Welcome, humans.

You gotta kinda feel bad for Sam and company at OpenAI. They launched Operator a week ago, and TBH, the DeepSeek news totally buried it. And that’s us saying that—our entire job is to read and watch AI stuff. We tried—we’ve barely seen anything on it.  

Chief Product Officer Kevin Weil did his best to pivot the convo back to Operator while the company was demoing their new models in DC (more on that below), but the media pivoted back to DeepSeek almost as fast as they pounced on Martha and her salad.

On top of that, the memes reacting to OpenAI’s claims that DeepSeek stole their data are completely out of control. Here’s what we mean:

There’s also this. And this. Oh, and this. They just keep coming!

All that said, according to OpenAI’s VP of Global Policy, GPT o3 (their new reasoning model) should come out today, so be on the lookout for that—we expect an announcement will post here—if it happens. This is OpenAI we’re talking about…

Here’s what you need to know about AI today:

• Open signed more government deals while other AI CEOs got in a scrap.

• Microsoft made Think Deeper free for Copilot.

• Google pulled Gemini 2.0 Pro after brief release.

• The Authors Guild launched a human writing cert.

OpenAI may about to be worth $300B+…but first it has some science to do.

The WSJ just reported that OpenAI could raise up to $40B in a new round that would value it at $300B+. Even wilder? They're raising this money while losing a lot of other money. SoftBank, the main investor in the deal, doesn't seem worried though; they're reportedly ready to pour in $15-25B.

Meanwhile, OpenAI just inked a major agreement with the US National Laboratories that would give their 15,000 scientists access to o1 (and o3). OpenAI plans to deploy these models on Venado, a beefy NVIDIA supercomputer at Los Alamos.

If you're wondering “why should I care?”—well, imagine having 15,000 of the world's smartest scientists using ChatGPT-on-steroids to solve some of the world’s hardest problems:

• Developing new treatments, disease prevention, and biological research.

• Innovating on power grid security, renewable tech, and materials science.

• Experimenting with physics, astrophysics, and advanced mathematics.

• Oh yeah, and working on nuclear weapons research (safety, of course).

In the background of all this, there’s a heated debate happening between two other AI CEOs: should AI be more like Linux (open to everyone) or iOS (locked down and controlled)?

Closed source folks (quarterbacked by Anthropic CEO Dario Amodei) be like:

• We need to control who gets powerful AI.

• National security matters.

• Government partnerships require trust.

• Export controls keep ‘Murica competitive.

Amodei wrote a whole piece on this where he called out DeepSeek for actually spending $1B on 50,000 AI chips to build its models, not the ~$6M some claimed, and argued these hidden chips demonstrate why the US needs to prevent China from acquiring millions more.

Hugging Face's CEO Thomas Wolf, on the other side, disagrees. His rebuttal? “Ok boomer” (not his actual words, but that energy).

He pointed out that DeepSeek published everything openly, and now teams everywhere are building on it:

• Allen AI in Seattle just released Tülu 3 (supposedly better than DeepSeek-V3).

• Mistral in Paris launched Small3.

• And hundreds of other developers are making their own versions.

Open source gang says AI belongs to everyone, more eyes = safer code, shared costs = faster progress, and no single company—or country—controls AI.

Our take: Government projects and defense work will stay closed (like OpenAI's new deal)—and that $40B will help Sam seal the deal. Just maybe change the name, Sam?

But open source will keep pushing commercial AI forward, like how Linux powers most of the internet…and drive down AI prices for the rest of us.

FROM OUR PARTNERS

Introducing Postman AI Agent Builder

Say hello to a game-changing way to build AI agents. Postman’s new AI Agent Builder is the ultimate toolkit for developers creating reliable, high-performing AI Agents.

Discover trusted APIs and models in Postman’s centralized repository. Use the no-code visual canvas to build workflows and transform APIs into agent-ready code.

Test locally for accuracy and reliability, and monitor metrics like response times and success rates.

Ensure data privacy by deploying agents on your own infrastructure.

Postman gives you everything you need to design, test and launch AI Agents—all in one platform.

Start building

DeepSeek for absolute beginners

Corbin Brown is one of our favorite AI Youtube creators, and this tutorial makes using DeepSeek extremely easy—he even has a second video where he explains how to run it locally (and safely) on your computer here.

Treats To Try.

1. Google Labs actually has a lot of tools in its Experiments page now—like tools that write your code, edit your photos, create videos, and make music.

2. SuperOps streamlines your IT management by combining ticketing, monitoring, and security tools into one platform (raised $25M).

3. Qwen 2.5, Alibaba’s rival to DeepSeek, has a chatbot you can use to try it out.

4. LM Studio is our preferred tool to download and run local AI models on your computer.

5. SimpleBench is an AI benchmark that measures how well AI models handle common-sense questions that any high schooler can easily answer—try here.

6. We don’t want anyone in trouble with Congress…but for everyone who still has TikTok, if you @CookTok in the comments of a food video, you get the recipe…

7. Someone made Seefood from the Silicon Valley TV show, which lets you find out if something is hot dog or not hot dog…

See our top 51 AI Tools for Business here!

Around the Horn.

• Microsoft made Think Deeper, its version of the reasoning model o1, free for all Copilot users.

• Google quietly released, and then quietly unreleased, its new Gemini 2.0 Pro Experimental—probably because they thought o3 would go live yesterday.

• Meta apparently sold 1M Ray-Ban smart glasses last year, and could release a new pair with a full on heads-up display called Hypernova for later this year.

• The Authors Guild now has a “Human Authored” certificate to let authors certify that an actual human wrote their book and help combat AI slop ebooks.

• IBM plans to cut $3.5B in 2025 through job cuts and redirect spending toward AI initiatives to grow AI revenue to $5B.

• ChatGPT got a memory upgrade through June 2024, it can now better understand spatial relationships in images, and there’s supposed to be a new think button, though users said it disappeared.

FROM OUR PARTNERS

Mastering AI Agents…

Learn how to create powerful, reliable AI agents with Galileo’s in-depth eBook! Enjoy 100 pages of expert content on:

• Choosing the right agentic framework for your use case.

• Evaluating and improving AI agent performance.

• Tips for identifying failure points and production issues.

Get your copy now

Intelligent Insights

• China has made a lot of progress in overlapping industries like AI, drones, lidar, robotics, and batteries, which forms a mutually reinforcing feedback loop.  

• In a recent study, Chinese researchers demonstrated that two large language models could create functioning replicas of themselves with 50-90% success rates across multiple trials (full paper).

• In case you’re curious… here’s a round-up of some of Sam’s other bets:

  • Sam Altman’s Retro Biosciences wants to raise a $1B Series A round to use AI to extend the human lifespan, joining other “longevity” upstarts like Altos Labs (raised $2B in 2022 from Jeff Bezos). 

  • OpenAI-backed robotics startup 1X acquired another Norwegian humanoid robotics startups called Kind Humanoid.

  • Sam Altman’s Hellion Fusion startup raised $425M in Series F funding + activated its seventh prototype reactor, Polaris, for the goal of supplying Microsoft with fusion-generated electricity by 2028.

• This is a fantastic read about AI art and the creative process, and why we need computers that “increase the work we do” (when it’s work we WANT to do) and not the other way around. Remember: the struggle is the point, people! 

A Cat's Commentary.