Most AI News Is Noise. Here's How to Find the Signal

Confession from people who curate AI news daily: most of it doesn't matter. Not "matters less" — genuinely doesn't matter, to anyone, within a fortnight. The launches that fizzle, the threads that age like milk, the "breakthroughs" that were demos.

Yet a small fraction matters enormously — the capability shifts and pricing changes that quietly rearrange how work gets done. The skill of our era isn't following AI news. It's telling these two categories apart in seconds. Five questions do it.

Question 1: Can Someone Use This Today?

The cleanest filter. "Available now" is news — it changes what you can do this week. "Coming soon", waitlists, cherry-picked demo videos and research papers are futures contracts, mostly worthless at the headline stage. The gap between announced and shipped in AI regularly runs a year, and shipped versions routinely underperform their demos. Discount accordingly; revisit when it ships.

Question 2: Capability, Price, or Opinion?

Three story types, wildly different values:

• Capability news — "models can now reliably do X" — is the rarest and most valuable. A few genuine capability shifts per year change workflows everywhere.
• Price news is criminally underrated. "The same capability now costs 90% less" transforms what's feasible for small businesses — and never trends, because cheap isn't exciting. Watch for it deliberately.
• Opinion — someone's take on the above — is entertainment. Fine with your feet up; it just isn't information about the world.

Question 3: Who Benefits From Me Believing This?

A vendor's benchmark chart, a founder's "this changes everything", a doom thread from someone selling doom consulting — every AI story has an economic author. That doesn't automatically make it false; it tells you the direction of the exaggeration. The corrective habit: for anything that matters, find one source with no stake in it before updating your beliefs.

Question 4: Does It Survive the Two-Week Test?

Imagine reading this headline fourteen days from now. Would it still matter? Model-beats-model drama and launch-day discourse almost never survive; pricing changes, capability shifts and regulation almost always do. If you only remember one filter, this is the one — it's also the argument for a daily curated page over a live feed, since curation applies the test for you. (That's the design brief of our AI News page.)

Question 5: Signal for Whom?

The final filter is personal. A genuine breakthrough in protein folding is world-class signal — for biologists. For a marketing manager in Colombo, it's high-quality noise. Signal isn't a property of the story; it's a property of the story times your work. This is why the "so what" note in our 10-minute daily routine matters: it forces the multiplication.

The Noise Diet, Summarised

Shipped over announced. Capability and price over opinion. Follow the incentives. Two-week test. Multiply by your job. Run those five and the daily AI firehose drops to two or three items worth your attention — which is what the calm, genuinely-informed people were reading all along.

The infrastructure for the habit: one curated daily page, ten minutes, and — if you have a team — a 15-minute weekly briefing to spread the signal without spreading the noise. Understanding how the underlying tech works helps the filters too: How LLMs Actually Work makes hype measurably easier to smell.