AI Signal

A curated, structured reading layer for the AI field.

Not a link dump. Not a Twitter list. A community-maintained set of sources with depth ratings, audience tags, and landmark post annotations — built for people who want signal, not noise.

→ Browse the site → Download OPML — import into Feedly, Reeder, NetNewsWire, Inoreader

---

Why this exists

AI is moving fast enough that:

• A blog post from 6 months ago may already be obsolete
• Good signal is scattered across arXiv, Substacks, GitHub, lab blogs, and Twitter threads
• "Awesome AI" lists are alphabetical link dumps with no curation signal

This repo is the data layer. Community PRs keep it updated. GitHub Actions flags stale sources automatically.

Structure

The entire dataset lives in sources.yaml. Everything else is generated from it.

- name: "Lilian Weng"
  url: https://lilianweng.github.io
  rss: https://lilianweng.github.io/index.xml
  type: individual        # lab | individual | applied | systems | paper-explainer | news
  depth: deep             # shallow | medium | deep
  audience: [researcher, ml-engineer]
  tags: [transformers, rl, diffusion, survey, agents]
  activity: active        # active | slow | archived
  last_checked: 2026-05
  landmark_posts:
    - title: "Extrinsic Hallucinations in LLMs"
      url: https://lilianweng.github.io/posts/2024-07-07-hallucination/
      why: "Best taxonomy of hallucination types — required reading before building any production LLM system"

Generated outputs

File	What it is
sources.yaml	The canonical data — edit this
sources.opml	Generated — do not edit by hand
index.html	The website — its data block is generated; run generate_opml.py to keep it in sync

Source types

Type	What it means
lab	Official research blogs from AI labs
individual	Individual practitioners or researchers
systems	Infrastructure, inference, serving
paper-explainer	Newsletters and blogs that explain research
applied	Company engineering blogs building with AI
news	Weekly digests and news

Depth ratings

Rating	Means
deep	Assumes strong technical background; rewards careful reading
medium	Accessible to engineers without ML specialisation
shallow	Good for staying oriented; low time commitment

Activity status

Status	Means
active	Posts at least once every ~6 weeks
slow	Infrequent but posted in the last 18 months
archived	No longer updating — kept only if landmark posts exist

---

Using the OPML

Import sources.opml into:

• Feedly — Add Content → OPML import
• Reeder (iOS/macOS) — Add Account → OPML
• NetNewsWire — File → Import Subscriptions
• Inoreader — Preferences → Import

The OPML is grouped by source type so you can subscribe to just the categories you want.

---

Contributing

Everything flows through sources.yaml. No special tools required.

The one rule

After any edit to sources.yaml, regenerate the OPML before committing:

python generate_opml.py
git add sources.yaml sources.opml
git commit -m "your message"

That's it. The site redeploys automatically on push.

Adding a source manually

1. Edit sources.yaml — add your entry following the schema above
2. python generate_opml.py
3. Open a PR

Required fields: name, url, type, depth, audience, tags, activity, last_checked. rss is strongly preferred — sources without it won't appear in the OPML. landmark_posts are optional but highly valued.

Updating activity status

If a source has gone quiet, open a PR:

• No posts in 90+ days → activity: slow
• No posts in 18+ months → activity: archived

Removing a source

Sources are removed only if they've been archived for 6+ months with no landmark posts. Open an issue first — sometimes someone has context on a temporarily quiet source.

---

Automation

A GitHub Actions workflow runs every Monday and:

1. Checks every RSS feed for recent activity
2. Flags sources with no posts in 90+ days
3. Opens a GitHub issue with the stale list for community review

---

Optional: AI-assisted curation

If you have Claude Code (Claude Pro), two slash commands can run a full curation pass — checking every source via web search, surfacing new candidates, and editing sources.yaml directly.

This is completely optional. The manual workflow above is the primary path.

Command	What it does
/scan-sources	Checks all sources for recent activity, updates activity and last_checked
/curate	Full pass: activity scan + landmark detection + new source discovery

After either command runs, review and commit:

git diff sources.yaml          # read what changed
# edit sources.yaml directly if you want to adjust anything
python generate_opml.py        # always regenerate after any edit
git add sources.yaml sources.opml
git commit -m "curate: YYYY-MM-DD"
git push

See SETUP.md for the full setup and workflow.

---

Licensed CC BY-SA 4.0. Not affiliated with any listed organization.