The Great American AI Race

We’ve been given a machine that turns energy into intelligence. It’s a remarkable, wild, and fun time. I get asked all the time how the US wins the AI race. If we turn this into a race toward centralization and seeing who can shovel more coal into GPUs, America loses. Central planning isn’t America’s strength. If, however, winning the race relies on people’s ingenuity to put intelligence into every product–then America cannot be beat. American people equipped with AI could lead to the improvement of education, health, defense, and almost every industry making the world a more free and wonderful place for our children.¹

So what can we do to spread American AI to every corner of the solar system? Open source and buying American technology are key elements. Open source quality is catching up to closed frontier models. When I wrote the Linux moment for AI, the closed source was far ahead in reasoning, e.g., a score in the 70s on key benchmarks like MMLU–the open source models felt like they were barely working models, only at 8 in the same benchmark. My colleagues thought I was crazy when I talked about open source benchmarking because there was “one, not very good, model.” Now, the open source is right there in quality–and improving! New form factors are coming fast and furious: audio, video, text, and DNA–on devices big and small. You’ll want intelligence in your phone, your car, but also your next generation medical devices, consumer products, and more.

Open source is not inevitable. As mentioned above, there was a time when it looked like we would centralize AI behind a few early winners. My view is this would cause real harm for the US. Thankfully, I think the debate of whether we should have open source is over. I think open source AI is here to stay. The new question has now come into view:

Should the world’s AI run on American or Chinese rails?

My view: AI should be on American rails–and we (American companies and academics) should make it happen. We’re in a race that’s bigger than the space race. This is not to say we shouldn’t use Chinese models or learn from our friends in China. If they put out open source models, we should use them! ² However, without a parallel open source ecosystem the world’s economy will depend on their models for updates, particular choices baked into the model, and cede American influence over AI products. I owe America so much; I love it. I also love to compete, and I hate to lose. We should feel this way as a country about AI. The US is still the best idea on the planet. Let’s bet on it.

America’s academics and industry have had long and successful partnerships that led to groundbreaking AI. Industry needs to take a long view.

Everyone knows that we have the best frontier model companies like OpenAI, Anthropic, Google, and more. The US invented the models that everyone uses, the transformers, and they are best optimized in the US with things like Flash Attention and Thunderkittens (both from our lab!). The new architectures from China that are “mamba” based also came from our lab, and see Simran’s note about their history. Note these architectures are used in the closed world (even Google!), but our companies are quiet about these ideas–because they view them as competitive advantages or means to promote. The American system has been broken in AI for a long time, which is the subject of a longer rant–but basically academia contributes great ideas, and some are massaged for promotion by bigger industry labs with PR machines. Don’t believe me: then why was the Transformer, the bedrock of modern AI, called “Attention is all you need?” rather than “here is this new thing attention?” It wasn’t new! We already knew what Attention was in this context… Yoshua and Chris Manning’s teams, two awesome academics, had talked about it, and most of us had coded it up! This takes nothing away from Google’s breakthrough and deployment, this is how science works! Their work directly led to OpenAI’s work and breakthroughs, which led to so much amazing stuff!

My view is that we need to work together, we need industry to fund basic science–Americans have picked up so much of the tab, but it’s time. Companies necessarily have a narrow view of how this technology can be used: Google builds a search engine, and OpenAI also builds a search engine–it’s an amazing business. Only recently have we seen how AI transforms how people can build applications with wonderful startups like Cursor or Cognition. Artificial intelligence is big, it will be infused everywhere in our lives. American companies and universities have a role to play to help the world be better and freer by building the foundations of this technology.

My view is that public-private partnerships are a clear path forward. Academia has the long-term vision, basic research, and ability to prevent technical surprise as we explore paths. Industry has brilliant people and can also easily stomach the CapEx spending to explore the many alternatives. This will also help industry identify new sources of talent and new use cases from the enthusiast communities. My bet is that there are many faster and cheaper ways to get to what we call AGI now; we’re just in the first innings. We need open source as a foundation to share ideas and progress concretely among our academic and industrial friends as has been done in so many other areas.

Regulation is hurting our startup and innovation ecosystem.

We should let M&A flourish as sometimes it’s the route to growth of technology. Not only does going to a larger company often give distribution to technology that allows it to benefit more people, the second order effects are large too. I know this first hand. When I sold my first company, I immediately started more, hired more, and invested in more. You want a dynamic ecosystem of rapid growth. People like me who start things aren’t always the people who should take them to scaled markets. Our large AI companies aren’t evil, they are great American companies and the envy of the world… let them flourish.

We should remove barriers to competition and the ability to sell technology to build an American foundation for AI.

The AI Race is a chance for American Academia to redeem itself.

Our great American universities can help lead in this race. Many Americans feel that universities don’t care about them or improve their lives. It shouldn’t be this way. At their core, universities have good, smart people who educate the best and brightest our country has to offer. They make smart new Americans and educate the AI workforce that the US needs.  Universities are great places with experts in a huge number of disciplines–like biology, medicine, physics, and chemistry–that could be changed by AI. We need to revitalize these universities, or dare I say it, make universities great again! MUGA!

We should get rid of the needless complexity of compliance and allow professors to engage more flexibly with industry. We need new models for collaboration, as Stanford no longer benefits as much as it could from brilliant former professors like Daphne Koller, Andrew Ng, Sebastian Thrun, and Bill Dally (CTO NVIDIA). We miss their experience and viewpoint in whatever capacity they have. It’s a great job. I personally would do the job for free. 

The AI race can also bring talent back to academia. Academia is at its best when it's a place to work on hard, multidisciplinary problems. For example, how will AI be a new tool for biology, physics, or the arts? Google doesn't have world experts in each discipline–universities do! Much like the space race fueled a generation of talent working tirelessly on not only rockets but a range of disciplines, the AI race could bring talent back to centers of excellence in academia. Just like the space race, academia’s contributions could be obvious to average Americans and a way back to America’s good grace and esteem. My view is clear: we owe the American people, they are the reason academia exists in its current form. Academia has done a bunch, but success is rented not owned: Let’s earn our keep!

More recently, we’ve seen that private organizations want to step up with public-private partnerships. I think more should happen! My personal favorite is the Arc institute that allows young faculty to have fantastic resources like compute and even collaborate with their senior, meme-loving colleagues. This leads to results like Evo, which I love, that could aid science and industry. These efforts grow the pie for academia and AI, but often are viewed with suspicion by administrators. 

I don’t claim to know the exhaustive list of ways professors and graduate students can contribute. They contribute in so many ways: some teach great classes, some mentor great graduate students, some have wonderful technical breakthroughs, and some build and connect communities in industry. There is a broad range of ways to contribute, and we need flexibility. To win the race, we need the ability to flexibly deploy our talent–and trust them in their bets!

There are already bright spots in Industry

There are a few people who I think deserve special praise. 

• Elon is very popular in my house. He is our best chance to go to Mars, and my kids love watching SpaceX launches together. When he took up the mantle of open source AI, I was thrilled. I subscribed to Grok (and everything else!). I hope he pushes this and builds more community, he can do it! 

• Meta and Mark Zuckerberg made a commitment to open source and the Llama models that put American opensource in the game and started the wave our friends at Hugging Face started! I also still love his above July 4th post (maybe that title robot is already here?!), which is also popular in my house.  We need more great leaders with technical vision. We have so many more form factors for intelligence to make sure the world runs on American AI rails. 

• I’m personally grateful to Vipul Ved Prakash, the CEO of Together, and Ce Zhang the CTO who lead with many others building the cloud for open source AI – and believed in it when it sounded insane to most people. Now, we’re at the cusp of helping to build that American foundation–shill or money where my mouth is? Let’s make it unambiguous: try the chat app! 

Open Source provides a way forward

We need a holistic AI effort from academia, industry, and the US government to have the best shot at a freer, better educated, and healthier world. I’m a mega bull on the US and open source AI. Maybe we’re cooking something bigger… stay tuned or contact us.

Our Discord channel: https://discord.gg/mvbk62T7CC