AI news moves faster than anyone can read, and most of the volume is repetition. This list collects eleven public newsletters, blogs and podcasts that sit on our own source list, with a plain note on what each is good for. These are our picks from public sources, not a ranking, and we make no claim about readership or quality beyond what each source says about itself. Some pieces are partly paywalled; check each site.

Import AI

A weekly research and policy newsletter by Jack Clark, at importai.substack.com. It is a good fit if you want a short weekly read that connects papers to policy implications. We list it as a weekly check-in rather than a daily feed.

Last Week in AI

A weekly roundup that also runs as a podcast, at lastweekin.ai. Its value is compression: if you missed a week, it gives you the main stories in one place. Use it as a catch-up tool and follow its links to the original announcements.

Interconnects

Nathan Lambert’s newsletter, at interconnects.ai, covers post-training, open models and reinforcement learning from human feedback. It suits readers who want to understand how open-weight models are built and compared. Our guide to open vs closed models is a useful companion.

Ahead of AI

Sebastian Raschka’s magazine, at magazine.sebastianraschka.com, explains large language model research in depth. Choose it when you want to understand a technique rather than just hear that it exists.

Simon Willison’s Weblog

A blog on hands-on language model tooling, at simonwillison.net. We find it especially useful for developer-facing news, since posts tend to describe what a tool does in practice. It is a good place to see how new releases behave when someone actually tries them.

Latent Space

An AI engineering newsletter and podcast, at latent.space, with the podcast at latent.space/podcast. It is aimed at people building products with models, and covers the engineering side of the industry.

Dwarkesh Podcast

Long interviews with researchers and founders, at dwarkesh.com. It suits listeners who want unhurried conversation with the people building AI, rather than news summaries. Treat claims made in interviews as the guest’s views until independently confirmed.

The TWIML AI Podcast

A machine learning research and practice podcast, at twimlai.com. It is useful for hearing practitioners describe applied work and research results in their own words.

Practical AI

An applied AI podcast aimed at developers, at practicalai.fm. Reach for it when your interest is deploying and using models in real systems.

Epoch AI

A data-driven source for compute, data and benchmark trends, at epoch.ai/blog. It is helpful when you want figures behind the narrative, for example how training compute is changing over time. Its numbers are best read alongside our explainer on numbers in AI announcements.

The Gradient

Essays and interviews on AI research, at thegradient.pub. It suits slower, essay-length reading when you want context rather than headlines.

How to combine them

Start by deciding what you need from news. If you work with models every day, you want speed and practical detail, so a developer blog and an engineering podcast matter more than a research essay. If you follow policy or business, weekly synthesis and data-driven analysis save time. If you are learning, favor sources that explain a technique step by step and revisit older posts, because the fundamentals change more slowly than the headlines.

Set limits as well. Following more than five or six sources usually produces duplicate coverage of the same announcement, with each outlet adding its own framing. A simple test is to ask, for each item, what new fact it contains. If the answer is none, skip it. Keep a short list of links you actually acted on over a month, and drop the sources that never made it onto that list.

A workable routine is to pick one daily skim source, such as a link roundup, one or two weekly newsletters for depth, and one podcast for long-form context. Then go to primary sources for anything that matters to you: a lab’s own announcement, the paper on arXiv, a regulator’s text or a leaderboard page. Newsletters are commentary; they can be wrong, selective or early. When a claim comes only from social posts or forums, treat it as unconfirmed until a primary source backs it up. We label such stories that way on this site.

What to watch next

  • Whether your favorite sources start covering new areas such as agent security or inference hardware.
  • Shifts toward podcasts and video for technical interviews.
  • Changes in what is free versus paywalled at the newsletters you use.
  • Primary-source feeds from labs, which you can add directly to a reader.

Key takeaways

  • Mix one daily skim source with weekly depth and one long-form podcast.
  • Weekly roundups are catch-up tools; follow their links to originals.
  • Developer-focused blogs show how releases behave in practice.
  • Data sources like Epoch AI add numbers; read them with caution.
  • Confirm important claims against primary sources.

Frequently asked questions

Are these ranked?

No. They are our picks from sources we monitor, in no order of quality.

Do I need to pay?

Some are partly paywalled; many are free at least in part. Check each site.

What should I read first if I am new?

Start with one weekly roundup and one explainer-style newsletter, then add a podcast.