Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: AI Grant – Get $2,500 for your AI project in minutes
323 points by danicgross on April 5, 2018 | hide | past | favorite | 173 comments
Hi HN,

Last year, we launched a non-profit AI research lab called AI Grant (http://aigrant.org). Our goal is to fund promising people around the world working on AI. No strings attached. We've since given away over $100,000 to 30 teams working on different projects. You can see some of them here: http://aigrant.org/#finalists. We just started accepting applications for our third batch!

The academic grant application process is burdensome. It's only avalible to a choice few. Applications take days to complete.

AI Grant is open to anyone on the internet. You can apply in under an hour. Fellows get:

  - $2,500 in cash.
  - $20,000 each in Google Compute Engine credits.
  - $5k in CrowdFlower data labeling credits.
  - ScaleAPI data labeling credits.
  - Access to the network of other Fellows
Our long term goal is to build an online community of people working on interesting side-projects in AI, supporting each other. We've built some infrastructure around this (chatrooms, group video conferences). We'll continue to expand this over time. I'd be very curious to know what HN thinks about this idea, and how we might make it better. Thanks!

Oh man, I just spent the last few hours writing my application and now it's asking me to complete "a 10-minute personality questionnaire, followed by a 20-minute brain teaser"

"No strings attached" seems a bit disingenuous too if it requires taking aptitude and psychometric tests.

I read it as "no strings attached on how you execute or spend once funded". They can still do their selection any way they see fit (otherwise it would be either first come first serve or random, neither of which is a good funding model). My 2c.

Correct! Most importantly, there are no strings attached in terms of any IP rights on whatever you create.

Actually, randomly could be a good funding model. https://arxiv.org/pdf/1802.07068.pdf

Well, they need to train their AI project on something, right?

"We can pay you $2,500, and you'll still be the product"

The beauty of AI work today - the technology has finally become accessible. You've got OpenCV, TensorFlow, Caffe - with tons of well-documented models and working examples. You can run them on your consumer GPU and get real, practical results.

Now, I can understand Google trying to be the platform of choice (when in reality, you don't "need" them to build AI), and I can understand startups like CrowdFlower trying to get embedded in that process - but these .orgs need to be more upfront about their agendas.

I guess "in minutes" is subjective.

Well, 7 hours are also some amount of minutes.

How many minutes are there in a day?

doesn't Einsteins theory of relativity tells us as much?

Thanks for that feedback! Since we’re hoping to scale the program one day, we’re looking for better ways to identify scientists we should support. That additional test is just an experiment.

We need to do a better job communicating that upfront. Will fix now. Thanks again. Some things you can only learn from talking to your users...

I guess there's no free lunch...

A few thousand smart people worldwide surrendering their more or less smart but however interesting AI ideas and their very own private psychometric data for a lottery ticket (say 1-2% winning chance) worth $2500 + unnecessary internet services? Come on YC, you can do much better than that.

Hi DrNuke! A few thoughts:

> surrendering their AI ideas

Nobody is surrendering anything. The money is literally a gift. We have no rights in whatever they produce.

> private psychometric data

We have a personality test we’re experimenting with for future batches. We’re still making funding decisions based on the application for this batch. If you’d prefer to avoid filling it out, by all means!

> unnecessary services

I disagree! Money and $20,000 in GPU training time is worth a lot. Most of our prior Fellows put it to good use.

> Come on YC

This non-profit is unrelated to my role at YC. Please direct any frustration at me, not YC!

You'd be a cool kid, if you'd sign an NDA. (edit. personally I do believe that you are truly just trying to push the A.I. tech forward and to stay current with the best ideas in the field.)

Mnyeah. I guess you could call building a profile of your users (say, to sell to third parties) "staying current with the best ideas in the field".

Kind of like "the best minds in a generation are trying to make people click on ads" etc.

I think its cool. I also think there is a healthy culture of skepticism and mistrust on this message board. Will we be able to learn anything about the results of this raffle?

> surrendering their more or less smart but however interesting AI ideas

What does this mean? Telling AI Grant about the idea? That doesn't warrant describing it as "surrendering" if so.

Yeah, you won‘t get far if you never tell anyone about your idea...

> unnecessary internet services

GCE credit might be useful for running GPU instances for training

You can get these from Google directly, or through one of the many accelerators and tech incubators - without providing all this data. As for CrowdFlower credits - I don't see why an aspiring AI company would want to build anything on top of another startup's technology stack, as opposed to building their own core tech.

"You can get these from Google directly, or through one of the many accelerators and tech incubators"


... by applying to their Startup program:


Problem is that you have to be a part of one of the "current partners" so if you haven't joined an accelerator or have VC funding you don't get access.

Unless I am mistaken. I'd be interested if I was able to find more credits for AWS/GCP/etc

I completed the whole application and then it presented the quiz/survey. I didn't have time to complete it so I closed the window. I never created an account...does your application save automatically? Seems like poor design to have users fill out an application without logging in and then showing them a survey that they may not have time to complete.

Thanks for the feedback! Your application progress does save when you advance. You should have a link in your email to restore it. Please email me if you don’t: daniel@aigrant.org.

We will also communicate the upfront time investment more clearly.

Part 2 of this application is weird.


Experimental Assessments

Thank you for completing your project description. Next you'll be presented two assessments: a 10-minute personality questionnaire, followed by a 20-minute brain teaser.

This part is an experiment. Selections will be driven by your project description, not your score on this test. However, you're required to complete this section to submit your application.


I agree it's weird. It seems like a classic personality test and pseudo (part of an) IQ test combined. Even collecting these results in the first place raises questions for me. Say that the results do indicate a personality disorder and/or a low IQ. Having that information tied in some random database you applied to for a "side project" raises privacy concerns. Note that the website has no clear (one I can easily find) privacy policy related directly to this aspect of the application. That's a red flag for me since this is highly personal data. In fact it's pretty unprofessional. A couple sentences saying you won't use the data is not sufficient (also why are you collecting it if you're not going to use it? More transparency please).

I would suggest to the creators of this that they immediately delete existing data collected through this part of the application until they can put together a proper privacy policy and be more transparent about how they're going to use the data.

I didn't care why, I cared because it was decietful. Eventually by the time they wanted an additional '20 minutes' AFTER already doing a ten question spatial awareness test I quit. Could go on forever. Very private stuff too. To not be upfront with things like that is wrong.

Yeah, transparency matters in this. Laying out the required steps at the beginning is easy to do.

Totally agree. All current data should be deleted.

Meyers-Briggs testing strikes again! /s

Please be transparent about the process, maybe add a "progress tracker". I'm grateful for the HN community saving me much time informing me there's a "personality questionnaire" (wtf is that?) awaiting at the other end of the funnel.

[E: this should have been down there sorry]

Is this open to Europeans and have you heard of GDPR?

Thanks for the feedback! Some extra context: Since we’re hoping to scale the program one day, we’re looking for better ways to identify scientists we should support. Per the description, it’s very experimental and we’re open to any feedback you have.

Personally my views are you have a right to ask for whatever you want. Don't lie about it and call it a 'grant in minutes'. Sure there is 525,600 minutes in a year. That many?!! Spell out everything you want upfront. This is a dark design pattern and it's unethical and especially coming from you all you should know better. With sensitive information don't lead people on.

I think as long as you make it clear that there is a 10-minute questionnaire and 20-minute brain-teaser at the start of the application people probably won't mind.

Plus you might want to change your claim from 'minutes' to 'hours'. You won't put off people who really want to apply and you'll get less people who just want easy money, trying to apply and reacting angrily.

Is there a name for the pattern recognition section? I'd be interested in reading more about it and how you use it as a measurement/how the results correlate with the decision process. Does it tie in with the behavioral aspect at all or are they completely independent?

Agree that you would need a way to filter out applicants who are just half way there.

But I think such tests have no relation to the mission. Mandating a short video on capabilities and need of problem area is likely to filter out most candidates. Making video requires effort.

This will get you people who are serious.

Further you could add option to either take test or 15 min video. Just a suggestion.

I think the process is great! It's super fast. I hope you do get to glance at the "IQ" test results - if someone is four standard deviations (160 IQ - doesn't describe me) above the mean, then you should fund them even if you don't understand one word of their description. Maybe nobody does in the entire world except for them.

Kudos on the process. Nothing wrong with it.

all that IQ tests validate is how well someone can take an IQ test. It means absolutely nothing in terms of validity of ideas, business sense, self motivation, discipline, leadership skills etc.

That is not true. IQ correlates with basically every good life outcome.

Here's an example discussion of how this works, drawing on various studies: http://slatestarcodex.com/2015/08/02/stalin-and-summary-stat...

You try a rebuttal, which I commend, but the approach of your rebuttal isn't strong enough! Height also correlates with "basically every good life outcome" but the application form didn't ask for our height: nor should it.

IQ is inherently different. I really support the IQ test they decided to give and I definitely think they should glance at the results - especially to basically approve any applications from people with great results even if the evaluators don't understand the proposed application exactly.

yeah? well I say: all a normal funding application does is test how well you can fill out a funding application!

So adding an IQ test changes that to something meaningful, that correlates with AI (what these applications are for). After all, how is a low-intelligence person supposed to make a project in artificial intelligence!

I stand 100% by my support of their process.

Wow that is actually disgusting.

There might be a lower limit IQ you might need to be able to hold enough data in your head to archive specific thoughts but i don't think that this limit is very high.

There are enough dedicated human beings out there who are not that smart but dedicated. They are easily able to do projects in artificial intelligence.

AI is not that complicated. It takes probably more stamina than IQ as a lot of research does.

I am certain that you are factually mistaken and there is not one single AI researcher anyone in the world who has ever personally coded a single project without having a high IQ. Sorry.

I'm an AI researcher, I have been coding like crazy for the last ten years and I'm dumb as bricks.

Sorry to burst your bubble.

What is your IQ? If you don't know go ahead and test it online.

I will go ahead and guess 127: nearly two standard deviations above the mean. (i.e. IQ-wise in the top 5%.)

Call it dumb all you want: that's just you being modest.

I find it very funny that you would think I'm, like, super smart. That has literally never happened to me before :)

I'm not being modest. I'm lucky to have smart people around me and I know how smart they are and how smart I am, and I'm not that smart. Anyway I've learned not to worry about it. I'm tenacious. That's taken me a long way already.

If it helps, though, I once cheated on an online IQ test for a job. I took it with a friend and we got 130 points together (I didn't like the job).

I guess that means my IQ is somewhere around 65 :P

oh okay, so I guessed 127 for your IQ and you already got 130 on a test you took with a friend. don't take an IQ test for me if you don't want to but I'm done proving my case.

Oh, alright then- you win, I'm super smart :|

But modesty is a virtue :) it just doesn't help the current thread. :) cheers,

I have had access to the research version of Cyc (http://www.cyc.com/) for over a year, and its capabilities are breathtaking. If aigrant.org had a subcommunity of people who were working on Cyc-related side-projects, I would definitely be interested in contributing to it.

What would be helpful is if aigrant.org could provide an account on a Cyc instance to community members.

I'm curious what you saw it doing. Cyc has been a vaporware product for many many years. It's had corporate backing on and off for a long time and produced next to nothing.

I spend 1-2 hours each day learning Cyc, and each day I see it do something new. For example, I just asked Cyc "Where did the 2012 Summer Olympics take place?" and this was its response:

        (YearFn 2012))
         #  ?VAR                          
        --- ---------------------------------------------------------------------
          1 UnitedKingdomOfGreatBritainAndNorthernIreland
          2 England                       
          3 EasternContinent              
          4 ContinentOfEurasia            
          5 Occident                      
          6 MilkyWay-Galaxy               
          7 PlanetEarth                   
          8 NorthernHemisphere-Region     
          9 ContinentOfEurope             
         10 CityOfLondonEngland           
  .   %/output

Google, Bing, Wolfram Alpha, and Siri all parse that query properly and answer it perfectly. Do you have an example of something Cyc does that none of those services do?

Cyc is used in the MathCraft application, which is currently in beta:


I don't think Google, Bing, Wolfram Alpha, or Siri can be used for applications like this.

Wolfram Alpha + Wolfram Mathematica certainly could.

And it's worth noting that Google and Bing generate that answer without human intervention, so no complex data engineering etc..

Not really, Google Knowledge Graph is an ages-old project and the side-panel "fact sheets" have involved manual curation of the automated results.

There's manual curation of the schemes which map some human-eneterd data on web pages to facts, but the extraction is automatic.

See [1], noting that human annotation is responsible for 250K of the 100 million high confidence facts in the system.

I believe they have some manual overrides in a few cases, but the system itself is automatic.

[1] https://research.google.com/pubs/pub45634.html

Do they really though?

That is exactly replicable in SPARQL against DBPedia (and probably Wikidata).

It's also fairly useless in that it doesn't give the "commonsense" answer of London as the first result and commonsense is what Cyc has always been supposed to do. (And yes I understand that CityOfLondonEngland probably has a plain English label too).

City of London is one of the central boroughs of London. It's basically the historic city, now turned into a financial district. So I wouldn't say that's the correct answer either.


Hmmm. The "City of London "is the "square mile", the but around St Paul's. Technically the 2012 Olympics were held at Stratford, what was once known as Stratford-atte-Bower.

Although London is good enough for me. William Shakespeare would be spinning in his grave.

> Technically the 2012 Olympics were held at Stratford

To be pedantic, only some of the Olympic events were held at Stratford. Horse dancing was held in Greenwich, for example, and many of the over-water events were held on Eton Dorney ("bloody miles away from Stratford").

> 4 ContinentOfEurasia

> 9 ContinentOfEurope

> 10 CityOfLondonEngland

Given it also has 4 & 9 I wonder if that's more of a tag than the name and it just unfortunately lines up with an actual place name. Don't know Cyc to check but I'd wager that if the location was City of London it would output CityOfCityOfLondonEngland.

Could you email me so that I could learn more about providing this? daniel@aigrant.org. Thanks!


Do you know if there are any open-source projects that are similar?

Doug Lenat (who started the Cyc project) says Cyc is the "last man standing" of the large logic-based AI projects. So I don't think any open-source projects that are similar to it exist.

If aigrant.org was able to provide accounts on a Cyc system to community members, would you be interested in having one?

Just read a twitter thread about how American culture is negative about everything. https://twitter.com/Noahpinion/status/981341257124397056

And then I found comments on this. Most of them are very cynical.

These are guys who want to give money with no strings attached and its so hard for everyone to even believe something like this is even possible.

I'm more interested in reading thoughts of people who were previously part of aigrant.

Please don't get discouraged, there are several people like me for whom the money and the other credits are very valuable. These are definitely not nothing. Even though you can get cloud credits in other ways, they are not that straight forward. From what I see they are offering, everything is valuable.

"I was accepted by YC when I was an uncredentialed 18 year old. It changed my life. I’ve since been fascinated by systems that find promising outsiders and bring them into the fold. YC does a great job at this for companies. I’m wondering if it would be useful to do it for research ideas as well."

What I would like to know is, from the previous batches, how many do you guys think that fit this category.

I applied last time, but I don't know if I would apply now, mostly time constraints.

Complete bobbins...avoid avoid avoid

Unless some of the existing fundees come forward with verifiable credentials I wouldn't give this a moment

This is a fantastic idea, and I think I’ll definitely be applying! Have you considered offering two tiers of grants? Most AI side projects don’t necessarily require ~$30k in resources, but I’m sure many people, including myself, could benefit from ~$3k in GCE credit. Opening up the same total resources to more people would allow for a greater number of moonshots, which are what this grant seems to be about. :)

Great idea, and something we’re considering for future batches. We may even do it this batch, depending on the final mix.

I could use the cash and some of the credits, but most of it wouldn't be used. But maybe I'm one of few doing AI projects not involving a deep neural net? Hehe.

>maybe I'm one of few doing AI projects not involving a deep neural net

You're not. While progress in machine learning is astounding and is going to do great things for humanity, it is kind of frustrating that "AI" to many people these days means neural nets, deep learning, high-end GPUs, and huge datasets. Not every problem needs machine learning as a solution, and there are still some of us solving hard problems with classical AI and old fashioned data analysis. But it's not trendy right now.

What kinds of problems are best solved by AI, but not deep learning/neural nets?

Natural language processing is a big area where classical AI is heavily used and machine learning is lagging behind. ML is terrible at teaching computers to read and/or produce human writing. Compare what the company Automated Insights is doing for news articles versus neural networks writing books and movie scripts (with humorous results).

Anything with a known and well-defined set of rules can also be programmed more efficiently with decision trees/Markov chain than through machine learning. There's no need to use tons of data to train a model to do something over time when all it needs is the rule book.

Machine learning is basically advanced pattern matching, neural nets often require you to brute force the problem with tons of data. There are plenty of places where you're not matching patterns and you don't have tons of data but the computer still needs to make a decision after analyzing the data that is available.

Of the finalists:

> Zbigniew Wojna (co-author of Inception-v3, one of the first better-than-humans perception models), object detection and instance segmentation for small objects

This looks really important. At the moment there's no way that Tesla's can avoid small animals and birds. This means that both the animal gets killed and it's likely that the human will crash after from the panic of hitting an animal.

The list in general is interesting as it shows what people are working on, so it's insight into current research even before a paper or results are released.

It's incredibly dangerous to brake, swerve, or otherwise alter your course for a non-life-threatening object in the road. People have died from this, and people have gone to jail doing this [1]. If people are likely to crash from panic after hitting a small animal, they're more likely to crash when the car swerves and then can't regain control afterwards. When you're about to hit a deer, the advice is to brake hard, but never swerve. Because you're never going to keep control of your car when you're both braking and swerving, and if you hit someone in the oncoming lane while swerving, you are at fault.

I don't like it either, but occasionally hitting animals crossing the road is the price we pay for millions of miles of high-speed pavement crossing the country. Hitting small animals is the best outcome in that situation, as opposed to hitting larger animals like deer (where the driver can be killed or seriously injured too).

Please don't brake or swerve for small animals. You could very likely end up killing another person in the process.

[1] https://www.usatoday.com/story/news/world/2014/12/17/canada-...

Precisely why we want collision-detection to be able to spot the animal and brake earlier than the human could.

I agree swerving is a problem, but I see no problem with braking. If you brake and the car behind you crashes into you that is their fault not yours. Cars should be maintaining a 3-4s gap precisely because the car in front could brake suddenly at any point.

Agree. And I find it surprising that the women in the article was sentenced to anything. It is certainly a moronic decision to stop on the motorway on the wrong line to protect ducklings, but what about a motorcycle that hits a static object? If it didn't have the visibility to have the time to stop then it was going too fast.

In a lot of places, there is a minimum speed on the freeway. In a lot of places, it's illegal to stop on the freeway. You may see cars broken down along the highway with tickets on the windshield and then getting towed after a period of time.

Freeways are very safe as long as traffic is moving at a more-or-less steady speed. That's why the speed limits can be so high, higher than would be safe if you needed to regularly stop. And people are not used to stopping on the freeway. It takes a long time to come to a stop from 70, 75, even 80mph, and if you're not expecting to have to stop, it can take longer than normal to realize you have to stop. Even in easily-recognizable traffic jams, many people have to emergency brake because they don't realize how long it takes to stop when you're going at freeway speeds.

Freeways are very dangerous places for stopped traffic. It is your responsibility to keep moving in traffic, and if you have to stop, to make sure you stop safely and only when absolutely required.

It was not absolutely required for this woman to stop, it was illegal for her to stop, and this terrible and unnecessary decision cost someone else their life. The fact that the motorcyclist also made a bad judgement call doesn't negate the fact that this driver made a completely unnecessary and illegal decision that cost the cyclist his life.

That's why I linked to an article where a woman went to jail for stopping on the freeway for a small animal. There is a problem with braking, and it's not always the fault of the car behind you if they hit you.

The law disagrees with you, and that's what matters.

> There is a problem with braking, and it's not always the fault of the car behind you if they hit you.

The driver is too close if they cannot stop in time. It's not complicated. It is always the following driver's fault, unless the car in front has defective brake lights.

You guys keep saying that like your opinion on the law outweighs that of a judge.

It obviously is not always the following driver's fault, or that woman wouldn't have gone to jail. You're wrong. It doesn't matter your argument. I provided proof that you're wrong and you're still arguing your wrong argument wrongly and continuing to be wrong.

You can argue with me until I give up and then you can declare victory but you're still wrong, and not only that, you're completely oblivious to the idea that you're wrong. The fact that a woman went to jail for literally this exact thing proves that you are wrong.

Hopefully that makes it clear, even to those of you who didn't read the link that I took the time to find and provide to you. I'm not trying to be a jerk, but you're so obviously wrong and yet still arguing against the law that I feel like I have to make it very clear that you are wrong. Go back and read the link I provided. You're wrong.

She wasn't braking. She was parked. The car was stopped, there were no brake lights. She had no hazard lights on.

The article is about a woman literally parking on the freeway and trying to ferry ducks off the road.

There is a significant difference between that and braking sharply.

'in minutes' lol. Read the fine print folks!! This ain't no minutes.

Perhaps they're over stating it. But it's still probably an easier/faster way to get funding than traditional research grants.

To me it wasn't the time commitment, it was the deception. Felt like a sweepstakes eventually because they are not upfront, and you give very private information. They know what's up with how it was designed and marketed (headline) and if they don't I humerously question whether they should be doling out AI grants.

Thanks for the feedback. Will make the upfront time investment more clear.

minutes = more than one minute

Before I fill the form, can you please give a quick disclosure about any other interest that you have in this, if any, specially about the data. There are lot of threads about some suspicion here, but I can't find any direct response.

Sigh. Hacker News is a suspicious place these days. We have no rights to anything you produce, or datasets you use. We just want to help budding scientists.

We are asking people to do a 30 minute quiz once the application is filled out. It’s 100% experimental, but the long term goal is to try and see if there are any ways we can scale this program one day. For example: if it turned out that all the Fellows we funded did terribly on the test, we’d know the test is bad. If they did really well on one question, we’d know we found something that would help us find more people in the future.

We’re not going to sell the applications in any way. Please email me if you have any concerns: daniel@aigrant.org.

> Sigh. Hacker News is a suspicious place these days. We have no rights to anything you produce, or datasets you use. We just want to help budding scientists.

It's not HN, it's the world. We live in a society where you have to beg and scrape for healthcare and a job and shelter. Then when you have it, the only way for most to keep it is to sell their labor to a corporation, who doesn't pay you for what it's actually worth (they take a profit). Then, when you go home to improve you life on your own time, the corporation tries to claim ownership over your ideas you work on in your spare time.

What you're proposing, "no strings attached" is counter to almost every expectation we've had our entire lives regarding "a free lunch". i.e. there's no such thing.

I am so sorry to come as rude. But my question is least about reminding you the things that you are doing wrong. I wanted to ask you about the involvement of YC/Google/any third party, and if there is some expectation of them from this initiative.

I really believe that you are doing this for the best of reason and want to see progress in this field, but there are other parties involved here.

Just finished questionnaire. The "brain teaser" at the end is full blown IQ test. Which got me an idea. I know its all nice to pretend that everybody is equal and all that jazz, but if you want to make real scientific progress and not just let some code monkey glue couple libraries together and fixing problems on stackoverflow, you need someone really smart. And IQ test is good filter for that. So if I would want to grant money to random people and I have 100'000 applicants, I would just sort them by IQ, keep top 1% and manually sort them and pick a few. If that's your strategy then I'm screwd, I was always too slow on IQ tests.

> If that's your strategy then I'm screwd, I was always too slow on IQ tests.

Right, this approach would just select those people who are good at completing IQ tests. Which is a pretty common criticism of IQ tests.

That's a common lay criticism of IQ tests, but it's not at all common among clinical practitioners or psychologists. It's also completely at odds with the outstanding literature on the subject[1].

There is broad clinical and professional consensus that IQ tests are both statistically reliable and valid for a number of generally useful measurements of cognitive ability.

I think you can make a compelling case that IQ tests can be misused as a hiring filter - for example, you might not need to hire people who score highly on it. But as a standalone assessment, modern literature doesn't leave much room for debate on its psychometric merit.


1. https://www.apa.org/pubs/journals/releases/amp-67-2-130.pdf

Why wouldn't the ability to quickly process information be considered an important part of intelligence? A proper IQ test is very general and just tests processing speed and working memory.

IQ has a higher correlation with future income than parent's earnings. If you had a choice between having 99th percentile IQ and being born into the 1% you'd be better off taking the IQ. The issue is that because IQ is highly heritable many in the 1% tend to be intelligent as well so the rich get richer.

I don't know about you, but if I had the choice between having a 99th percentile IQ and being born into the 1%, there is absolutely no way I'm taking the IQ. The streets and fast food franchises are littered with exceedingly smart people. Dumb kids of the mega-rich are still rich. A lot of things need to go right for success, those things have a better chance of going right if your parents are loaded.

It's important to note here that we're not discussing the well-off, two vacations a year types here, we're talking about the 1%. I would be extremely surprised if the correlation between your future earnings and your parents' earnings (if they are 1%) weren't -a lot- stronger than your having a 99th percentile IQ. Any sources?

> IQ is highly heritable many in the 1% tend to be intelligent as well so the rich get richer.

I'm sorry, do you have a citation for this? I find this highly suspect. Everything I know about money and power and society suggests the reasons the rich get richer are money, power, and society. Nothing to do with IQ. This idea that the poor are poor because they are dumb is something rich people (who don't want you to be rich) want you to believe.

Well, it's rather obvious that an IQ measures how good you are at completing IQ tests. The question is whether IQ is good predictor of success in whatever kind of work you are doing. If it is, then I can't blame them for making you take the test.

While smarts are generally useful, is not creativity a stronger influencer in innovation? But I’m not sure what kind of test could measure creativity.

Creativity, and grit. Sometimes provided by different people.

As long as IQ only means how quick you are able to understand something and not how motivated / how much stamina you have, sorting by IQ is a stupid/ignorant idea.

That doesn't mean sorting by IQ is ignorant. It means that you shouldn't rely on it entirely. Hypothetically speaking, unless there is literature indicating that a population of high IQ individuals is more highly correlated with lack of stamina or motivation than the general population, sorting by IQ alone as a first pass would still provide you with plenty of candidates who have the motivation and stamina you're looking for.

I think there are credible arguments against using IQ for various sorts of job filters. But I don't think your argument is among them. If we're at the point where we're implicitly taking that an IQ can sort people by "how fast they learn something", the only reason not to sort for that (if that's what you're looking for), is if that quality is more correlated with something else you don't want. I'm not familiar with any research suggesting that high IQ individuals are more prone to stamina or motivation problems in a large population.

Which is precisely why we’re not making decisions based on that test for now. We’re just experimenting with it to see what data we get. We’re still going to grade applications manually for now.

We’re trying not to fall into the “just IQ” trap, but learn the broader picture of what makes great Fellows great.

>And IQ test is good filter for that

[citation needed]

I am cynical. What's your angle? I presume that the idea here is to find open source projects to become heavily "invested" in, and then find ways to extract out a commercial offering, whilst engaging in (free) open source labour? Sorry for the cynicism, but offering only $2500 (+things of variable value) sounds like you are trying to target desperate people to take advantage of almost. Ie. If an open source project grew a commercial element, it's going to be worth a bunch more than that. Please, I stand waiting to be corrected. Thanks.

Hi laythea! That is an incredibly cynical take. Why are you so default cynical? Anyway, I’ll share my hidden motivations with you, if you like. It’s twofold:

1. I am intellectually curious about AI research. I enjoy talking to researchers. I do it in my spare time. For fun. I would like to do more of that.

2. I was accepted by YC when I was an uncredentialed 18 year old. It changed my life. I’ve since been fascinated by systems that find promising outsiders and bring them into the fold. YC does a great job at this for companies. I’m wondering if it would be useful to do it for research ideas as well.

Lastly, I completely disagree that GPU training time is “of variable value”. It’s a big deal for many researchers.

Hi, thanks for your reply. Yes, sorry for sounding a bit cynical, I will take a look at what you do in more detail, but the sponsors of this venture surely want some return? My, as yet uneducated, layman outsiders guess as to what that return may be would be some IP, derived from these projects? Kind of like growing a tree for birds to lay eggs in and then shaking it to see what drops out. No offense intended. Cheers

It’s a non profit! The IP you develop is yours!

Charities can and do steer activities in a commercially beneficial manner to suit certain stakeholders. I suppose, non-profits are similar.

>> I enjoy talking to researchers.

In that case you should be used to the cynicism (although it's usually called "skepticism" in the field).

I am also curious about the CrowdFlower connection.

I am one of the AI grant fellows, and I have to say that they never asked me any thing in return - just our opinion and some help regarding the AI grant. I personally think the AI grant is particularly interesting to get some extra visibility (not only the money)!

BTW: our project is already available out there! :) https://datasets.freesound.org/

Is there any solid evidence of this being legit? The web sites [1][2]seem kind of thin, and maybe the whole game is to collect a bunch of contact info for AI researchers. I imagine one could sell that list to recruiters.

[1]http://www.cyc.com/ [2]http://aigrant.org/

No! I don't know anything about Cyc, but AI Grant is very real. I run YC AI (http://ycombinator.com/ai). Nat is the VP of Developer Services at Microsoft. We're not doing this to sell info to recruiters. We're doing this because we're passionate about AI research. And because we think the best ideas often come from the fringe. We want to encourage others to take their side projects more seriously.

Feel free to message any of our existing Fellows and ask about their experience.

Pls consider changing the 'grant in minutes'. I am busy so I read the headline and jumped right to the page URL. It said nothing of the substantial requirements and commitments both data and time. That should be spelled out up front. This felt very un-comfortable, to keep being asked for more. If nothing else, spell out what you want upfront.

- Application form - 100 question personality test - 10 question spatial test prelim - 20 min spatial test - who knows

The website is very light on contact details. Specifying things like physical addresses of where you're registered/located are fairly simple things to differentiate yourselves from scam artists.

Sorry I voiced that musing. Guess I was just being true to my username.

Nah, these days, you can't always be sure that someone's running a grift. The internet sure as hell makes it easy and approaching $0 to scam.

And being cranky as well, I get it. But this is legit.

Wouldn't the data be covered under HIPPA (the data regarding personality could be construed as medical information, therefore opening this up to federal criminal charges?

Yep, I vouch for Nat Friedman.

I applied last time it came up, and had a good discussion that how I wanted to approach wouldn't quite have worked within the scope of the grant. Certainly no hard feelings at all (it was an oddball 3d scanner, like really weird). I felt that Nat was respectful and sincere. I never got even an inkling that this was some scraping scam or something.

Tl;Dr. I've seen a lot of scams through the years. This certainly wasn't (and isn't) one.

Dani, thanks for doing this and I like your idea of building an online community working on interesting side projects in AI. I would apply this batch and recommend it to other researchers who are working on cool stuff to apply now, "only" if you take our feedback, which you asked for, into consideration which are:

1-Delete the personality questionnaire and the brain teaser. Keep it simple as you promised above "in minutes". Seems like everyone here hates it, including me.

2-Either in the application process or on your website, state how would you protect my idea. Ideas become things. In the application, I will give you my full idea that I am working on. You might get inspired by it, and build it your own. If you are really trying to attract top-notch side project, you need to do this.

How often are these running? I'm currently taking the Udacity Deep Learning nanodegree and have some ideas I want to try out once I'm done. Having credits to train and generate labeled samples would be really helpful.

We hope to do a few more, but don't know exactly when. When in doubt, just apply! At the very least the process will sharpen your thinking on the project you're considering.

Why limit yourself to AI? And what exactly qualifies specifically as "AI"?

Focus helps one make a great product (for example, domain-specific content and community). In time, we'll expand. If this strategy works, it can be an accelerant to almost all areas of research.

We're liberal with the definition of AI.

This personality test is interesting. I can see why YC would want this data.

Does this apply to work that is deeply theoretical as well? (e.g. studying/experimenting with tiny Turing machines in an attempt to approximate AIXI-like algorithms)

As long as you can articulate why your project would be useful, we're interested. Theoretical or not. It just needs to be useful to other humans. Or AGIs. (Kidding. Mostly.)

Build us an atomic bomb and get $2,500.....

AI is dangerous.

If an AI becomes aware, what happens then?

I used to think that it was insanely unlikely that an AI would become aware. But after many deep thoughts on what consciousness actually seam to be. I believe it could happen at any time now just by random chance. We are closing in on the number of neurons that i believe is required.

1. I'm not sure we have properly simulated even a single neuron yet. There are no signs of anything approaching general AI yet.

2. If a human is aware, what happens then? Not much.

What is a awareness and how will you tell?

Why is awareness necessary for an AI to be dangerous?

A swarm of armed, autonamous, UAVs is plenty terribad already, and Google's making them smarter. It doesn't take much smarts to "find squishy warm mobile things and put missile there".

> What is a awareness and how will you tell?

I have no idea :) But i am Aware and dangerous, but if that awareness was taken from me and i became unaware a mouse could kill me.

Not saying that awareness (whatever it is) is necessary.

What exactly was your thought process between "It's unlikely that AI will become aware" and "It could happen at any time now just by random chance"?

First it was like nah it will never happen "we don't even understand how our own consciousness works, or even what it is."

Then it became "we don't even understand how our own consciousness works, or even what it is." So it could be anything. Even a 302 neuron network like the one in the C. elegans worm could be conscious from what we know.

It was more of an instant realization. A realization about how little we actually know.

We need grants based on the outcome of the project, not based on what technology is used to achieve it.

I don't think it's 100% of either of these (outcome vs tech used). If its true research, the outcome isn't known. There are major benefits in developing technology for it's own sake, which then provides tools that can be used towards other applications.

This is awesome, I love the idea that you can be working on a side project and still get support.

This is awesome! The apply link (https://apply.aigrant.org/) is not working for me however.

Sorry, our DNS provider was having issues. Please take another look now. Thanks!

No problem, works now. Thanks!

Has CrowdFlower renamed? Their website now redirects to https://www.figure-eight.com/

Yes, they are now FigureEight.

The description and intentions seems disingenuous from what I see and what others have commented. Please flag this.

I would apply but I'm not sure how that will be considered as I'm on a work visa :(

You can always decide not to take the grant if you're accepted. At the very least, the application process will serve as a potential way to validate your idea.

Good to know, thanks!

I am not a huge fan of NDAs but I wouldn't give your idea away to someone just for validation.

Not just the idea, but a large chunk of personal data via the quizzes.

In fact, coming right after the Facebook/Cambridge Analytica "Personality test" news, I wouldn't be surprised if this just a large Honeypot/social engineering demonstration to demonstrate how the HN crowd isn't immune to the same tactics "Take this short quiz and win $XXX".

I, for one, am a believer that ideas are a dime a dozen.

I actually had my idea stolen, by a better funded company (yes it involved AI). But that made me even more of a believer.

Even though this other company had far more engineers & was able to replicate a seemingly copycat product after I refused their buyout offer, their execution sucked(which was a key reason I refused, rather die broke than work for a company that think engineers need daily beatings to be exploited properly by great “business” minds)

However, they were VC backed & their board made them shut the product down and we ended up getting all the customers they brought on in the copycat product.

The people behind AIGrant.org have solid geek creds & given their backgrounds can probably come up with more creative ideas than trying to scrape ideas off the web the offering large bounties.

What’s really awesome to witness is that here’s a coupe of geeks actually “executing” on their vision. I’m a little jealous because I should’ve stared something like this, but instead I’d rather just support AIGrant.org because building & executing even benevolent ideas take a ton of effort.

Shallow thinking. I personally believe the idea has value only in relation to how well it can be executed upon. If can be built easily, then it will be copied anyway.

I agree that it is shallow thinking, but this is some random application for the potential of a small amount of money. Some one else can validate the idea easily

Alright, so I applied the other day. Just one comment on the process the size limits on the answers are way too small.

So much for "in minutes"... "Applications take days to complete."

Cloud credits are easy to get. But you are giving away cash - what's the catch?

I am one of the grant recipients. In my experience, there really isn't any catch to the sponsors giving away cash. It's simply to help further the ongoing research and it has been one of the easiest and fastest ways for me to get resources for my own research. Also, it has helped me connect with other grant recipients who might be able to offer advice, IF I need it.


Yes, it's frustrating - but if the predominant response is doubt and confusion, you guys clearly didn't think this all the way through. AI tech is now easy and accessible, and you need to make a very good case why an AI startup would "need" you - along with the platforms you're promoting.

How many grants will you be given out this year?

How many hours are there in a week?

no confirmation email after applying?

hello! I had the same concern. I saved a PDF of my application just in case. Please let us know if there's anywhere we can confirm receipt of the application :-)

I copied my responses to a .txt file before submitting just in case

hmm.. unless the sponsors publish the names of successful recipients, there is nothing to say this is what it is presented as.. (I did not visit the site)

Try visiting the site, I think you'll be pleased!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact