The Stance Map · Free
Every AI source.
Plotted by lean.
018
sources tagged · across 7 editorial stances
You shouldn't have to read twelve newsletters to know whether the latest AI thing is real, hyped, or quietly dangerous. Here's how I tag every outlet I pull from — and what each tag actually means.
Legend
What each tag actually means.
AI is overblown or dangerous
Sources arguing AI capability is overstated, deployment is reckless, or societal harm outweighs benefit. Examples: Gary Marcus, Emily Bender, AI ethics academics. Tagged when the source's core argument is anti-deployment, not merely cautious.
Skeptic
Stories that lead with what's not working. The counterweight to launch posts. Read these BEFORE you decide what to ship.
Academic
Papers and research blogs. The truth is in here somewhere, often before anyone else has it. Slow to parse, high payoff.
Safety
Stories framed around model behavior, misuse, alignment, and policy. Slow-cycle but high-stakes. The future depends on these being right.
Builder
Stories from people who ship. Concrete details, working examples, real tradeoffs. The highest signal-to-noise category.
VC
Stories about market structure, funding, and competitive moats. Trustworthy on strategy, less so on technical claims.
Hype
Stories framed for emotional response and shareability. Useful for what everyone is hearing. Discount the adjectives.
The Map
Every source. Every tag.
- Hacker News ↗Tier 1
The articles are mixed. The comments are where the skeptic stance lives. Often the best critique of a hype post is in the top reply.
Princeton researchers documenting where AI claims don't survive contact with evidence. The grown-up skeptic position.
- Google DeepMind ↗Tier 1
Deep research, often consciousness- and safety-adjacent. Slower cadence. Worth reading in full when it lands.
The paper-of-the-day signal. Upvote count is a usable popularity metric. Read the abstract, skim the discussion.
- arXiv (cs.AI, cs.CL) ↗Tier 1
Firehose. 50–200 papers a day. Read by lab, not by feed. Filter to labs you already trust + papers with social-media velocity.
Economist who reads more AI papers than most AI researchers. Not all AI, but the AI takes are always worth reading.
- Import AI (Jack Clark) ↗Tier 2
Weekly. Long-context safety + policy takes by Anthropic's co-founder. The most thoughtful single newsletter in the space.
- Anthropic ↗Tier 1
First-party model and research announcements. Treat as primary source. The framing is honest but it's their product.
- OpenAI ↗Tier 1
Product launches and model cards. Higher marketing-density than Anthropic. The capabilities claims are usually real; the wording is always optimistic.
- Google AI ↗Tier 1
Gemini, Gemma, Workspace. Heavier on enterprise framing than DeepMind. Watch for capability vs. availability mismatches.
- Meta AI / FAIR ↗Tier 2
Llama releases and FAIR papers. Strategy is open-weight-as-commodity — read for the moves more than the moments.
- Microsoft AI ↗Tier 2
Copilot, Azure AI, enterprise integrations. Almost always partnership news. Skip unless it changes pricing or distribution.
- Latent Space ↗Tier 2
Podcast and newsletter. Builders talking to builders. Transcripts are excellent. The 'how it actually works' counterweight to vibes posts.
- TLDR AI ↗Tier 2
More substantive than the engagement-optimized dailies. Often surfaces tools and infra news before the bigger feeds.
- X / AI researcher list ↗Tier 3
Where releases get debated in real time. Highest noise, also highest leading-indicator signal. Curate a tight list or it'll eat you.
Strategy and market-structure analysis. Paywalled. Read when a market-structure question matters more than a capability question.
- Ben's Bites ↗Tier 2
Daily AI newsletter optimized for clicks. Useful as 'what is everyone else covering.' Discount the framing by half.
- The Rundown AI ↗Tier 3
Similar to Ben's Bites — daily, breadth over depth, formatted for sharing. Useful as a coverage map, not as a take.
How I read it
Three reading orders, depending on what you came for.
If you have 10 minutes
Read the Tier-1 builder sources only. Anthropic, OpenAI, Google AI. You'll miss the takes but you'll know what changed.
If you have 30 minutes
Builder sources, plus one skeptic source (Hacker News comments, or AI Snake Oil). The skeptic pass keeps you from over-indexing on launch-post optimism.
If you're trying to ship
Builder + academic. Skip the hype dailies entirely — they will cost you more time than they save. Latent Space transcripts are the highest hourly-return read in the entire list.
Or skip the list
Let me read all 18 for you.
The Monday digest: one email, the week's stories, every angle tagged, an actual opinion. Free. Unsubscribe in one click.
Read carefully. Ship carefully.
— Samwise