How to actually keep up with AI (without the doom-scrolling)
A filtered, non-hype way to track what's changing: a handful of sources worth the attention, and the noise that's safe to skip.
Most “AI news” is either a model release with a scoreboard nobody outside a lab cares about, or a hot take about what it means for jobs. Neither tells a team running a real business what changed this month. Here's a leaner way to filter it.
Track capability changes, not release dates
A new model number isn't news. What it can now do that the previous one couldn't is what matters: longer context, cheaper inference, reliable tool use, a lower hallucination rate on a specific task. Skim release notes for the capability line, skip the benchmark chart.
Watch price-per-token, not just quality
The most business-relevant AI news of the last two years has mostly been about cost, not intelligence: models that were frontier-only a year ago are now cheap enough to run on every request. That shift quietly turns “too expensive to automate” into “worth automating,” and it's the kind of change that doesn't trend on social media.
Ignore anything that can't answer “compared to what?”
“AI will replace X” and “AI can now do Y” claims are only useful next to a baseline: replace it at what cost, do Y how reliably, under what conditions. If a claim doesn't survive that question, it's marketing, not information.
A short, honest list
The model providers' own changelogs (not their blog posts), a couple of independent eval leaderboards for the specific task category that matters to your work, and (if it's useful to say so) this newsletter, which exists to filter exactly this noise down to what's worth recommending.
Get new posts by email
AI updates, trend breakdowns, and picks for the certifications worth your time, a few times a month, when something's worth sending.
Have something like this to automate?