Hey there, Tech Trailblazers! 🤖✨

Quick pulse check before we dive in. On Nov 6, 12–4 PM ET, Section is hosting a free “AI Builder in a Day” with hands-on micro-workshops and a certificate you can add to your profile. Claim your free spot if you want practical frameworks you can use immediately.

Inside this issue: an all-AI research conference that tests agents against human peer review, Nvidia’s desk-friendly “personal AI supercomputer,” and Windows Copilot upgrades that let you talk to your PC and let it “see” your screen to help.

📰 Upcoming in this issue

• 🤖 AI Agents Write and Review a Whole Conference

• 🤖 Nvidia’s Tiny “Personal AI Supercomputer” Lands on Desktops

• 🖥️ Windows Copilot gets Voice and Vision

📈 Trending news

• Veo 3.1 in Gemini and Vertex AI

• Why the AI Economy Isn’t a ’90s Replay

• Why AI Startups Are Owning Their Data

🤖 AI Agents Write and Review a Whole Conference

Nature reports on Agents4Science 2025, an experiment where every paper and every peer review comes from AI. Humans attend, but the work under discussion is generated and evaluated by model “agents,” then compared with human standards.

Read the full article here »

Key Takeaways:

• 🧪 A First-of-Its-Kind Trial: The online event debuts 22 October 2025 with 100% AI-written submissions and AI-written reviews, set up to test quality head-to-head with humans.

• 🧠 From Tools to “Agents”: Organizers say the shift is from single-task LLMs to coordinated agents that act like research teams across the pipeline.

• 🧰 A Safe Sandbox: The conference is designed to try new submission and review processes without risking standard venues, then share what works.

• 📏 What Will Be Measured: Reviews by models will be compared with human peer review to gauge rigor, bias, and usefulness before wider adoption.

🤖 Nvidia’s Tiny “Personal AI Supercomputer” Lands on Desktops

Nvidia’s DGX Spark is a $3,999 mini-PC that delivers up to 1 petaFLOP of AI performance in a small, 240W box with 128GB of unified memory. It can run inference on models up to 200B parameters locally and fine-tune mid-size models without the cloud.

Read the full article here »

Key Takeaways:

• 🧠 Big Models, Small Box: 128GB unified memory lets developers run far larger models than typical consumer GPUs can handle.

• ⚙️ Grace–Blackwell Inside: The GB10 superchip pairs a Grace CPU with a Blackwell GPU, tuned for local AI dev and inference.

• 🔌 Desk-Friendly Power: About 240W, standard outlet, DGX OS with the full CUDA stack for plug-and-build workflows.

• 🛒 Broad Availability: Orders begin mid-October through Nvidia and partners like Dell, Lenovo, HP, and others.

🖥️ Windows Copilot gets Voice and Vision

Microsoft is baking AI deep into Windows 11 so you can literally talk to your PC and let Copilot see your screen to help, explain, and even take actions you approve. Think wake word control, on-screen guidance, and task automation that turns your desktop into a true assistant.

Read the full article here »

Key Takeaways:

• 🎙️ “Hey Copilot” arrives: Voice becomes a first-class input so you can start conversations hands-free on Windows 11. Opt in under Settings.

• 👓 Copilot Vision: With permission, Copilot can view your screen and coach you through tasks or generate context-aware help.

• 🛠️ Copilot Actions: Microsoft is testing agent-style actions like making reservations or simple edits on your PC, limited to resources you authorize.

• 🔐 Privacy controls: Features are opt-in, with explicit mic and screen-share permissions and easy off switches.

Why It Matters

AI is shifting from single tools to orchestrated agents, from cloud-only to powerful local hardware, and from app-by-app add-ons to OS-level assistants.

Your next steps are simple: try one agent workflow, prototype something locally, and audit your privacy settings before you enable vision or voice features.

Until our next issue,

Samantha Vale
Editor-in-Chief
Get Nerdy With AI

P.S. Interested in sponsoring a future issue? Just reply to this email and I’ll send packages!