Sources and method

Which papers

Notable data and AI research papers from the labs and companies the Glyph sites follow (OpenAI, Anthropic, Google, Google DeepMind, Meta, Microsoft, NVIDIA, Databricks, Snowflake, DeepSeek, Alibaba, Ai2 and more every week) and the universities behind the field's core methods. The first papers were picked by hand: the most cited work that shaped today's models, and notable recent papers from those labs. Each paper sits under one lab, the one its authors or the paper name first.

New papers each week

Every Monday at 12:15 UTC the refresh reads the last seven days of Hugging Face daily papers, plus two older weeks, walking back until it reaches the first daily papers, so the archive deepens every run. A paper is added when its Hugging Face organisation is a tracked lab and readers gave it at least 25 upvotes. The same run reads a few labs' own Hugging Face papers lists, labs with no papers here first, and adds their papers from the last year or with 10 or more upvotes. New papers carry an Unverified badge until checked by hand.

New labs each week

The lab list is not fixed. Every company in the companies.fru.dev registry is looked up on Hugging Face, and looked up again every few months when it had nothing; a company whose own organisation there lists papers becomes a lab, and its papers are picked up the same way. Research passes add companies that publish on their own research pages: each paper is taken only from the company's official channels (research page, engineering blog, official GitHub, its own Hugging Face organisation) and checked against arXiv.

What is not covered yet

• Papers without an arXiv id (technical reports and conference papers published only on a company's site).
• Papers from companies with no Hugging Face organisation, until a research pass reads their own research page.

The one-line summaries

Hand-written from the abstract for the first papers. New papers start with Hugging Face's own summary, marked as such on the page, and are rewritten in plain words by one batched model call a week when a model is available (no web access; it reads only the abstract). The line is a pointer to the paper, never a substitute for it.

Citations and upvotes

Citations and influential citations come from the free Semantic Scholar Graph API in one batch request a week; "influential" is Semantic Scholar's count of citing papers that build on the work. Upvotes come from Hugging Face papers. Both are signals of attention, not of quality.

History

Nothing that changes is overwritten silently. Every new count, rewritten line, topic change and review decision writes a row with the old value, the new one, when, and the source. The Changes page lists them and /api/changes serves them. Rejected papers stay in the database, hidden.

Across fru.dev

Lab slugs follow the shared registry at companies.fru.dev; lab pages link to the same company on sibling sites, and papers that name a model link to its product releases on releases.fru.dev. /api/companies lists every lab with its registry slug, domain and dated papers so the registry can add them to each company's timeline.

Disclaimer

Papers from the labs' own pages, arXiv and Hugging Face, found by a Monday agent and reviewed by hand; unverified ones are marked. Citations come from Semantic Scholar, and the list grows as new papers appear. One-line summaries are for orientation, not a substitute for the paper. Logos via logo.dev; trademarks belong to their owners. Anonymous usage stats via Google Analytics. Papers belong to their authors. Use Suggest a correction on any page.