The two most advanced AI models on the market, a voice mode that finally lets you think, and a US state that hit pause.
Inside a month: Anthropic and OpenAI each released the most capable model they've ever made public, OpenAI rebuilt how you talk to ChatGPT out loud, and New York became the first American state to halt construction of new data centres.
That pace is exactly why this series exists.
What this is
TechCurious This Week is a short weekly briefing on what actually happened in AI — what it means for your life, and what, if anything, to do about it. Not every announcement, and never the ones designed to make you feel behind. Three stories, decoded, with my honest read. The standing deal: if something matters, I'll tell you why. If it doesn't, I'll tell you that too — some weeks the most useful thing I can say is ignore this one.
1. The two most advanced models ever released — and what happens if you try one
Two releases, both at the very top of the range. Anthropic shipped Claude Fable 5, the most capable model it has ever made public. OpenAI shipped GPT-5.6 — really three models: Sol (the flagship), Terra (the everyday one), Luna (the fast, cheap one). The word you'll see everywhere is frontier model. It just means the most capable model a company has built to date — working at the very front.
Here's something worth trying. Next time you have a genuinely tangled problem — the kind you'd block out an afternoon for — switch to the more powerful model (the model picker sits near the bottom right of the box where you type) and run the same question through both. Does it feel smarter to you? Your answer counts for more than any benchmark chart. And even if you never touch the picker, you still win: the model you use for drafting emails and planning trips is more capable this month than it was last.
Amanda's take: You don't need to remember a single one of these names. When the top models jump, everything underneath jumps with them — the same chat window you used last month quietly got smarter. The upgrade already happened to you. All that's left is to give it something harder to do.
2. ChatGPT's voice mode was rebuilt so you can pause and think
On July 8, OpenAI launched GPT-Live, a new generation of voice models now powering ChatGPT Voice. The old voice mode worked in turns: you spoke, it waited for silence, then decided you were finished — so every time you paused mid-sentence to find the right word, it jumped in. The new one listens and speaks at the same time. You can interrupt it. You can ask it to stay quiet and just listen. It handles pauses, does live translation, and quietly hands harder questions to a bigger model without breaking the conversation.
🎬 PAUSE — honestly, watch the launch video. I posted part of it on IG — I get a kick out of the "mature" actors they use. Watch it.
Why this matters: plenty of us work things out by talking — it's why we call a friend before the decision, not after. A tool that interrupts every silence is useless for that. A tool that waits (even feels intuitive) is something else entirely. To find it: the small microphone in ChatGPT is just dictation; the button beside it in the bottom-right corner — the little row of sound-wave bars — is the conversation. Look for the Live label once you're in (no voice button at all? update the app). Then just talk. When it wanders off, "stop, go back" works exactly the way it would with a person.
Amanda's take: This could be the moment voice AI stops feeling like an automated phone menu and starts feeling like a patient thinking partner. I'm not declaring it solved — but it's the first version I've wanted to use on a walk or in the car.
3. New York hit pause on data centres. Hamilton chose to keep building.
On July 14, Governor Kathy Hochul made New York the first US state to pause approvals for new large data centres — 50 megawatts and up, for one year, after residential electricity prices climbed nearly 68% since 2019. The next morning, an hour down the road from my house, Hamilton council voted 10–6 the other way, declining what would have been Canada's first pause — councillors pointed to badly needed investment in a city hit hard by struggling steel. Mississauga and Vancouver debate their own versions this month. And this isn't a distant story: Ontario's grid operator says data-centre projects totalling roughly 7,000 megawatts — enough for about six million homes — have active requests to connect. That figure doubled in months.
Here's what I keep running into. Go looking for evidence that data centres damage local environments and you'll find it — peer-reviewed, credible, alarming. Go looking for evidence the water panic is overstated and you'll find that too — also peer-reviewed, also credible. Both piles are real. They're measuring different things, at different scales, in different places — and each side reaches for the pile that suits it. Two things can be true at once: your own AI use is a rounding error (I ran those numbers in What We Should Actually Be Worried About with AI — an average prompt works out to about five seconds of Netflix), and the strain near one particular building in one particular town can be severe.
The part we should all be able to agree on: most of this isn't disclosed. A peer-reviewed paper this February found data centres are among the least transparent industrial water users in the US, and Ontario has its own version of the gap — the grid operator can say 7,000 megawatts, but not how many projects or where. You can't have an informed argument about numbers nobody publishes. Transparency isn't a partisan left or right position; it's what has to come first for anyone's argument to hold — including the one you already agree with. So if a project lands near you, get specific: which water source, and how much? What happens to rates? What did the company commit to in writing? Either those questions have answers, or the silence tells you something useful. That's the level where a person is actually effective — and it's the kind of question we're already good at asking.
Amanda's take: The loudest voices on both sides need this to be simple, and it isn't. I won't tell you the Hamilton vote was right or wrong — anyone certain after a week isn't reading closely. But staying out of the argument doesn't keep you neutral. It just means the decision gets made without you.
Consider it decoded.
Written by Amanda · Midlife TechCurious
Sources: Anthropic, Claude Fable 5 launch (Jun 2026) · OpenAI, GPT-5.6 (Jul 9) and Introducing GPT-Live (Jul 8) · Artificial Analysis Intelligence Index (Jul 2026) · TechCrunch & CNBC on the New York moratorium (Jul 14–15) · CBC News on the Hamilton vote (Jul 15) · Privette et al., Data Centers Water Footprint, AGU Advances (Feb 2026) · IESO 2026 Annual Planning Outlook · Pew Research via TechCrunch (Jul 2026).