering logic is opaque
• Navigation is chronological only, with no evident search or subfield categorization in the README itself

Verdict

Worth bookmarking if you want a low-friction pulse on weekly ML highlights without managing your own arXiv alerts. Skip it if you need code, reproducibility notes, or detailed reviews stored directly in the repository.