🧠 Claude's Voice Grows a Brain

Since launch, Claude's voice mode ran on Haiku, the fast but lightweight model built for quick back and forth. That was fine for simple questions, useless for anything that needed real thinking. Anthropic just changed that.

Voice mode now runs on Sonnet and Opus too, and you can switch between all three mid-conversation from the model picker. It also reaches tools you've already connected, like Gmail and Slack, so you can ask for an inbox catch-up or a calendar check without touching your keyboard. Language support got a lot wider as well, so you can think through a problem in whatever language actually feels natural to you.

This is rolling out in beta across mobile, desktop, and web, though Anthropic says it works best from your phone.

Health in ChatGPT is starting to roll out to U.S. users 18 and up, across every plan including free. Open it from the sidebar and you can connect Apple Health or supported medical records, then browse a timeline of your synced data and recent trends.

With your permission, ChatGPT can pull that context into any conversation, comparing a new lab result to an old one, tracking changes since your last appointment, or factoring in a recent injury before suggesting a workout. OpenAI says none of this connected data trains its models or feeds ads. More on this here.

Black Forest Labs just introduced FLUX 3, a single model trained jointly on image, video, audio, and action prediction. The pitch is that generating a realistic next video frame and predicting a robot's next physical move are basically the same problem wearing different outfits.

FLUX 3 Video is live in early access now, and BFL is already using the same backbone with robotics partner mimic and Audi to predict robot actions, not just pixels. Full image and open-weight versions are expected to roll out over the coming months.

If this holds up, it's a preview of AI models that don't just create content, they understand and predict the physical world too. Read more here.

Creator Peter Yang open sourced a free Claude skill called /no-ai-slop that strips out the tells everyone's sick of reading: fake profound endings, "it's not X, it's Y" contrasts, throat clearing openers. You copy the GitHub link, ask Claude or ChatGPT to install it, and it edits your draft while explaining exactly what it changed.

It's been tested against the popular "humanizer" skill from a few months back and reportedly catches more of the obvious patterns. If your writing still reads a little robotic, this is worth five minutes. Watch a quick demo here.

Head to the ChatGPT app, open Plugins, and install the PDF plugin. Once it's in, you can upload a stack of forms, think rental applications, intake paperwork, anything repetitive, and ChatGPT fills them out for you instead of you doing it by hand.

If you've ever lost an evening to identical forms with slightly different formatting, this one earns its spot on your phone. See how this looks here.

If you like the idea of Wispr Flow but not the subscription, OpenWhispr is a free, open source dictation app that does the same core job: press a hotkey, talk, and your words get typed into whatever app you're using. It runs locally, so your voice never has to leave your machine.

We’ll talk through how to choose the right model for the job, where token usage tends to creep up, and how to get better results without defaulting to the most expensive option every time.

🗓️ When: Friday, July 24th @ 10AM PT (1PM ET)
 📹️ Where: Virtual