The Big Story

OpenAI just released new voices for its Voice feature in ChatGPT, Codex, and dots. The update brings better accents and fewer awkward “checking...” moments while speaking in French, German, Hindi, Mexican Spanish, Korean, and Japanese. Think of it like upgrading from an old robot voice on a toy to a friend who actually sounds like they’re from the country whose language they’re using. The voices now flow more naturally, so conversations feel smoother whether you’re practicing a language, brainstorming ideas out loud, or just asking quick questions while walking the dog. For students this means better help with homework or language practice without the voice breaking the flow. For anyone who uses voice mode on their phone, it turns AI into something you might actually want to talk to for longer periods instead of typing. The change is rolling out to all users right now, so open the ChatGPT app on your phone, tap the voice icon, and try switching languages or asking it to tell a short story in one of the new voices. You’ll notice the difference immediately in how little it pauses to “think.” The new voices also appear in Codex and the dot agents, giving the same smoother experience when you ask those tools to handle tasks by voice. OpenAI noted the update is already live for everyone and will reach the API soon, so developers can bring the improved speech into their own apps. This matters because voice is often the easiest way for younger users or people on the go to interact with AI without staring at a screen the whole time. Source: x.com

Explain Like I'm 14

You know how your phone’s keyboard guesses the next word while you’re texting, sometimes finishing the whole sentence before you finish typing? Now picture that same idea but for sound instead of text. The AI listens to the tiny pieces of audio it has already made, then predicts what the next tiny slice of sound should be so the voice stays smooth and natural. It also learns patterns from real people speaking those languages, like how French rises at the end of questions or how Korean has specific rhythms, so the guesses get better and better. When the system used to pause with “checking…”, it was basically saying “wait, let me double-check my guess,” but the new version has practiced enough that it rarely needs that pause anymore. The result is a voice that feels like it’s actually thinking and responding instead of stitching together pre-recorded clips. Because the model now handles six new languages with the same improved prediction, the same smooth flow works whether you’re practicing Spanish homework or asking for a Korean recipe while cooking. Next time you hear an AI speak without stumbling, remember it’s just doing an extremely fast version of autocomplete, but with sound waves instead of letters.

Cool Stuff & Try This

Build tiny games and share them instantly on social Bluesky’s Attie AI tool now lets you describe a simple game, poll, or dashboard in plain words and it builds the whole thing for you. You can then share it as its own little website that friends can open and play right away. It’s perfect if you’ve ever wanted to make a quick quiz or mini game without learning any code. The tool started as a way to create custom feeds and has now expanded so anyone can turn a short description into a working social app like a two-player game or a shared poll. Go to Bluesky, find the Attie AI feature, and type something like “make a two-player rock-paper-scissors game with funny animal characters.” In minutes you’ll have something you can post and play with others. Because the apps run as their own small websites, your friends don’t even need a Bluesky account to try them. This is especially handy for students who want to create quick classroom polls or fun challenges for a group chat without any setup cost. Source: theverge.com

Shop across the whole internet from one cart Google’s new Universal Cart tool lets you add items from different stores while you’re searching or chatting with Gemini, then check out everything in one place. It works right inside Google Search and the Gemini app. The feature appears automatically when you search for products, so you can keep browsing without opening dozens of tabs. Open the Gemini app on your phone, search for something you want to buy, and look for the cart option that appears. It’s a simple way to see how AI can keep track of shopping across many sites without you juggling tabs. You can add a pair of sneakers from one store and a phone case from another, then review the combined total before paying. The tool is already available across the United States, so anyone with the Gemini app can test it today by searching for everyday items like headphones or snacks. Source: techcrunch.com

Quick Bits

Anthropic now lets outside safety experts look at its models like employees do The company gave a nonprofit called METR the same deep access its own team gets so they can check for risks and publish their own findings. This is a step toward having independent watchdogs who can actually see what’s happening inside powerful AI systems. METR will release its own reports on any issues it finds, and Anthropic has agreed not to block those publications. The move follows earlier work between the two groups and includes similar access talks with other organizations like SecureBio and Hugging Face. Source: anthropic.com

YouTube Music will now label AI-generated songs The platform is adding clear tags so listeners know when a track was made by AI instead of a human singer. The labels are already appearing and will soon show up on regular YouTube too. This helps listeners quickly spot which songs were created with AI tools instead of traditional recording sessions. Source: arstechnica.com

💬 Reply to this email — Patrick reads every one.

Share: X · LinkedIn · WhatsApp

Forwarded this email? Subscribe here — it's free.