Information streams

A caveat: my days are spent experimenting with this stuff professionally, and my spare time is spent musing, experimenting more, and following the kinds of people listed below. Your mileage may vary.

X.com is unreasonably good.

I know how that may sound. The politicisation, the image problems, the discourse about the discourse. I get it. But it turns out the best minds in this space are very active and engaged there. It’s the best place to keep track of what’s happening, by a country mile.

The catch is that it only works if you follow the right people. Which is what this post is about.

The 80/20 Sources

Before I list accounts, let me save you some time. If you want to keep up with what’s happening in AI with minimal effort, these two sources will get you most of the way:

  • Simon Willison’s blog - Prolific, curious, gentle, early to most topics, yet careful. If you read one thing, read this.
  • swyx’s newsletter - Comprehensive coverage, good interviews, consistently high signal.

This might seem counterintuitive after raving about X above. X is absolutely the best primary source if you follow the right people - which is kind of the point of this post. But Simon and swyx utterly get this. They already follow the right accounts, read the papers, catch the discourse. Their blogs are the best regurgitations of X and similarly high-signal sources. You get the curation without having to do the curation yourself.

You could probably stop there and have the bulk of the payoff.

People Worth Following

That said, if you do want to go to the source:

  • Simon Willison - @simonw - See above. His X is as good as his blog.

  • Steve Yegge - @steve_yegge - Prolific and insightful. Famous for the Platforms Rant, but he never actually stopped writing. His LLM stuff is equally excellent.

  • swyx - @swyx - Runs Latent Space, hosts the AI Engineer Summit, and his book The Coding Career Handbook is terrific. Good at synthesising what matters.

  • Andrej Karpathy - @karpathy - A pretty good common-sense bellwether, keyed into the community through his excellent tutorial videos on ML. Coins neologisms we all end up using. Also has what I think are underappreciated nuanced views on things - underappreciated in spite of his fame.

  • Dwarkesh Patel - @dwarkesh_sp - The single best podcast for understanding the pace of acceleration and its implications, while maintaining reasonable doubts about the “when.” Honourable mention: the Manifold podcast as a runner-up.

  • Matt Turck - @mattturck - Creates the MAD landscape and hosts the MAD podcast. Unusually engaging and technical at the same time, where usually that’s a trade-off.

  • The Sourcegraph crew - Beyang Liu, Quinn Slack, Thorsten Ball - Beyang has a terrific Sourcegraph podcast, but more recently the team working on Ampcode has become an excellent source for day-to-day, down-in-the-weeds questions about coding with LLM agents. Proudly opinionated. I think of them as intellectually moving first on a lot of things rather than simply being reactive. Special shout-out to their Frequently Ignored Feedback page - they’re not going to shower you in options, they tell you what they think and you can use another product if you disagree.

  • Yacine / kache - @yacineMTB - Aggressive but feels very honest. Pretty insightful regardless of how raw his public-facing persona seems. I suspect it’s just the real him - open and unrestrained. Top tier for me even just for getting me using Tailscale. That alone would be good enough.

  • Boris Cherny - @bcherny - Works at Anthropic on Claude Code. I find him quite sensible in his positions. Interesting how none of the hype around Claude Code has actually come from him.

This isn’t exhaustive. I’ll add more as they come up.

A Health Warning

If you’re anything like me and prone to dopaminergic loops of checking and rechecking feeds, you can lose an awful lot of time trying to stay on top of things.

There’s also Cal Newport’s point about context switching - it’s far more cognitively costly than people realise. Studies show people self-report far smaller disruptions than they actually exhibit on cognitive tests when briefly interrupted. You can do surprising amounts of damage to your ability to function if you’re checking social media all day.

This is why the 80/20 sources matter. Simon’s blog and Latent Space give you most of the signal without the slot-machine pull of a feed. For those of us who are susceptible: consider whether you actually need X, or whether the curators have already done the work for you.