Interesting pivot. They used to call themselves a "non-profit distributed AI research lab" that gave small chunks of cash to AI researchers. Now they've changed to much larger investments for products/startups.
Given that the Hacker News frontpage has been taken over by GPT-3/DALL-E/Stable Diffusion in the past couple months, consumer AI popularity definitely is not fading away.
The HackerNews community is not reflective of the general tech, consumer, or business environment. First-hand experience in delivering and selling AI solutions to enterprises and governments plus activity at conferences, venture capital investment, and other measures are showing that interest in AI is generally cooling off, even if there are interesting things happening with generative transformer models, which is where the interest is currently.
DALL-E mini/Craiyon, Midjourney, and Wombo all blew up because of the easy/free accessibility of AI generation tools, to the point of John Oliver doing a sketch about it: https://www.youtube.com/watch?v=3YNku5FKWjw
The consumer applications will incidentially provide more of an incentive into investing into AI research since there's a good chance of ROI, which will trickle down into better open source AI like Stable Diffusion.
I’d be interested in what you mean by that - VC investment has slowed down, but investment by large firms (tech or otherwise) is only increasing.
And with large language models, generative models, diffusion models, etc becoming more and more common its becoming less novel, but certainly not less popular.
As someone on the outside of AI tech, I wonder if this is also a reflection of how difficult it can be to make create a differentiated product in the AI space on a small budget. The mission statement also indicates that they would like to find products that are complex, likely to create a moat and make it difficult for competitors to move into the same space quickly.
What exactly are they giving back? They're taking a cut of your company using a MFN SAFE note, same terms as many seed investors. They just are investing a fixed amount into a very focused niche. It's a seed investment finance play, same story as happened in many other industries from small SaaS-funds to biotech to crypto to quantum and beyond.
Okay thanks. Just wondering how it compares to YC. I know YC says mfn safe too but they say 7% (and it’s 500k and obvious differences in experience, mentoring, process etc)
I’ve always wanted to setup a nonprofit to give motivated people a grant (not investment!) that enables them to build an app/product. Like maybe someone who didn’t go to college but wants an opportunity to build a product.
How hard is it to set this up? The process to start a nonprofit doesn’t seem inviting.
it depends on how big you expect the endowment to be - aka how much capital is your nonprofit going to be working with
are you going to be the sole donor and board member, or are you going to be actually raising donations because thats a full time job
giving grants for coding is hard, it has to match your mission statement that got you the non-profit status to begin with, and the IRS is not enlightened on the concept of coding as a charitable class, but you can have other things like educational purposes, or "sustainability" and give grants to organizations that are ultimately doing coding.
grants and scholarships to individuals are much more complicated, but they can incorporate and nobody cares that the new company is just wrapped around an individual
also, all of this can be circumvented due to the fact that non-profits can invest. growing the endowment is simply asset management and there is no difference in investing in shares on the stock market and investing in shares of a private C-corp that was just formed that just your non-profit filled the seed and series a round for. non-profits can make wild profits, no different than Harvard's endowment growing indefinitely. there are also some cases where investments can count as the grant requirement (if your type of non-profit is one that has an annual grant requirement, via regulation)
yeah the regulations actually make the individual charity part of non-profits really hard and you can do lots of profit making activities! but you can try your goal
Alternative view: this is a pre-pre-seed and fills a much needed gap in the VC ecosystem. Well-connected founders raise very early money from "friends and family". Most people don't have wealthy friends and family, so getting the first money in to kickstart the venture is difficult.
> So like 1 experienced researcher or 2 college grads for 1 year's worth of salary. Putting it through that perspective, it does not seem like much.
Well, I think the point is that you would be the one doing the work at first and you wouldn't have to pay yourself market rates because you also get to own the vast majority of the company.
> Wondering now how much hours of compute this would buy.
It also comes with $250k in Azure compute credits, which would buy
~70,000 hours on a machine with an A100 GPU, for instance.
I participated in another Daniel Gross program, Pioneer, and it was _excellent_. I highly recommend it or this for folks that may be more 'outside the band' when it comes to raising money or getting a startup from 0 to 1. I would note that if you're trying to bootstrap this is probably not the program for you. These guys are incredibly connected in venture capital and this program ought to be a great way to leverage that.
"8k-ish" I read this as k8s-ish... which makes me thing maybe 8k-ish could be your product name, as a play on k8s-ish. That said... ive been learning kubernetes lately and it doesn't necessarily relate to AI.
I like that. It's a very on the nose name right now. I'm considering a flight to Anguilla to see a man about a parked domain (after the competition), if that gives any context to the brevity and nature of the name.
The fundamental product is large and physically involved, with the 'AI' under the hood to sort of Pixar the whole experience and make the 'game' more satisfying, connected, and fulfilling. Sort of a means to progress, in a general sense.
> Spending 8k-ish to build a product demo for my friends rich uncle is too uphill for me to risk.
That’s something I could potentially fund. No equity, just pro-bono / repay if it works out. Feel free to reach out and chat if it’s a dream you believe in.
I could definitely get off the ground for much less, but having rent/bills covered to go all-in, and capital for speed-to-market / competitive advantage is real nice. As long as I have a legally maximized control of the charter considering the investment, I'm happy to give up future cash. To me, realized gains is a byproduct of viably bringing the product to the world. But doing the thing is the purpose.
And it would free up my time to focus on scaling and experimenting instead of assembling.
I'm genuinely taken back by that offer, and I'll be reaching out to your email with an introduction. Thank you for reaching out.
I'm working on building a search engine specifically to help with prompt engineering https://pagebrain.ai/promptsearch/
Is this useful enough to be a startup?
Appears super similar to Lexica.art, but less data; for example, appears you’re missing even the limited about of parameters listed by Lexica, which still are incomplete.
Is anyone actually using co-pilot? They claim “millions of users” but I seriously doubt millions of programmers opted into paying $10 per month for it. A lot of these stats seem highly suspect.
I am using it for Elixir and Python. It surprises me at times, and sometimes it just misses. Most of the time I end up with a template that is just a little bit off which is fine.
Yep, it's been a game changer and I happily spent the money.
As someone who has strong programming intention, but is not always able to easily remember syntax, or visualize a solution, it's been a force multiplier.
It often helps with either typing something out or having to search for something online. As an example, it really helps with switching in between languages where I sometimes forget the exact method name or syntax. Co-pilot helps generate code that I can easily verify but would be annoying to write.
For me having to fix weirdly incorrect code is much more annoying than writing new code. Similar to the issue that I have with driver-assist tools: supervising a system that does ok much of the time but then does something batshit insane every once in a while is a lot more stressful than just doing it myself.
I much prefer predictable tools like code templates where I know exactly what I will get when I trigger it.
Unpredictability is the fundamental problem with these tools.
MFN SAFE means it converts to equity in the next funding round, at the price set by the lead investors in that round. E.g., if the next round had a $2.5M post valuation, this would become 10%.
Hey folks, this is Evan from AI Grant! We're excited to support founders building new AI products.
AI Grant is investing $250k on an uncapped, no-discount, MFN SAFE, as well as providing more than $300k in cloud credits.
AI Grant used to do small grants in lots of AI research projects. Now, we think there's plenty of research, and what's needed is products: so we're doing bigger investments in founders building products. Name's the same, but now we can deploy larger checks, and focus on what we think will help the most in 2022. Frankly we used the name "AI Grant" because we already had it!
If you're working on something cool, please apply! And if you have a special situation, or need help, reach out to support@aigrant.org
How does the "uncapped, no-discount, MFN SAFE" work? What does that mean in terms of the percentage of equity you'd take?
I'm self-funding fakeyou.com from a previous exit. My AI/ML startup is called "Storyteller" and my team is about to launch real time voice conversion (you -> Trump, Adele, etc.), music generation (instruments + vocals), and real time motion capture (for film and animation, essentially super high quality VTubers) in a matter of weeks, plus a bunch of other stuff by EOQ3. From there it's AI musicians, story generation, etc. A takeover and democratization of Hollywood and Nashville.
I have no idea what I'm doing in terms of fundraising. We have 5M+ MAUs and I don't even have a deck. I'm in Atlanta grinding out product and I'm not plugged into this VC/fundraising world, but I know I need to grow this.
The $300k cloud credits sounds right-sized. I also started buying a metric ton of 3090s and running k8s both in the cloud and on-prem. :)
Apologies if this comes off as harsh, but I'd recommend doing your own research on fundraising and common terms rather than trusting potential investors to educate you.
Very high level explanation of a SAFE is that it converts on the valuation of your next funding round. Uncapped means that if you raise at a $200m valuation in one year, the SAFE will convert to equity at 250k/200m or 0.125% of the company.
If it is capped then you replace “200m” by the minimum of “200m” and some agreed on amount, e.g., 10m. This protects the investor in case the first equity round has a surprisingly high valuation, which would dramatically dilute them. At the same time, instead of agreeing on a valuation for the initial investment, you are agreeing on a theoretical upper bound on the current valuation.
Capped means the investor is putting in a floor for the minimum stake they’ll accept in the next round. So regardless of next funding round they would get own (original investment)/(new valuation with maximum value of cap) = % of company owned. Uncapped means the denominator will be whatever new valuation is.
How it matters to the founder - if it's uncapped then as the next round goes to infinity, the equity from the SAFE round goes to zero (see the 200MM example). If It's capped, then the SAFE will convert at no lower than X%.
Uncapped is better for the founder, but the difference only will ever really matter if it takes off so fast that the next round is at a much higher valuation than expected.
Beyond your specific question, which appears to have been answered, highly recommend the book “Venture Deals” to learn more about equity based funding:
Hmm… their weights vary a bit more than I would have expected; in a little searching, I’m finding figures ranging from 1372g for a Gigabyte (which would be ~729 to the ton) to 2157g for the Founders Edition (which would be ~464 to the ton).
(Aside: this is the first time I’ve reached for Google in well over a year, because DuckDuckGo categorically failed to find anything even vaguely useful or relevant for my queries, whereas Google found plenty of meaningful results.)
I'd guess most people wouldn't know the meaning, so here it goes:
MFN Clause: In a most favored nation (MFN) clause, if subsequent convertible securities are issued to future investors at better terms (e.g., a lower valuation cap), the better terms will automatically apply to the investor's SAFE.
We're working with AI to help human writers write financial content, and we're not far off rolling our own models. Is that too run-of-the-mill for this? We do have some traction.
Sorry if this is obvious, but it's $250k for what percentage?
I have something that might be the type of startup (integrating AI into algonfts.art) except it's not a very popular niche and probably wouldn't make back the $250K. Although that reality doesn't seem to stop most people.
As I understand it's at the best terms they give on their next raise, ie if they raise at a 2.5 Million valuation, this investor will get 10%. That's the mfn clause.
I love this, there are so many interesting questions and challenges in the area of how humans will interact with AI. To me the correct model is to think of intelligence as a resource like compute as opposed to a personified computer, which opens up the opportunity for so many more applications.
An example is the recent text-inversion paper where they found a way to extract the embedding of a visual concept from several images. Or how Midjourney is able to somehow integrate their model with StableDiffusion to get the realism and control of SD but the aesthetics of Midjourney.
Basically, it seems like deep learning networks are a lot more composable and controllable than I would have expected, while still demonstrating really interesting features. There could be entirely novel neural net architectures with no analogue in the human brain that still do really interesting and useful things. It’s exciting to see what people are coming up with!
It amazes me how many VCs or angel investors will put out all this information about what they want to see in a company or founder they invest in, but close to nothing about themselves on their website.
“We have money to invest come and apply to us and we will determine if you are good enough.”
Bro please.
If we are building game changing AI before we take your money we want to know what you stand for. What are you made of. How do you want to shape the world. What’s your track record of dealing with small companies. What’s your philosophy around investor relations. Prove to me you’re not a complete dick to work for or to have on the board.
There’s a pretty decent set of info and links to personal profiles on their website. If that’s not enough for a founder to do due diligence, they probably aren’t cut out for VC fundraising.
And someone writing small checks like this is almost certainly not going to be on the board unless the founder is clueless.
Not enough. This “apply to us” format should be deprecated. There’s a solo investor on twitter who says “DM me would love to chat if you’re building something.” And that’s much more approachable. She’s flooded with high quality founders and startups.
People who run their own thing are looking for partners not a boss or anything that resembles one. This “fill up this form job/university application format so some alleged superior being judges you” is exactly what they ran away from. Speaking from experience.
Yes, they offering seems to be quite weak than other renowned incubators like YC etc where you get more more support and possibly better funding deals.
You have a fair point. It's interesting because I was literally just going through government programs that provide startup funding, where the application burden is high, the value is pretty low, and the format overall skews hard towards acting like the funding org has all the power and applicants should be thankful that someone deigned to review their application.
It's actually similar to the job market. There is government, where only the most desperate, process oriented folks bother to go through the silly bureaucratic process, there's "traditional" application-interview-offer where the employer has the power, and there's a pull model for top people where they get hunted for jobs.
The more power and brand you actually have, the more you can skew traditional / bureaucratic. If you're an unknown startup, you can't just post a job as and expect to get good candidates (I've tried that) - but you get a lot of mediocre ones. And if you're an unknown investor, I'd expect the same.
These investors either have done similar enough things before that they have a sense of the candidates / companies they'll get and are cool with that level, or will learn this time about the volume and quality this kind of solicitation generates.
(
Products, not papers. Tinkering, not training. Apps, not Arxiv
Personly I think we're still at a stage where real research is important if you want ML to be an actual differentiator, and just e.g. cramming GPT into something is going to give you something very shallow, but let's see. I've been thinking about AI products a lot recently because I have an AI startup, and what I've concluded is that for AI to actually be a core differentiator and not a peripheral feature requires fundamental ML work specific to the product. But as I say, let's see what people can do)
Forgive me for this unrelated comment to the post. This website feels like it was designed and optimized to avoid typical HN commenters saying: Please change your site font size! This website is tracking me! This website isn't loading in X.
Having spent a good deal of time trying to productize "AI" at both large companies and at an AI startup in the enterprise/b2b space, anytime I hear about a new AI fund or a new AI startup, I am reminded of this A16Z blogpost. It resonates very well with my own experiences: https://future.com/new-business-ai-different-traditional-sof....
@flaque How many projects are you looking to fund in this cycle? Is the $10M over the course of many years and many batches? Are you doing rolling decisions or interviews only after October 1?
I would love to apply on these positions but it takes a colossal effort to pull anything. Those willing to bet into the future, I have few suggestions.
1. Design an app that takes in series of text prompt and generates images for corresponding prompt that is coherent across all images. These images will be keyframes.
2. Design an app that acts like database of videos, so that you can query any kind of clip.
3. Try to use view- consistent generated images to get 3D models.
4. Define agents that can play for a given rule and gradually adjust the rule or even form new rules!
I want to be that post in hn which inspires multi million dollar company!!
100 comments
[ 223 ms ] story [ 3456 ms ] threadSee their HN announcement 5 years ago: https://news.ycombinator.com/item?id=14851806
The consumer applications will incidentially provide more of an incentive into investing into AI research since there's a good chance of ROI, which will trickle down into better open source AI like Stable Diffusion.
And with large language models, generative models, diffusion models, etc becoming more and more common its becoming less novel, but certainly not less popular.
>> Wait, is this a grant or an investment?
> It is an investment.
Easy to misunderstand. The word "investment" does not show up until way down the page.
They're not saying that they give grants now.
(from "open" in the article)
How hard is it to set this up? The process to start a nonprofit doesn’t seem inviting.
are you going to be the sole donor and board member, or are you going to be actually raising donations because thats a full time job
giving grants for coding is hard, it has to match your mission statement that got you the non-profit status to begin with, and the IRS is not enlightened on the concept of coding as a charitable class, but you can have other things like educational purposes, or "sustainability" and give grants to organizations that are ultimately doing coding.
grants and scholarships to individuals are much more complicated, but they can incorporate and nobody cares that the new company is just wrapped around an individual
also, all of this can be circumvented due to the fact that non-profits can invest. growing the endowment is simply asset management and there is no difference in investing in shares on the stock market and investing in shares of a private C-corp that was just formed that just your non-profit filled the seed and series a round for. non-profits can make wild profits, no different than Harvard's endowment growing indefinitely. there are also some cases where investments can count as the grant requirement (if your type of non-profit is one that has an annual grant requirement, via regulation)
yeah the regulations actually make the individual charity part of non-profits really hard and you can do lots of profit making activities! but you can try your goal
Wondering now how much hours of compute this would buy.
Well, I think the point is that you would be the one doing the work at first and you wouldn't have to pay yourself market rates because you also get to own the vast majority of the company.
> Wondering now how much hours of compute this would buy.
It also comes with $250k in Azure compute credits, which would buy ~70,000 hours on a machine with an A100 GPU, for instance.
https://pioneer.app/
As someone who has a lot of tinkering and research underway on a big idea, this motivates me to scrap and scrounge to get the demo party together.
Spending 8k-ish to build a product demo for my friends rich uncle is too uphill for me to risk. But for this, I mean fuck yeah.
The fundamental product is large and physically involved, with the 'AI' under the hood to sort of Pixar the whole experience and make the 'game' more satisfying, connected, and fulfilling. Sort of a means to progress, in a general sense.
That’s something I could potentially fund. No equity, just pro-bono / repay if it works out. Feel free to reach out and chat if it’s a dream you believe in.
But $250k from this might go a lot further.
And it would free up my time to focus on scaling and experimenting instead of assembling.
I'm genuinely taken back by that offer, and I'll be reaching out to your email with an introduction. Thank you for reaching out.
Appears super similar to Lexica.art, but less data; for example, appears you’re missing even the limited about of parameters listed by Lexica, which still are incomplete.
https://lexica.art/
i was not expecting github to start charging for such a mediocre product. obviously i decided we don’t need it.
but i expect we’ll get something open source and better relatively soon considering the pace of innovation in this field.
As someone who has strong programming intention, but is not always able to easily remember syntax, or visualize a solution, it's been a force multiplier.
It's been pretty invaluable for vector math and reasoning in Unity.
It often helps with either typing something out or having to search for something online. As an example, it really helps with switching in between languages where I sometimes forget the exact method name or syntax. Co-pilot helps generate code that I can easily verify but would be annoying to write.
I much prefer predictable tools like code templates where I know exactly what I will get when I trigger it.
Unpredictability is the fundamental problem with these tools.
$250k for how much equity?
AI Grant is investing $250k on an uncapped, no-discount, MFN SAFE, as well as providing more than $300k in cloud credits.
AI Grant used to do small grants in lots of AI research projects. Now, we think there's plenty of research, and what's needed is products: so we're doing bigger investments in founders building products. Name's the same, but now we can deploy larger checks, and focus on what we think will help the most in 2022. Frankly we used the name "AI Grant" because we already had it!
If you're working on something cool, please apply! And if you have a special situation, or need help, reach out to support@aigrant.org
I'm self-funding fakeyou.com from a previous exit. My AI/ML startup is called "Storyteller" and my team is about to launch real time voice conversion (you -> Trump, Adele, etc.), music generation (instruments + vocals), and real time motion capture (for film and animation, essentially super high quality VTubers) in a matter of weeks, plus a bunch of other stuff by EOQ3. From there it's AI musicians, story generation, etc. A takeover and democratization of Hollywood and Nashville.
I have no idea what I'm doing in terms of fundraising. We have 5M+ MAUs and I don't even have a deck. I'm in Atlanta grinding out product and I'm not plugged into this VC/fundraising world, but I know I need to grow this.
The $300k cloud credits sounds right-sized. I also started buying a metric ton of 3090s and running k8s both in the cloud and on-prem. :)
To address your question, an "uncapped, no-discount, MFN SAFE" roughly amounts to getting the same terms that will be given to your next investors: https://www.dentonsventurebeyond.com/resource-center/safes/
More details here https://capbase.com/how-do-safes-work/.
How would it be different if it was capped? I went to your link but there are no mentioned of "capped" or "uncapped"
Uncapped is better for the founder, but the difference only will ever really matter if it takes off so fast that the next round is at a much higher valuation than expected.
That contract seems like a really bad deal unless you’re certain to be raising a lot of money & need capital to cross some barrier to get there.
https://www.amazon.com/Venture-Deals-Smarter-Lawyer-Capitali...
Hmm… their weights vary a bit more than I would have expected; in a little searching, I’m finding figures ranging from 1372g for a Gigabyte (which would be ~729 to the ton) to 2157g for the Founders Edition (which would be ~464 to the ton).
(Aside: this is the first time I’ve reached for Google in well over a year, because DuckDuckGo categorically failed to find anything even vaguely useful or relevant for my queries, whereas Google found plenty of meaningful results.)
I'd guess most people wouldn't know the meaning, so here it goes:
MFN Clause: In a most favored nation (MFN) clause, if subsequent convertible securities are issued to future investors at better terms (e.g., a lower valuation cap), the better terms will automatically apply to the investor's SAFE.
I have something that might be the type of startup (integrating AI into algonfts.art) except it's not a very popular niche and probably wouldn't make back the $250K. Although that reality doesn't seem to stop most people.
Edit: already explained in several posts above
An example is the recent text-inversion paper where they found a way to extract the embedding of a visual concept from several images. Or how Midjourney is able to somehow integrate their model with StableDiffusion to get the realism and control of SD but the aesthetics of Midjourney.
Basically, it seems like deep learning networks are a lot more composable and controllable than I would have expected, while still demonstrating really interesting features. There could be entirely novel neural net architectures with no analogue in the human brain that still do really interesting and useful things. It’s exciting to see what people are coming up with!
[who?] [citation needed]
“We have money to invest come and apply to us and we will determine if you are good enough.”
Bro please.
If we are building game changing AI before we take your money we want to know what you stand for. What are you made of. How do you want to shape the world. What’s your track record of dealing with small companies. What’s your philosophy around investor relations. Prove to me you’re not a complete dick to work for or to have on the board.
And someone writing small checks like this is almost certainly not going to be on the board unless the founder is clueless.
People who run their own thing are looking for partners not a boss or anything that resembles one. This “fill up this form job/university application format so some alleged superior being judges you” is exactly what they ran away from. Speaking from experience.
It's actually similar to the job market. There is government, where only the most desperate, process oriented folks bother to go through the silly bureaucratic process, there's "traditional" application-interview-offer where the employer has the power, and there's a pull model for top people where they get hunted for jobs.
The more power and brand you actually have, the more you can skew traditional / bureaucratic. If you're an unknown startup, you can't just post a job as and expect to get good candidates (I've tried that) - but you get a lot of mediocre ones. And if you're an unknown investor, I'd expect the same.
These investors either have done similar enough things before that they have a sense of the candidates / companies they'll get and are cool with that level, or will learn this time about the volume and quality this kind of solicitation generates.
(
Personly I think we're still at a stage where real research is important if you want ML to be an actual differentiator, and just e.g. cramming GPT into something is going to give you something very shallow, but let's see. I've been thinking about AI products a lot recently because I have an AI startup, and what I've concluded is that for AI to actually be a core differentiator and not a peripheral feature requires fundamental ML work specific to the product. But as I say, let's see what people can do)1. Design an app that takes in series of text prompt and generates images for corresponding prompt that is coherent across all images. These images will be keyframes.
2. Design an app that acts like database of videos, so that you can query any kind of clip.
3. Try to use view- consistent generated images to get 3D models.
4. Define agents that can play for a given rule and gradually adjust the rule or even form new rules!
I want to be that post in hn which inspires multi million dollar company!!
Good luck.