The most reliable way to keep up with AI is to read where announcements start, then add a few newsletters and podcasts for interpretation. A routine of about 90 minutes a week is enough for most professionals: a 10-minute daily skim of primary-source feeds, one weekly newsletter, one long listen and one paper or technical post you read properly. This guide explains how to set it up, why primary sources come first and how to avoid the common traps.

Why start with primary sources

Most AI headlines trace back to a handful of originals: a lab’s blog post, a paper, a model card, a pricing page or a court filing. Secondary coverage is useful, but it adds a layer of summary, and summaries drop caveats. Reading the original takes minutes and lets you judge claims yourself, such as which benchmark was used and who ran it. Our own newsroom works the same way: we link the primary source first and treat vendor claims as claims until a neutral party confirms them.

This guide complements our curated list of podcasts and newsletters, which focuses on what to follow. Here the focus is how to combine sources into a habit.

Build a primary-source feed

Use any RSS reader and subscribe to a short list of feeds. Good starting points:

  • Lab and company blogs. OpenAI publishes an RSS feed of its news, and Google DeepMind and Hugging Face run blogs you can follow the same way.
  • arXiv. The cs.AI recent listing shows new preprints daily. Treat preprints as unreviewed unless a venue says otherwise.
  • Curated paper lists. Hugging Face Daily Papers surfaces papers the community is discussing, which saves sorting through hundreds of submissions.

Keep the list under about 15 feeds. More than that, and you will stop reading. Mark items as read without guilt; the goal is awareness, not completion.

Add interpretation: newsletters and podcasts

Primary sources tell you what was announced; newsletters and podcasts tell you what it might mean. Pick one of each rather than five:

  • A weekly newsletter. The Batch from DeepLearning.AI is aimed at practitioners, while Import AI by Jack Clark focuses on research and policy. Choose by your role.
  • A long-form show or newsletter on building. Latent Space covers AI engineering for people shipping products.

Read newsletters for the links as much as the commentary. If an item matters to you, open the original it cites.

A 90-minute weekly schedule

  1. Daily, 10 minutes: skim the feed, open at most two items, save the rest for the weekend.
  2. Midweek, 20 minutes: read your weekly newsletter and follow two links to originals.
  3. Weekend, 30 minutes: listen to one podcast episode, ideally while walking or commuting.
  4. Weekend, 30 minutes: read one paper, model card or technical post properly, and write three sentences about it.
Headphones resting next to a smartphone
A podcast episode on a commute fits the weekend listening slot in this routine. Photo: wu yi takeshi2 / Wikimedia Commons, CC0

Adjust the mix to your goal. A founder may swap the paper for a pricing page and a competitor’s changelog. A student may read two papers and skip the podcast.

Check claims before you share them

A few habits prevent most mistakes:

  • Look for the number’s owner. Is a benchmark score reported by the lab, a third-party evaluator or a public leaderboard? Our guide on judging AI predictions and forecasts shows how to weigh confident claims.
  • Check the date. Old posts resurface constantly, and model names change quickly.
  • Separate announcement from availability. A launch post may describe a waitlist, a research preview or a region-limited rollout.
  • Treat community posts as leads. Forums and social threads are good for spotting stories, but confirm them before repeating them.
  • Note conflicts of interest. Authors who work for, advise or invest in a company have an angle, which does not make them wrong.

Common mistakes

  • Subscribing to everything and reading nothing.
  • Treating a press release summary as the finding.
  • Confusing a preprint with a peer-reviewed result.
  • Following only voices that agree with you.
  • Never writing anything down, so nothing sticks.

A short notes file helps. For each item you read, record the source, the claim, who verified it and what you would need to see to believe it. Review the file monthly and drop sources that never produced a useful note.

What to watch next

  • More labs offering feeds, changelogs and model cards in structured form.
  • Newsletters and podcasts adding clearer disclosure of affiliations.
  • Leaderboards and evaluation sites becoming standard sources for verifying claims.
  • Reader tools that summarize feeds, and how reliable their summaries are.

Key takeaways

  • Start with primary sources, and use newsletters and podcasts for interpretation.
  • Keep your feed to about 15 sources and your routine to about 90 minutes a week.
  • Open the original behind every claim that matters to your work.
  • Treat preprints and vendor benchmarks as unconfirmed until verified.
  • Write short notes so reading turns into knowledge.

Frequently asked questions

What is the best free way to follow AI news?

Subscribe to lab blogs and arXiv listings in an RSS reader, then add one free weekly newsletter. That covers announcements and interpretation at no cost.

How many AI newsletters should I follow?

One or two is enough for most readers. More tends to produce duplicate coverage and skipped issues.

Are preprints on arXiv reliable?

Preprints are posted before peer review, so treat them as early findings. Check for a venue, replication or independent coverage before relying on them.