Show HN: AI Grant – Non-profit, distributed AI research lab
We have an idea for helping AI research and we’d like to hear your thoughts on it. We want to help get a lot more people working on ML projects they find interesting. If you've been thinking about or working on a side project or have some idea that won't let you go, you're who we want to reach.
Why try to help? Nat and I are passionate about AI. We want to see more Show HNs that use machine learning. We've been rewarded by pursuing our own shower-thoughts and want to remove any barriers from others thinking of doing the same.
Our plan for this is AI Grant (https://aigrant.org/), a non-profit distributed AI research lab. We're issuing grants to the smartest people we can find, doing interesting work that might otherwise not happen, and connecting them to mentors, experts, and each other. We ran our first round this spring, and awarded $50k in grants to 10 projects.
Filling out the form should take less than five minutes. Grantees get:
- $2,500 in cash.
- $20k each in Google Compute Engine credits.
- Q&A with AI experts including Andrej Karpathy (Director of AI at Tesla and previously at OpenAI) and researchers at Google.
- Access to the network of other grantees
- 250 Tesla K80 GPU hours from FloydHub.
- $1k in ScaleAPI data labeling credits.
- $5k in CrowdFlower data labeling credits.
Please let us know if you have any ideas or suggestions on how we might improve, either on the specifics of AI Grant or the general goal of spreading AI research to everyone smart who wants in.
- Daniel Gross & Nat Friedman.
P.S. I'm a partner at YC. This is Nat's and my side project, not a YC effort.
41 comments
[ 3.7 ms ] story [ 105 ms ] threadhttps://github.com/Latently/DeepLearningCertificate
Free access to ridiculous amounts of hardware and the opportunity to implement important and hot scientific papers and conduct original research in deep learning.
Latently, Bitfusion, and IBM Cloud enable democratized access to deep learning
https://www.ibm.com/blogs/bluemix/2017/07/latently-bitfusion...
> What's your background? In particular anything that would help convince us that you're actually capable of finishing your project.
> Project description: Describe your project, including: where you got the idea, how you think others might use it, and how it is new/different/better than what already exists. This is the main answer we use to judge applications.
These questions make the grant seem like a Catch-22. The people who are skilled enough at machine learning/deep learning to receive the grant would be able to be employed at any relevant company for their research.
Thanks to modern machine learning/deep learning tooling, the educational bar is much lower for more creative projects (e.g. Show HNs) which would have more necessity for the spare resources. But from the application, that's not the sense I'm getting.
1. We're looking for any information that will convince us that you're likely to complete the project. Academic background is merely one type of proof. A Github profile is another. Or recommendations from peers. Or a record of running marathons. You get the idea. We're open minded.
2. We funded a neural network to generate puns in the last batch. Non-conventional creativity is fine! At the end of the day, the project has to be useful to some humans. Even if it just makes them laugh.
It doesn't seem like this is something like the Thiel fellowship that's supposed to replace an education, it's just a way to make AI research affordable. I'm sure if I wanted to I could scrape together enough money to do a lot of cloud computing and be able to aggressively try out ideas, but that would cut into money for living expenses. With a grant like this, I would feel much more comfortable just experimenting, not feeling like if I train a model overnight and it doesn't turn out perfectly I lose $10.
I don't think a hobbyist can justify using Google/Amazon's best GPU processors, and instead must settle for their cheaper options to feel like they're being financially responsible. I see this AI Grant as a way to solve that problem. AI projects aren't like web-dev projects-- it can take significant amount of money to train a state of the art model in the cloud.
Yes, but maybe some of those people want to work on a side project in their free time. And if you work in deep learning you need some resources to train your models, so any help is welcome.
This isn't completely true. There are plenty of smart people who eg do Kaggle, and don't have the 300+ hours to invest in twiddling ensembles to actually win but do have the skills and knowledge to develop something new.
We detached this comment from https://news.ycombinator.com/item?id=14852088 and marked it off-topic.
My understanding: you get nothing tangible from the grant ("The money is a gift"), no contracts or IP... To me it seems like a bet that the networks effects and the gratefulness of the grantees will pay back somehow. I think this is a smart move because investing in smart individuals is itself smart, although the devil is in the details.
As for value capture, there's no master plan. Getting to know smart people is enough of a reward.
One of the things I've finding difficult to find in my process is actually mentoring support. You can learn a lot from online resources, but often I have hit walls where I wish I could ask for help and guidance.
It'd be cool to a version of this with more defined mentoring path, especially for those who are new to the space!
You will probably have to make an effort to differentiate yourself from crackpots and demonstrate pre-existing knowledge of the field. In my case I had to answer a number of questions of the form "I think you may not be aware of XYZ, which is quite similar to your idea." with "The paper on XYZ inspired my idea, but I'm trying something different."
A targeted venue that pre-selects for researchers willing to spend their time on mentoring others (who are not officially their students) could help, but it would still have to deal with the crackpot problem.
Tom Kalil https://www.linkedin.com/in/tom-kalil-a581/
I know I'm not the only one.
And let me use this opportunity to pull out this gem I found once from my quotes file:
Getting to a point where you control your own research and have financial stability for research that could span 3+ years with very little output (paper wise) is something comparatively few academics achieve. Even fewer of those who have, did so by conducting longer term research. Most academics I know are just trying to produce as many papers per year as they can and do whatever research in their field they can do, to do it.
Like groupcurrency.org?
But nobody finances this. Maybe an investor reads this understands that AI is not the next social network or JS framework. It is the pinnacle of human invention and requires that amount of financing.
There could be a bounty for find bugs and suggesting features.
Maybe charge $2/mo. It would have a fiduciary duty to you as a user to act as an agent in your best interest. Allow people to contribute to the project using distributed GPUs(with legal contracts to minimize fraud).
Open budget with bidding and open evaluation criteria.