
Welcome back to the Neural Net! Let’s jump right into the latest and greatest in AI.
In today’s edition: AI weather forecasting quietly outperforms the best physics-based models, Meta considers going closed source after Behemoth’s poor performance, Google and Cognition dismantle Windsurf, and more.
▼
The Street

note: stock data as of last market close
▼
🔥The Hottest AI You Haven’t Heard Of: Weather Forecasting

While everyone’s busy arguing over chatbot IQ’s, one of AI’s most useful applications is quietly floating above our heads, literally.
Meet WindBorne, a startup that’s reinventing weather forecasting with a mix of helium balloons and deep learning. Their tech might sound like a science fair project (long-duration balloons drifting through the upper atmosphere) but it’s producing serious results.
WindBorne’s AI-powered forecasts recently outperformed Europe’s gold-standard weather model by 37% on next-day temperatures. That same system flagged last month’s East Coast heatwave 15 days in advance, beating rivals by up to four days.
How? The secret is twofold:
More data — Their balloons stay aloft longer and gather hard-to-get atmospheric readings.
AI-powered models — They feed that data into deep learning systems that learn patterns instead of simulating physics directly.
As WindBorne’s head of A.I. put it:
“When you do a physics-based model, you’re being smart. When you do a deep learning model, you don’t have to be smart. We’re not writing down the physics. We’re just having it learn.”
⚡ Deep Learning Doesn’t Need the Rules to Win
This quote gets to the core of what makes AI such a game-changer for any type of forecasting. Traditional models are built on known equations — you handcraft the logic, and the model follows it.
Deep learning, on the other hand, skips the human-defined equations. It takes in enough data that it begins to infer the relationships on its own. Here’s how deep learning stacks up against traditional forecasting models:
🤖 Deep Learning | 📐 Traditional Models |
|---|---|
Learns patterns directly from data | Based on physics and hand-crafted equations |
Higher potential accuracy with enough data | Reliable, but limited by known assumptions |
Fast to run once trained | Slower and compute-intensive |
Can be brittle or hard to interpret | Transparent and explainable |
Doesn’t show its work | Traceable logic and outputs |
In summary: AI can spot patterns we’d never notice, even if we stared at the clouds all day. But the models poised to take our jobs come with one major flaw: they can’t explain how they do theirs.
But hey — at least this means we won’t need to rely on Apple Weather forever. If WindBorne says there’s a storm coming, you might actually believe it.
▼
💡How To AI: Claude Takes Small Step Towards Ambient AI

One of the biggest limitations of current Generative AI is that it starts every conversation like it's your first. You have to re-explain what you're doing:
"I'm working on a product launch."
"Here's the spreadsheet with our metrics."
"This column is revenue, this one's churn."
Then you paste in the data, attach files, and spell out the format like you're onboarding a new intern.
➡️ Why the copy-paste era might be changing: Claude gets context.
Claude just released a new set of tool integrations that allow it to pull info directly from many of the tools you already use. It now supports direct connections to tools like Notion, Canva, Stripe, Figma, Jira, Confluence, and others.
With this update, Claude becomes more like a teammate that lives inside your tools, not a chatbot that lives in your browser tab. It’s a small step toward ambient AI: something that quietly supports you in the background.
Read how to get started here.
▼
In Partnership With SelectSoftware
Get a List of the Best HRIS Software for Your Company
Stop wasting time on endless research and confusing options.
Our HR Software experts provide you with tailored recommendations from our database of 1,000+ vendors across HRIS, ATS, Payroll, and HCM.
✅ 15 minutes vs. hours of demos
✅ 1:1 help from an HR Software expert
✅ No spam, no sales pressure
▼
Heard in the Server Room
Meta’s new Superintelligence Lab is reportedly considering ditching its open-source AI model, Behemoth, for a closed version—a major pivot from its famously open-source roots. Behemoth has been fully trained but underwhelmed in internal tests, leading teams to halt further work. The shift would mark a big philosophical change for Meta, known for making its AI tech public. While discussions are still early, the ultimate call will of course rest with Zuckerberg.
The Velvet Sundown, a retro folk-rock band that seemingly came out of nowhere, racked up 1.2M monthly Spotify listeners in June with its hit Dust on the Wind. But plot twist: the band isn’t real. Every song, member, backstory, and album cover was generated by AI. Billed as an “artistic provocation,” the project claims to blur the line between human and machine-made music. Or it’s just big tech harshing our vibe, man.
OpenAI just scored a $200M contract with the U.S. Department of Defense to prototype next-gen AI tools. It’s the first move under the company’s new “OpenAI for Government” initiative, which aims to put secure AI in the hands of public agencies, from streamlining paperwork to boosting cyber defense. The one-year pilot builds on existing partnerships with NASA, NIH, and the Air Force, and signals a growing government embrace of generative AI.
▼
🦈 Windsurf Feeding Frenzy Shows Brutal Tactics In AI Race

In a stunning plot twist in the AI coding arms race, Google DeepMind pulled off a $2.4 billion reverse-acquihire, scooping up Windsurf’s CEO Varun Mohan, co-founder Douglas Chen, and top researchers—mere hours after OpenAI’s $3 billion acquisition offer for Windsurf expired.
While Google doesn't own a stake in Windsurf, it does hold a nonexclusive license to key tech and, perhaps more importantly, access to elite AI coding talent. It's a familiar Big Tech tactic to bypass regulatory heat: grab the brains and license the IP. But the move left most of Windsurf’s 250-person team behind.
But just days later, AI agent startup Cognition swooped in, acquiring Windsurf’s remaining staff, product, and IP. The deal includes full employee payouts and plans to integrate Windsurf’s IDE with Cognition’s own AI coding agent, Devin.
Why it matters: The scramble for Windsurf shows how high the stakes are in the AI coding wars. As Big Tech races to control the developer experience, talent has become the hottest commodity. With this move, Google arms itself with elite coders. While Meta has dominated the news recently, this is a reminder that the other big players aren’t going quietly into that good night.
▼
That’s it for this week! Have a great week, and we’ll catch you next time with more neural nuggets.

