Products, not papers. Tinkering, not training. Apps, not Arxiv.
The AI Grant program:
- $250,000 for your AI-native product startup
- $250,000 in cloud credit from Azure
- A week in San Francisco with amazing people
AI researchers have given the world a revolutionary new platform in the form of large models trained on internet data. And the game has changed dramatically with the recent release of advanced open source models.
Some people think that the model is the product. It is not. It is an enabling technology that allows new products to be built. The breakthrough products will be AI-native, built on these models from day one, by entrepreneurs who understand both what the models can do, and what people actually want to use.
There are exciting early examples, like Midjourney, Copilot, and Jasper – new applications that have grown explosively, reaching millions of users and creating huge businesses practically overnight. This is just the beginning.
Our goal is to find and support the great entrepreneurs who are exploring this new frontier. We want to raise the status of building AI-native products. Research is important, but creating a product that people love on a new platform requires unusual levels of creativity and ingenuity – arguably more so than raising money and spending it to scale up a training cluster. And the rewards to the entrepreneurs who create these products will be enormous.
Researchers have raced ahead. It's time for entrepreneurs to catch up!
Advisors and speakers:
- Riley Goodside, GPT-3 whisperer
Organizing team: Funded by:
Can anyone apply?
Yes – no credentials required.
What type of company or project qualifies?
Anything that leverages AI models in a useful or engaging way. In particular, we're looking for technical & pragmatic founders who want to build great products. If you get a thrill from making something that other people love to use, and you understand that building a new product is 1% idea and 99% iteration, we want to support you.
What do I get if I win?
- A $250,000 investment via a no cap, no discount MFN SAFE (available here).
- Spend a week with peers and industry experts at the AI Grant summit in San Francisco, November 7th - 13th.
- $250,000 in cloud credits from Azure, along with other startup benefits, membership to the Microsoft for Startups Founders Hub, and go-to-market support.
- $50,000 in credits from the fine folks at Cohere
- Stability.ai has graciously agreed to provide early access to their new models. Some startups building on those models may also be able to get selective access to the Stability.ai ultracluster.
Wait, is this a grant or an investment?
It is an investment. When we started AI Grant in 2017, we handed out much smaller grants for fundamental research. We awarded grants to 36 individuals, who went on to do important research, and to found companies like Cohere, Helia, and Cresta.
Five years later, it seems much has changed: AI research is abundant, but UX innovation is just getting started, and so we're relaunching the program with a focus on AI products & investments versus AI papers & grants.
Will you fund multiple projects working on the same thing?
Yes. We plan to fund many companies; some form of overlap is bound to happen. Ultimately, we don't think this will be an issue because we strongly believe the AI product market is enormous -- large enough to support both Lyft and Uber or Bing and Google.
How will you select the winners?
We're looking for smart and energetic people with actionable ideas that are clearly useful. Demos are a big plus; not just because they're fun to look at, but because they hint at the type of person that cares about the end-user experience, which is ultimately all that matters.
Do I need a company to apply?
No, but you will need one if you're accepted into the program. We only invest in Delaware corporations, but you can create one from almost anywhere in the world with Stripe Atlas (or other programs) in only a few days.
Can I apply if I've raised money already?
Yes.
I'm a single founder. Can I apply?
Yes.
Who is funding this?
Nat and Daniel have invested $10M in AI Grant. The company investments themselves are made by AI Grant LLC.
If you are raising more than this, come talk to us directly! We're interested in great companies raising between $2M and $200M.
I love this idea and I want to help! Can I provide additional funding, hardware, GPU time, datasets, mentorship, or help reviewing applications?
Yes! Thank you for being awesome! If you want to contribute in any way, please email us at sponsorship@aigrant.org.
Can I work at a company you fund?
Yes, please fill out the form here and we'll help match you.
I have further questions! How can I reach you?
support@aigrant.org
What will AI-native products look like?
We have some hunches, but no one knows for sure
- It’s a new world; there is no map
- The best products will play to the strengths of this new platform
- They probably won't be simple wrappers on models
- We're interested in all good ideas, not just ones that accord with our hunches!
The copilot model: an assistant that looks over your shoulder and removes drudgery
- Copilot for support, copilot for legal, copilot for sales
- Fine tune on your own data
- Odd that people aren’t trying more of these
Utility products built on LLMs have to cope with models that hallucinate
- It’s easy to design a great product for a model that doesn’t make mistakes
- LLMs today alternate between spooky and kooky
- Design to minimize the user cost of bad model output
Latency matters
- Scaling laws tell researchers that bigger is better
- But big models are slow, and slow mistakes are annoying
- Latency/IQ/preference curve on model size is not fully understood
AI-native social networks
- Is AI-generated content the new user-generated content?
- Midjourney discovered that multiplayer is fun; will AI join the chat?
- Hyper-targeted personalized advertising
Generative entertainment
- Generate custom stories and fanfic
- Make existing books interactive
- Gaming: assets, NPCs, items, dialog, levels, worlds
- "Desktop Pixar"
New creative workflows
- Imagine a UI for generating 100 variants, sifting, refining, mixing, remixing
- “Render this character in 100 poses”
- Logline to plot to script to storyboard to video
User interface for text generation
- Most products are a single forward pass with a little prompt engineering
- Seamlessly shift between outline and text? Pin key details in source materials?
Mobile + AI
- How do large models unlock creation on mobile?
Chatbots
- Uncool right now because so many bad ones have been made
- LLMs might make great chatbots possible
- Side channels to apps, expert advisors, tutors, assistants, remote employees, friends?
AI-native knowledge retrieval
- Search the web, search my company, search my stuff
Earn money on mturk/upwork
- Upwork does $3B in gross sales every year. How many of those tasks can be done by large models?
Won’t big companies just do this?
- Maybe – they have users, data, and money
- If it's a simple bolt-on to an existing tool, they are more likely to do it
- But big companies are not good at inventing new products
- It might start as a toy, or in a small niche
- If it overlaps with the responsibilities of two VPs it's harder to do
- It might pose too much reputation risk
- They don’t need to build a great product; startups do