AI is getting good at itself — and the models are too
The bettors now read the models instead of just using them. New tricks for steering AI with URLs, Amazon launching satellites, and Google quietly feeding us slop nobody's caught yet. The people who win are the ones learning to see what's happening underneath.
The people betting on AI are learning to read it
A group of forecasters who've been tracking AI progress for years just published a new model that actually tracks the frontier — not by chasing benchmarks, but by predicting which papers and papers will move the needle. It's the kind of thing that sounds small until you realize it's the difference between watching the game and reading the scoreboard.
What makes this one worth paying attention to is that it's built from people who've been doing this work for a long time. The forecasters have been writing about the industry since before the current wave — they've seen what works and what doesn't, and their model reflects that history. The result is a tool that's useful not just for the people who built it, but for anyone trying to figure out where this is going.
Why this matters for us: the people who understand how this technology is actually moving are the ones we'll need when the tools start changing our work.
Xbox just showed us the future and it's bleak
Microsoft wrapped Summer Game Fest with what looked like a win — Halo, Gears of War, Fable, a translucent Xbox, even some old-school Persona and Crazy Taxi games. It had that E3 energy, the kind that makes you feel like the industry is doing well.
But Andrew Webster sees past the party lights. The real story isn't the games; it's what Microsoft is doing with them. The company has been quietly turning Xbox into a subscription play, moving away from selling consoles to selling Game Pass. Consoles are getting thinner, the hardware matters less, and the platform is becoming more about the service than the box in your living room.
This is the future of gaming for the rest of us: less ownership, more monthly fees, games that live in the cloud and the app rather than on discs or drives. For families and communities who buy consoles and keep them for years, the value is real. For the rest, the subscription treadmill never stops.
Why this matters for us: if you're tired of paying for things that used to be yours, Xbox's pivot is the canary in the coal mine for every industry following Microsoft's lead.
The model isn't broken — it's just spread out.
— tech.instacart.com
#how-an-ai-token-actually-travels-through-a-model-and-why-variance-matters-fd45f7How we nudge models using URLs
Anthropic and OpenAI's models now read URLs directly in the prompt. You can point them at a spec, a blog post, or a PR and the model treats it as part of its context window — not as a link to click, but as something to reason about. That changes how we write for models, not just for people.
The trick is that the model doesn't just fetch the page; it parses it and weaves the content into its next response. Write the URL into the prompt the way you'd write a paragraph. The model reads it like a colleague who shows up to the table with a document.
Apple's foldable iPhone runs out of memory — again
Apple is making its first foldable iPhone and hitting the same wall that has bugged every company trying to scale these screens: memory. The Chinese chipmakers CXMT and YMTC, which supply the DRAM for the foldable's display panels, can't keep up with demand. The shortage is real enough that production is being held up — not a prototype problem, a manufacturing problem.
Folding a screen requires a lot of memory, more than a flat one. The display controller needs its own chunk to manage the layers, and the main SoC needs extra to handle the stress of a screen that bends.
Amazon has enough satellites up to launch its own LEO service this year
Amazon has now launched enough small satellites into low-Earth orbit to actually run its own satellite internet service — no partner needed. The constellation is big enough to cover the planet, and the company plans to turn the switch on sometime this year.
This is not a one-off test. It's a full network, the kind you'd use to connect a remote farm or a container at sea. Amazon has been building it for years, mostly quietly. The satellites talk to each other, route around failures, and talk to ground stations across the world. When the service flips on, it'll compete directly with Starlink and the other LEO players.
Why this matters for us: la gente in places with shaky broadband — the pueblos, the ranchos, the coastal towns — gets a real alternative. Amazon's network means more options, more competition, and hopefully less dependence on whatever the local telco is charging.
The 50 richest creators in 2026, ranked by Visual Capitalist
Visual Capitalist put together a list of the 50 highest-paid content creators in 2026. It's the kind of number you expect to see — the top spots are the usual suspects: MrBeast, MrBeast Gaming, MrBeast Shorts, the trio of creators with the same name but different channels, and a few others you've seen at the top of every ranking since 2022. The rankings move only slightly from last year; the real story is how concentrated the money is. A handful of channels now pull in more than entire media companies.
What matters is the structure behind the numbers. The list separates YouTube from Twitch, TikTok, and podcasters.
Google is feeding everyone the same AI slop and nobody noticed
Maria Loseva posted one metric on LinkedIn — a dropped number — and the comments flooded with the same sentence. Everyone was repeating what Google's AI generated for them, not what they actually thought. It's the slop problem: a hundred writers, one model, one voice.
The thing about this isn't that it's bad writing. It's that it's invisible. Your abuela reads it on Facebook and thinks it's real. Your primo reads it on X and retweets it. They don't know the difference. And the more the model gets better at sounding human, the harder it gets to tell.
The real test is whether the metric actually dropped — not whether the sentence sounds right. Because the metric either moved or it didn't.
Why this matters for us: la gente is trusting AI voices now. If we're going to write for them, we need to write like someone who actually lives here, not like a model pretending.