How I Track AI Without Drowning in the News
Every morning my feeds hand me dozens of AI headlines: a new model, a new benchmark, a new funding round, a new claim that everything is about to change. Reading all of it is a full-time job, and most of it will be forgotten within a week. So instead of reading more, I built a filter. This post is that filter, written down.
Start from primary sources
Model releases, pricing pages and official blogs are facts. Commentary about them is opinion, and second-hand commentary is often opinion about opinion. My rule is simple: if a claim matters, I trace it back to the original announcement before I let it change my mind about anything. Screenshots of screenshots are not sources.
Separate capability from marketing
A demo video shows what a system did once, under chosen conditions. What I actually want to know is different: Can ordinary users access it today? What does it cost? What are the rate limits? Does it still work on the tenth try, on a bad input, in a language other than English? The gap between a launch video and a reliable tool is where most disappointment lives.
Watch costs, not just scores
Benchmarks get the headlines, but cost curves change behaviour. When the price of a given level of capability falls by ten times, applications that were silly experiments become obvious products. I keep a rough note of what frontier-level inference costs over time; it has predicted more real change than any single leaderboard.
A weekly routine, not a daily panic
I skim headlines daily but only sit down once a week to decide what was actually important: usually two or three items out of hundreds. Those are the ones I write about. Everything else goes into a "wait and see" pile — most of it never comes back, which is exactly the point.
What I ignore on purpose
Anonymous rumours, engagement-bait predictions, and any chart without a source. Ignoring things deliberately is not laziness; it is the only way the important minority gets enough attention. The goal of tracking AI is not to know everything first. It is to understand the few things that matter, slightly earlier and slightly better than the noise suggests.