There's plenty of places to play chess already so presumably the main idea is to show off the AI generated pieces and default to the 3D view because it looks cooler.
so when I first tried the demo I wondered, “are these pieces generated meshes?” thinking about the ability to zoom / pan the board in 3D. The high fidelity of the pieces would seemingly be more in the domain of 2D image generation given current gen AI standards, which would be my educated guess as to why there’s only a handful of views.
I don't know how Google is internally organised, but maybe Google Labs does not have a dedicated legal team. And out of fear for possible legal implications, they just block it in EU.
It's quite understandable, because the relevant legislations (AI Act etc) are all very new, not been tested in courts and due to the lack of stare decisis, you can't rely on a uniform application of the law in all jurisdictions, so they just block it.
Frustrating, but I prefer they move fast rather than safe. And from their perspective, it means not releasing it in the EU.
You are generalizing extremly without any substinance...
1. its Googles decision to not allow this demo globally. Its a Chess Demo, what EU legal issue do you think makes this a problem? Probably NON...
And yes we are not America. Our privacy is getting better protected then that of americans. Move if you don't like it or at least take the time and effort to state clearly what you don't like about certain EU regulations.
When Meta prevented the EU from using meta.ai or even downloading its vision models, I sunk my head in the AI legistation.
Here, I am honestly not sure which part they rely on, to say that what they made might be unlawful.
The closest thing I found for Meta was that “emotion recognition systems” are classified as high-risk (paragraph 54), and high-risk systems must have their training data disclosed (Art 11(1))[0]. In theory, you could upload photos to meta.ai and ask it what emotions are displayed, but it is already a stretch. For GenChess, I’m at a loss; it doesn’t sound like you can do that. (Not that it prevented any vision chatbot from releasing.)
If someone has a better guess as to why they might have restrained it here, I am curious.
It's simple. Big tech doesn't like regulation. By pretending the regulation won't let them release "all these cool things" they can turn public opinion against regulation.
Nothing in the law requires a banner. It could even be handled in the browser by letting people choose what third-party cookies to accept (or none, hence the problem) and having that be negotiated during page load.
It's nice the law is being interpreted to require to be as easy to reject all local storage of other's data as it is to accept all local storage of other's data.
I'm allowed to ask for all of the data they have from me. This obviously includes all chess games which obviously is a pain in the ass for a dumb demo that will no doubt be archived next week in some obscure place.
What bullshit? Protecting the privacy? Making sure the playing field is at least somewhat level?
Sorry, but I'm a big fan of that "bullshit". And the big American companies are not, and they are making sure you know it, by pulling things like this.
Not available in the EU - so I guess the question is what are they collecting off you under the guise of playing chess that would require them to block EU users...
It’s probably not nefarious and just generally not worth the headache of an EU release and going through the checks/requirements for an experiment or early beta.
The many rules of the EU stifle speed and change the math for releasing something especially in a big company that has a variety of requirements built up over time to reduce litigation risk or being on the wrong side of one of the overreaching government officials there.
Typically the internal processes to verify what is/is not collected are mandated by launch region, not whether or not the initial version of the software does data collection.
At work all our products do not eat children by default and it would require legal review to add any such feature. Therefore we don't need a procedure to make sure that any child-eating components get disabled in regions that have some concerns about child-eating.
Except too bad you're launching a washing machine which needs to be checked for child-eatingyness on the tiny island of Zogdog, so you just decided not to sell it there.
> At work all our products do not eat children by default and it would require legal review to add any such feature.
I bet the people at Peloton thought that too until they made that treadmill[1]. I know you meant your critique to be absurd, but it turns out creating a child eating machine by accident is entirely possible. I also bet Peloton product development now includes a process to review child-eatingness despite that not being their primary market, just the usual twice burnt reflex.
Accidentally logging the PII can easily happen for a single engineer. I managed to do it on an product that was privacy-focused & the error slipped through review. The odds of such inadvertent errors rise linearly with the number of products and engineers, and the fines are probably superlinear with the size of an offending organization. If your 3-person consultancy chomps on a GDPR baby or 2, no one will ever know about it, but if Google does it, it's going to be news headlines an millions in fines.
Logging some PII by accident is also an issue and can also lead to compromises, but I think intentionally collecting it in bulk is the primary concern here. With google I assume bulk data collection is their default stance and that's why they need to carefully trim it down just enough that the lawyers say is justifiable rather than the other way around. That's the problem I'm gesturing at.
This mindset assumes that you know what question you want to ask before you have the data, as opposed to having data and then being able to generate hypothesis from it.
Legal basis is ever shifting based on the regional locale. As n+1 requirements pop up, it's only natural to release things like GenChess in the place that requires the least friction, especially when it is not a revenue generating event.
I suspect it’s a byproduct of Google’s internal launch process which may require more work or longer processes for launching something in different jurisdictions. So it’s probably not an active decision that they couldn’t legally do this in the EU, but a result of being extra careful around what they can “launch” in a jurisdiction with potentially high penalties.
(I am a Googler, but not on this team, or familiar with their launch policies)
I think one of the least wise things a person (or company) can do when faced with any law is to assume that it's "not complicated really."
Much, much wiser to assume "there be dragons" and only engage once qualified legal counsel has helped you understand what compliance means to you.
And along these lines... The second least wise thing to do in this scenario is listen to randos in a forum like this tell you, "but all you have to do to comply is..."
The problem with thinking like yours is that legislation like GDPR is _really_ made to be simple and straightforward, but since companies whose livelihood depends on them abusing your privacy will fight it tooth and claw, they will gladly make it look like it's more complicated and insurmountable than it really is. They will also devise ways to comply in such ways that's most cumbersome for the end user and will readily blame GDPR for it.
To devise such a way to comply, they definitely need a large and expensive legal department.
The privacy abusers are much like trolls on the internet who, upon seeing a code of conduct (previously known as "rules") consisting of only "don't be a dick", will spawn endless arguments about what a "dick" is and how it is or is not inappropriate word, what does it really mean to be one, or, indeed, to be, question the use of the indefinite article, and complain about "don't" being too assertive and arrogant.
> ...they will gladly make it look like it's more complicated and insurmountable than it really is
There are non-malicious explanations for the same pattern of behavior at large organizations - the motivation (malice or not-malice) that seems correct is a Rorschach test.
If I accidentally log IP addresses for EU users that opted out on some throw-away experimental page on my site, Brussels would never find out. If Google does it, it not only has to report the incident, but will most likely be fined. In order to avoid this outcome, they have internal review processes which makes ot "complicated and insurmountable", because how do you justify investing many hours of dozens of lawyers and technical reviewers time for a frivolous, niche AI demo?
> If I accidentally log IP addresses for EU users that opted out
If opting out is a possibility, then it's not essential data. If it's not essential data, what legitimate reason would you have to collect it in the first place?
Data safety is a not a binary thing (essential = okay to keep, non-essential = not okay). Instead , it is a context-sensitive, layered, multi-axis concern. IP addresses, for instance, are PII that are essential for network routing, but that doesn't mean address+based ad retargeting is fair game.
Accidentally logging data that's essential but was supposed to be ephemeral would turn it "not-ephemeral" and make it possible to cross-reference with other information in privacy-defeating ways. As I said, it was a privacy-focused product, and using data that is meant for one thing (e.g. abuse detection and prevention) for some other out-of-spec use - even accidentally - is a big no-no.
Ok... Let's assume this is true (which I'll reiterate, that I contend assuming so is foolish). What happens when courts have interpretations of this "simple law?" Do the courts make an effort to keep things simple and in plain language? Or do lawyers and bureaucrats do what they can to drive unintuitive interpretations, but favorable to their cause, of otherwise plain language? Are European laws such as this subject to the interpretive lens of case law? If so, the best intentions of legislators may only be secondary relative to the actual rulings and unintended consequence of their laws. The problem with thinking like yours is that it dismisses all of this messy reality in favor maintaining the idealism that might have motivated public support of the law.
Those that have to follow those laws need to care about the mess.
Second that. GDPR actually made those aspects clear and never caused a headache during implementations I've seen or participated in (more like a checkbox on a list). When I see any complaints, then it's clear some iffy user sniffing is happening.
For an experimental project every extra requirement is annoying/slows down release. It’s completely reasonable that they don’t one to add more work before releasing to a large customer base that doesn’t need the extra work.
From my experience helping my company with GDPR, IMO it's true that the principles of GDPR are straightforward. But there can be a fair amount of ambiguity in how certain parts are interpreted, so in practice if you're taking it seriously (which every company should), you'll want to loop in your lawyers. Then there are more and more conversations to make sure everybody understands what the company is doing and what their stance is on GDPR.
Sadly, GDPR is not a black-and-white (pun intended with the chess project) checklist with black-and-white checklist items.
Doesn't the EU also have an 'AI Act' that imposes additional rules, even when you're not tracking anyone?
And a lot of employers have legal teams who are extremely risk-averse, so even if it's obvious to you and me that rules about "deepfakes" don't apply to a tool for generating pictures of chess pieces made of cheese, doesn't mean legal will sign it off.
Not sure it is a GDPR issue – it isn't filtering my out on either mobile or the office connection. I'm in the UK, and despite brexit we are still (technically at least) covered by GDPR (there are some differences in UK-GDPR, and more will come, but IIRC they are not significantly substantive yet). Unless, perhaps, they are banking on our ICO being toothless so won't enforce anything.
Limiting access for EU users is an easy way to sidestep a hell of a lot of potential legal liability. It’s a fast-moving and complex regulatory environment that is focused on punishing/controlling big US businesses.
Option A is to burn a pile of cash and time to keep abreast of it.
It is very annoying at most big techs to go through launch approval for something all new and adding yet another approved or checklist for the EU is not worth the tradeoff of time for something experimental/early.
That department becomes the overhead every time since it imposes requirements or work on the team doing the actual creation. They make the teams building the product make changes or answer questions and “gate” releases. Every time you add someone extra things slow down if their whole job is related to reducing risk. Misaligned incentives.
I have worked at multiple of the FAANGs. I know quite well how much overhead these kind of releases take and how much overhead any centralized release team made to “help” end up adding. It’s not just a few small things that you mention.
> Limiting access for EU users is an easy way to sidestep a hell of a lot of potential legal liability. It’s a fast-moving and complex regulatory environment that is focused on punishing/controlling big US businesses.
Have you considered a less paranoid explanation, like... focused on protecting Europeans from the abuses of big corporations?
I wanted you and your sibling commenter to know that even though we're probably coming from opposite ends of the political spectrum, I meant no malice or schadenfreude by my comment.
Despite being a huge capitalist, I do think there's a sweet spot of very minimal regulation that works out best for everyone.
However, the blocking-the-EU outcome is obvious after a couple of minutes of thinking about it in game theory terms. Another one to look out for is companies (both physical and digital) that never even considered doing business in the EU, but who would have if it were easy.
Pretty cool, and also pretty fucking bad. Checks all the AI boxes, I guess.
I tried "a classic set inspired by Armored Core". It gave me a typical-looking Armored Core (which would be good except it was supposed to be the king), a Space Marine (which I rerolled into some small mech that's close enough), a Zoid (which it refused to budge from), a tower (which it refused to budge from), and a chess queen with purple lights (which it refused to budge from, except to sometimes give me the gigantic head of an AC).
Not to mention, sometimes pieces look directly at the camera instead of forward, or don't even sit on the plane of the board. One of my pieces even had a mini-chessboard for its base.
I asked it to draw me a creative chess set inspired by copyright violations. The queen is carved from what I can only describe as judicial oak, wearing a Jesus-style crown of thorns and being grasped from behind by a creepy hand that's also holding what appears to be a large banknote.
Look how easily they just fabricated 100% of a game's assets, with a consistent, high quality style, and then put them right into the game environment. That's the takeaway IMO. Very tight GenAI loop.
I‘m not a game dev so I might be missing something, but this hardly looks high quality to me. The set I get is very inconsistent and lacking in any central theme, or otherwise interesting or clean design. If I regenerate one peace in isolation, say a knight, I get something completely different that looks as if you lost the knight from your set and just took replacement knights from another set at random.
The time it takes to generate a set is also significant, as a web-dev this takes for ever, and I would be very reluctant to offer this experience to my users. Doesn’t feel fluent, nor tight at all.
Plus a silly mistakes like the knights facing the wrong way, different sizes, etc. Seeing this, I certainly hope game designers (at least in online chess) will stay away from generative AI, for a while at least.
I mean I love it but it's not exactly a thing that needs a proof of concept, and is more than a little surprising to see google still having fun with such a small toy! Maybe that's the better takeaway, google labs is allowed to have fun again.
So? Any video generation model must necessarily be able to do this. (consider the case of generating a pan-over of a chess board where the starting input frame is only the first pawn and rook, the model should know to generate the rest of the pieces in the style of input pieces)
Oh yeah, that's confusing wording. I just meant it's a simple image, not animated, no additional views of it.
I feel somewhat bad about my comment now though, it's delightful to play with something you made and that's the point, and I'm glad google is able to ship small fun demonstrations of stuff like that via google labs.
I hate how hard it seems to be for to follow a thread of replies on here, but it feels like you're just replying to my comment without having read what it was a response to.
And I know I'm in the weeds by replying more.
But damn it, my reply was to "Look how easily they just fabricated 100% of a game's assets" which is a statement that really implies a lot more than generating 12 images! Chess is a game with 12 static images. You can call these dynamic, that's fine, but the context matters.
It's that and someone's toy Next.js project. The chunks from Next.js's code splitter aren't the only chunks involved in the game though - the gameplay is horrible.
I'm fascinated by what prompts lead to 'Please try a different prompt.'
New Zealand or Australia lead to the aforementioned error, but The Antipodes generates a set.
I thought individual countries had been banned for possibly generating offensive content but it works for Great Britain.
Metallica works, Slipknot doesn't, Iron Maiden does, Nirvana doesn't.
Aperture Science works, so does Black Mesa (the proposed opponent, hehe), Cyberdyne Systems doesn't, Vault-Tec does, Weyland-Yutani also works.
Studio Ghibli doesn't work, Toei Animation does, so does Production I.G. and KyoAni.
Akagi works, JoJo doesn't but JoJo Bizarre Adventure does (so does Joseph Joestar, and it knows Dio Brando is the opponent), Uzumaki does not, BLAME! is accepted but not recognized (a shame), 20th Century Boys is not accepted.
Soviet Union works, Russia does not, neither do most countries. US States are accepted (California's opponent is Florida), so do many cities, with exceptions of Mexico City/Ciudad de Mexico, Pyongyang and Tehran.
Voyager-1 works, Apollo 13 doesn't (for any mission numbers), plain Apollo does (but it's not the space mission), Soyuz does, SpaceX Starship does not.
Vim works, so does Emacs, so does Visual Studio. (I've tried that just to see the proposed opponents)
Dota 2, S.T.A.L.K.E.R., Mass Effect, Half-Life, Deus Ex, Arma 3, BioShock, Elden Ring all work, even Wolfenstein does. Disco Elysium is accepted but produce something unrelated (a shame, again). Can't find a video game that won't be accepted.
Funnily, US Government worked partially for me, failing to produce a single piece.
"Warhammer 40K" generates perfectly good lawsuit material: a space orc on a bike (knight) and a space marine with a bolter (pawn) in a classic WH40k pose. Also some Eldar-like pieces I didn't recognise.
I tried entering cocaine and it was rejected, but tried coca leaves and it looks quite good. Also my opponents pieces were coffee beans which is a nice color contrast, and with that I beat the easy computer player. My first try a couple hours ago and I chose Grasshopper Pie Milkshake and the opponent it chose was Avocado Toast, and both were green, and I lost.
Holland, England and Scotland all work. So does Myanmar, but not Burma. Deutschland works, but not Germany. Probably just an overly literal filter based on a static list of countries.
I agree it is cool and fun and nice to see. But the investment here seems minimal. Preexisting image generation capabilities (with a link to Imagen-3...) plus some "prompt engineering" slapped on top of https://github.com/josefjadrny/js-chess-engine.
I was hoping that, with sufficient practice, a studious person might be able to beat a JS chess engine. However it looks like these engines are in the ELO range of 2500-3000, so unless you're a young teen with a few years to spare for improving your chess score, it's probably not possible. Even for a smart teen, it would be a stretch goal.
If that rating is accurate for these JS chess engines, even a motivated young teenager practicing and studying continuously for years STILL doesn't guarantee that they'd be able to beat it. 2500+ is the realm of GM level chess.
There's only a couple thousand chess grandmasters IN THE WORLD.
if "with a bit of practice" extends to "with a bit of research"
folks used to beat engines by playing lines that the engines could not calculate correctly. you could probably find lots options to play that give you an advantage, though youd still want a pretty good elo to pull it off
Maybe... but being able to play "anti-computer chess" (lots of subtle moves that have very small perceived advantages) hasn't been a particularly viable strategy since Kasparov's loss to Deep Blue in the 90s.
Jonathan Schrantz has videos beating stockfish (not full strength, more like the JS version) a couple years ago using specifically anti-computer preparation.
At least they're now willing to publish these kinds of fun and creative things. Which was almost guaranteed to be blocked by one of approval chains for several years.
I‘m confused. What is this and why is it impressive?
It feels like I‘m playing chess at a bar that has mixed peaces 6 different sets and I‘m forced to sit at an angle from the board because of space limitations inside the bar, but without the benefit of having a fun bar chat with my opponent during the game.
It doesn't seem to take orientation into account either, which is pretty important when creating a 2D sprite to represent a 3D object. The pieces face in arbitrary directions rather than always facing the opponent.
perhaps more of a weighting system on proper names and whether it’ll return copyrightable material, strange what works and what just doesn’t.
“lisa frank” doesn’t work, but “a clone of lisa frank” does. “deiter rams” doesn’t work, neither does “a clone of deiter rams”, “facsimile”, “design language”…
It’s an indication that most new products will (and usually do) launch somewhere less strict first. The EU can change that by cutting down on rules but we all know it’s going the opposite direction towards further stagnation with current plans.
Exactly. I'm not under the age of 18 and is located in the EU region, thus cannot use this service for no other obvious reason. I guess Google feels it is more convenient to just block EU users from services like this than it is to try to comply with EU regulations.
Remember the faceswap apps that were doing the rounds a while back?
The most egregious violators of privacy and ingestors of PII are often the most trivial of 'cool tech demos'. This being Google doesn't make it any different - particularly after their egregious violations regarding their Incognito Mode.
They're reigning themselves in before they're smacked down again for naked profiteering, not 'punishing' the EU for imposing a 'success-tax' as the reductive consensus from Americans seems to be.
There seems to be a weird relationship between what you prompt and what it actually generates.
"wonders of the world" works will with each being an individual piece type.
For instance when I prompt it with, "first generation pokemon starters" it will do an okay job using one pokemon per piece with inspiration but it's not smart enough to make the more powerful pieces the more advanced pokemon.
But when I prompt with with "types of simple machines", "branches of the US armed forces", "the wives of Henry VIII", "the original avengers", or "founding EU members" it just makes pieces that are a mismash of them instead of separating them along very obvious lines like oh here's an Iron Man piece and a Hulk piece, etc.
If I ask for trivial pursuit categories it just makes a generic board game themed set instead of following the prompt.
I generated a Rick and Morty set that was completely hilarious and played against cardboard. It was pretty confusing though because cardboard was white and the R&M set was dark. That isometric view is unplayable though!! It's hard to even click on the right piece and analyzing the board becomes confusing.
Not that there are actually legal blockers. It's just that for the team/person creating this thing, it's too much work to pesker legal to find out if there are legal blockers. Legal is going to CYA on these things anyway.
There is a group of countries where it's "probably fine", so those countries get to try demos like these. There are other countries where the policy is "talk to legal", and nobody wants to talk to legal. So those countries don't get access to the demo.
I'm not involved here, but I've seen these issues play out quite often lately.
Is this any different than prompting an image generator
"Give me a classic king chess piece in the style of <prompt>",
"Give me a classic queen chess piece in the style of <prompt>" for all the piece types?
Perhaps some tweaking to promote consistency between the pieces?
Very cool! Also makes me curious what will happen to this project next - will it morph into some other product or feature that we'll see a few months from now.
Anybody know of fun stories of what's happened to previous Google labs projects like this one?
I managed to make it generate a couple of pieces with boobs on them a single time but otherwise wasn’t able to make it do anything NSFS. I used the term “mammories” and after a couple attempts I guess it added it to a block list?
This is super frustrating to use. There's a character limit of 30 but it seems like whenever it tries to put a line break in the input it just deletes what I was typing.
At first I thought these were 3D models: they're just images with black backgrounds removed. If you look closely, you can see where the find/replace failed.
That was my first thought too, strong Battle Chess vibes!
For those who never played it, Battle Chess was a chess game from 1988. It had one different: when a piece took another piece, it would play a canned animation (gruesome, funny)
It also doesn't seem to let you generate chess sets inspired by many political figures or other popular names. While you can't generate a Kamala Harris inspired chess set, you also can't generate a Bob Ross inspired chess set...
Yes, my religion forbids me from interacting in any shape with any Abominable Intelligence. I am thankful to the Google corporation from protecting my eyes from this unholy threat.
No, AI chess is not banned here, but I suppose they - even in this app - collect every data possible and use it against the user - then yes, that is banned and this is not a fault of my country.
Noticed this pattern with Google delaying everything in EU by several months, apparently due to GDPR compliance? They also did that with Gemini when it came out.
It takes forever to get internal compliance signoffs, and typically the product has to be done enough to be clear about what will ship. Why hold it back elsewhere during that time?
All this energy being burned to create shitty images, shitty music, and shitty videos that all amount to essentially shitty memes. I don’t understand how the singular focus isn’t on solving humanity’s problems.
We already have artists, we don’t have a cure for cancer or the climate crisis.
Let's hope he is not the guy that checks the tickets for a roller coaster. They just send people from A to A using an ineficient path that burns a bunch of energy.
I hope he doesn't work at any tech company; I'm pretty sure you could say that about all of them. I also hope he doesn't have a car - that pointlessly burns energy when he could be taking public transit instead.
There’s an episode of Seinfeld that opens with some standup where Jerry is joking how scientists are working on seedless watermelon, and why aren’t scientists working on more important things. And I always think the same thing! “… so the comedian is complaining about how scientists spend their time…”
It’s a fine joke, but deep down the instinct is very top-down and suggest centrally planned economies, dare I say communism!
In some ways it does reduce the climate crisis. A person sitting on their couch browsing memes is less carbon intensive than someone driving to their friend's place or something outdoors.
Imagine putting all your work into creating an amazing demo and then drowning it with one design decision.
Edit: Ok I see that you can change the view in settings. They should make that option more visible.