Today's NYT Mini Crossword Answers for Friday, July 25

Forget the massive, energy-hungry AI models we're used to—researchers have developed a new type of AI called Hierarchical Reasoning Models (HRMs) that can solve complex problems 100 times faster than large language models (LLMs) like GPT-4, and it only needs 1,000 training examples to do it. Unlike today's behemoth AIs, HRMs are leaner, quicker, and far more efficient, making them a potential game-changer for real-world applications where speed and data scarcity are issues.

Editor’s Note: This isn't just another incremental upgrade—it's a fundamentally different approach to AI reasoning. If HRMs live up to their promise, they could democratize AI by making powerful reasoning accessible without needing massive datasets or computing power. That means faster, cheaper, and more practical AI tools for everything from healthcare diagnostics to financial modeling. It also hints at a future where AI doesn't have to mean "bigger is better."

If you're stuck on today's Wordle puzzle (#1429), this article offers handy hints and the solution to get you unstuck—plus yesterday's answer if you missed it. Think of it as a friendly nudge when your brain hits a word wall.

Big tech companies and scrappy startups are racing to develop specialized computer chips that could drastically cut the energy appetite of AI systems. Right now, training and running AI models like ChatGPT gobbles up enough electricity to power small cities—these new designs aim to make the process far more efficient without sacrificing performance.

Editor’s Note: AI’s explosive growth comes with a hidden cost: massive energy use. If these chips deliver, they could make AI cheaper, greener, and more accessible—while easing concerns about the tech’s environmental toll. It’s not just about saving watts; it’s about keeping AI’s progress sustainable.

Microsoft is bringing back a bit of that nostalgic Clippy charm with its new Copilot Appearance feature, but this time with a modern AI twist. Instead of a static assistant, Copilot now has a visual, expressive character designed to make interactions feel more personal and engaging—like chatting with a friendly digital sidekick rather than a cold, faceless algorithm.

Editor’s Note: Remember Clippy, the overly eager paperclip from old Microsoft Word? Love it or hate it, that little guy made software feel a bit more human. Now, Microsoft is trying to recapture some of that personality—but smarter and less intrusive—by giving its AI assistant, Copilot, a visual presence. This isn’t just about nostalgia; it’s about making AI tools feel more approachable and relatable, which could change how we interact with them daily. If it works, we might start seeing more AI with a "face"—for better or worse.

Researchers have uncovered a sneaky new hardware vulnerability called "SRAM Has No Chill" that lets attackers steal sensitive data from a chip's memory by exploiting power domain separation—basically tricking parts of the chip into spilling secrets they shouldn’t. Think of it like eavesdropping on a private conversation by messing with the room’s electricity.

Editor’s Note: This isn’t just some theoretical flaw—it’s a real-world weakness in how chips manage power, which could expose everything from encryption keys to personal data. If you’re into tech, it’s a wake-up call for better hardware defenses; if you’re not, it’s a reminder that even the silicon in your devices isn’t always as secure as you’d hope.