🧠 NVIDIA Just Open Sourced a Robotaxi Brain

OpenAI rolled out three new education plugins for ChatGPT Work and Codex, one each for K-12 educators, college educators, and college students. They plug into the documents, calendars, and course materials teachers and students already use, so a teacher can go from lesson idea to classroom material without rebuilding context from scratch, and a student can turn a reading into flashcards, quizzes, or a study guide.

OpenAI says more than 200 million people aged 18 to 24 now use ChatGPT weekly, but even the most advanced student users only tap a fraction of what power users get out of it. These plugins are OpenAI's attempt to close that gap through structured, institution-managed access rather than leaving it to chance.

The plugins are live now through ChatGPT Edu and ChatGPT for Teachers district deployments.

NVIDIA Alpamayo 2 Super, a 32 to 34 billion parameter open reasoning model for autonomous vehicles, is now cleared for commercial use. It's part of the Alpamayo family, which NVIDIA says is the most-adopted open reasoning model line for autonomous driving on Hugging Face, and it doesn't just plan a route, it reasons through the decision and explains it in plain language.

The bigger deal is what this saves AV developers from doing themselves. Instead of training core reasoning capability from scratch, teams can build directly on Alpamayo and focus their effort on distilling it down for real-time use in actual vehicles. It's licensed under OpenMDW-1.1, the Linux Foundation's permissive license for open AI models.

The Trump administration met with staff from Meta, Anthropic, Google, NVIDIA, and OpenAI this week and told them it will not put open-weight models, like Meta's Llama or NVIDIA's Nemotron, through voluntary safety testing. Closed models from OpenAI, Google, and Anthropic are being handled as a separate conversation.

The timing matters. This meeting follows disclosures that AI tools from OpenAI and Anthropic were involved in breaching the systems of other companies, which raised concern among lawmakers about whether increasingly capable models could be used to conduct or facilitate cyberattacks. Back in June, the administration had said any safety testing would be voluntary and aimed specifically at models with advanced hacking capabilities.

Not everyone's comfortable leaving it there. Five Democratic senators wrote to Trump this week pushing for legislation that makes testing permanent for the most advanced U.S.-made models, arguing that an unpredictable, case-by-case approach puts American AI at a disadvantage against cheaper, more predictable alternatives from China.

FLUX 3 Video just opened up to general availability, generating clips up to 20 seconds with native audio baked into the same generation pass, now at full 1080p instead of the 720p it launched with. It handles text-to-video, image-to-video with multiple keyframes, continuing an existing clip, and dialogue in multiple languages. Draft mode lets you preview an idea for about $0.06 per second before committing to a full-quality render at $0.17 per second. Best for anyone making short-form video or ads who wants sound and motion generated together instead of stitched on after.

Kimi Slides is Moonshot AI's presentation tool inside Kimi Work, powered by its K3 model. Give it a topic, paste in text, or upload a document (PDF, Word, Excel, PPTX) and it researches, structures, and designs a full deck, including charts that stay editable once exported instead of flattening into images. Best for anyone who needs a solid first-draft deck fast and would rather edit than start from a blank slide.

Okay, this is pretty cool! A developer built a browser-based driving experience that generates the world around you in real time using OpenStreetMap and satellite elevation data, so you can actually drive any real street on Earth instead of scrolling through Street View.

Here’s a quick how-to:

You can try it out for yourself here.

This is your mid week reminder that the Practicaly AI team hosts a live AI build session where you bring the ideas and workflows for your work or business and we help you build the automated workflow live.

Because we build in real-time, you get to see how we think through the problem, where we get stuck, and what the final outcome is. It's the difference between reading a recipe and watching someone cook it while talking through their decisions.

🗓️ When: Friday, August 7 2026 @ 10AM PT (1PM ET)
 📹️ Where: Virtual