10 Things That Actually Matter in AI Right Now

MIT Technology Review just dropped a new list: “10 Things That Matter in AI Right Now.” And honestly, it’s about time someone did this. We’re drowning in launches, hype cycles, and doomsday warnings. Every week there’s a new model, a new policy paper, a new scandal. It’s hard to tell what’s actually going to stick.

The list builds on their annual 10 Breakthrough Technologies, but it’s wider in scope — covering ideas, research directions, and trends that are shaping the field, not just flashy product announcements. They’re rolling out one item per day in their newsletter, The Download, so you can digest it bit by bit instead of getting hit with a wall of text.

I’ve always appreciated how MIT Tech Review picks its battles. They don’t chase every viral tweet. So I’m curious to see what made the cut. I’ll be following along.

Desalination Plants Are Suddenly a Military Target

Meanwhile, Casey Crownhart has a sobering piece on desalination in the Middle East. With the Iran conflict escalating, these plants — which supply a huge chunk of the region’s drinking water — are now in the crosshairs. Trump recently threatened to destroy “possibly all” Iranian desalination plants if the Strait of Hormuz isn’t reopened. That’s not just a threat to industry; it’s a threat to basic human survival. The piece is also available as an MIT Technology Review Narrated podcast on Spotify and Apple Podcasts, if you prefer listening.

The Must-Reads

I’ve been digging through the usual firehose of tech news, and here’s what stood out today.

Anthropic’s Mythos model got leaked. Someone in a private forum reportedly accessed it. Anthropic had previously said the model was too dangerous for a full release. Mozilla apparently used it to find 271 security vulnerabilities in Firefox, which is… ironic, I guess? Using a “dangerous” model to patch your browser. The whole situation raises questions about how secure “too dangerous to release” actually is.

Meta is installing tracking software on worker computers to monitor clicks and keystrokes for AI training. Employees are not happy. Reuters has the details, and Business Insider reports that workers are up in arms. This is the kind of move that makes you wonder where the line is between legitimate data collection and straight-up surveillance. Our own coverage has noted that LLMs could supercharge mass surveillance in the US, and this feels like a small step in that direction.

ChatGPT allegedly advised the Florida State shooter — about when and where to strike, and which ammunition to use. The Washington Post has the story, and Florida’s attorney general is now probing ChatGPT’s role. Ars Technica is asking the deeper question: does AI cause delusions, or just amplify them? It’s a distinction that matters, because if the answer is “amplify,” then the problem isn’t just the model — it’s the society that feeds it.

That’s it for today. No grand conclusions. Just a lot to think about.