😺 Sam Altman has a vision of "abundant intelligence"

Welcome, humans.

A few weeks ago, we put together a live “vibe bracket” comparing the top open voice models in HuggingFace (here’s the write-up on that live demo btw!).

But here's what nobody's talking about with voice AI: what happens when voice AI is everywhere? Sure, we hear about “ambient tech,” that’s out of sight, working in the background, but what about ambient AI that actually knows when to pipe down or pipe up?

Well, this demo from Attention Labs just blew our minds (you might’ve seen this as the sponsored section in yesterday’s Treats to Try; today, we’re talking it up on our own because it’s actually awesome).

Their on-device engine lets multiple AI assistants listen to group conversations, but only the one being directly addressed actually responds. No wake words, no cameras, just AI that listens and reads the room (well, hears the room) like a human would:

What we like about this: they're solving what one commenter called “the social layer missing from all current LLM apps.” Their system is basically a smart preprocessor that runs entirely on-device (privacy win!) with sub-100ms latency and 97% accuracy, figuring out who's talking and routing conversations to any model you want—ChatGPT, Claude, your custom fine-tuned model, whatever. It's the cocktail party problem solved in practice, working everywhere from noisy restaurants to chaotic office meetings. This feels like one of those quiet platform shifts… keep your eyes on this one.

Here’s what happened in AI today:

• OpenAI announced plans to produce 1-gigawatt datacenters a week.
• OpenAI, Oracle, and SoftBank committed $400B to five new US data centers.
• Alibaba released six new AI tools like content blockers and real-time translators.
• Microsoft discussed a new publisher marketplace for AI content use.

Advertise in The Neuron here

P.S: Join us tomorrow, Thursday at 10am PST/ 12 PM CST for The Neuron LIVE's first-ever voice interview showdown with ChatGPT, Gemini, Grok, and Claude—basically the AI equivalent of a late-night talk show (with maybe a surprise fifth guest…).

Is the era of “Abundant Intelligence” upon us?

DEEP DIVE: OpenAI wants to build a factory that produces a nuclear power plant's worth of AI compute every single week, and they just announced five new data center sites to make it happen.

Sam Altman dropped a new blog post titled “Abundant Intelligence” where he details that OpenAI plans to build infrastructure that can produce 1 gigawatt of AI capacity weekly. That's the equivalent of a nuclear reactor, except it's powering what he calls humanity's future “fundamental right”—access to AI on tap.

The vision got real when hours later, OpenAI, Oracle, and SoftBank also announced five new Stargate data center sites across America:

• Oracle's building massive facilities in Texas, New Mexico, and the Midwest that'll deliver 5.5 gigawatts and create 25,000+ jobs.
• SoftBank's fast-tracking two sites in Ohio and Texas for another 1.5 gigawatts operational within 18 months.
• Combined with their flagship Abilene campus and CoreWeave projects, they've hit $400B in investment commitments and 7 gigawatts of planned capacity.
• This puts them ahead of schedule for their full 10-gigawatt, $500B target by end of 2025.

Meanwhile, the other big news of the day: The team at Alibaba is absolutely locked in, because they just dropped 6 new releases:

• Qwen3Guard automatically blocks unsafe AI content in 119 languages—like having a bouncer for your chatbot (GitHub)
• Qwen Chat Travel Planner builds your entire trip itinerary—just tell it "5 days in Tokyo" and get a complete day-by-day plan
• Qwen3-LiveTranslate-Flash translates face-to-face conversations in real-time—it even reads lips and gestures (blog).
• Qwen3-Coder writes better, safer code and now accepts screenshots—show it a mockup and it'll build it (API, GitHub, HuggingFace)
• Qwen3-VL turns screenshots into working websites and can operate your computer by understanding what's on screen—this one’s BIG. (cloud, API, blog).
• Qwen3-Max is their new flagship model that rivals the best AI models at coding and reasoning (now available on OpenRouter!).

Is there such a thing as shipping too hard? Because TBH, at this point, Alibaba should hire Xzibit as their company spokesman…

FROM OUR PARTNERS

Confluence is the AI-powered collaborative workspace where knowledge becomes impact

Confluence powers teams for their most critical tasks. Equipped with intelligent AI agents to save time on tedious tasks, brainstorm new ideas, and help you quickly find the answers you need, Confluence is where knowledge becomes impact. Versatile content types like pages, live docs, whiteboards, and databases help teams move seamlessly from ideation to execution.

Confluence integrates seamlessly with the Atlassian suite of products like Jira, Trello, and Loom, and allows for endless customization with apps from our Marketplace.

Start for free. No credit card needed.

Context engineering, explained by Manus!

If you haven’t tried it before, Manus is a superagent that can do a ton of stuff. Way back in July, they published this brilliant blog post about how they do “context engineering” (where you give the agent all the context it needs to succeed). Here are some of the key insights from this blog if you plan to build your own agents:

• The money lesson: Everything revolves around that KV-cache hit rate. Think of it like browser caching; the more you can reuse previous computations, the faster and cheaper everything gets. Small changes like adding timestamps can kill your cache and 10x your costs.
• The memory trick: Instead of cramming everything into context (expensive and slow), teach your agent to use files as external brain storage. It's like giving your AI a notepad it can actually use.
• The attention hack: Agents get distracted in long conversations, just like humans. Manus’ todo list trick is brilliant: constantly reminding the model what it's supposed to be doing keeps it on track.
• The failure philosophy: Most people hide errors from AI. Manus does the opposite. Let your agent see its mistakes, and it'll get smarter about avoiding them.

Treats to Try.

*Asterisk = from our partners. Advertise in The Neuron here.

1. *Typing wastes time. Flow makes voice the faster, smarter option everywhere. Flow users save an average of 5–7 hours every week — that’s full workdays back every month. Give your hands a break ➜ start flowing for free today.
2. Howie lives in your email and schedules your meetings like a personal executive assistant, coordinating with others to find meeting times and handling all the back-and-forth (raised $6M).
3. Quo handles all your business calls and texts in one place, with an AI receptionist that answers when you're busy or asleep (raised $105M).
4. Inkeep lets you build AI agents that handle your customer support tickets, answer internal team questions, and write content for you, all without needing to code (raised $13M);
   1. P.S: this uses TypeScript language, so you can use it with the tool below for the backend.
5. Convex lets you write your entire backend (database, auth, APIs) in TypeScript and automatically syncs changes across all your users in real-time—when one user updates something, everyone else sees it instantly
   1. P.S: Ray Fernando likes Convex for the backend (instead of Supabase), and he says its good because its easy to use and built by former Dropbox people, who he’s met and thinks are “cracked”, which is Silicon Valley for very good sir yes indeed.
6. Google Mixboard turns your text ideas into visual mood boards; just type “cozy living room” and it generates images you can mix, match, and edit with simple commands.

Around the Horn.

This is fun! Try it here.

• Microsoft opened discussions with publishers to build a two-sided marketplace (called the Publisher Content Marketplace) that would compensate them when their content is used in Copilot (and eventually, other platforms).
• Google now lets Android users talk or text with their images in Google Photos to edit them.
• Meta now has a Super PAC (political lobbying group) dedicated to fighting AI regulation at the state level, as many states prepare their own AI regulations.
• Auterion is building Nemyx, a single software standard to enable fully coordinated autonomous drone swarms, and just raised $130M to do it.
• Check out our latest intelligent insights on the website for more insights!

FROM OUR PARTNERS

Become a Trusted AI Consultant

The AI consulting market is about to grow by a factor of 8X – from $6.9 billion today to $54.7 billion in 2032. But how do you turn your AI enthusiasm into marketable skills, clear services and a serious business?

Our friends at Innovating with AI have trained 1,000+ AI consultants – and their exclusive consulting directory has driven Fortune 500 leads to graduates.

Enrollment in The AI Consultancy Project is opening soon – and you’ll only hear about it if you apply for access now.

Click here to request access to The AI Consultancy Project

A Cat’s Commentary.