The company behind Stable Diffusion appears to be crumbling into chaos
futurism.com
futurism.com
The article smells like someone wants to build an argument against Stable Diffusion.
I don't think that any of the arguments on the article hold to imply that Stable Diffusion (company) is crumbling.
As an example, the "I sold 15% of the stake for $100" argument from the previous cofounder doesn't hold unless proven otherwise by a judge. The context always matter: is not the same selling 15% when there were no investors or IP, to selling it under coercion... but that a judge will be way better to assess than people external to the matter.
I wonder what's the gain behind publishing this kind of "yellow" articles (and to whom)
https://twitter.com/emostaque/status/1680774535342358528?s=4...
In a case where each side has a substantial vested interest in their story being accepted as true, it's wise to be skeptical of the claims on both sides and look for other, corroborating or contradicting evidence.
1. Did not inform him of pivot and generative AI art (false, even generated AI art for him)
2. Did not inform him of fundraise/inflows (also false)
3. Either of the above are not legal under Delaware law (also false)
We will find out in a few months I suppose.
A famous example of this is Ronald Wayne selling his 10% stake in Apple for $800 because he was worried that Jobs would incur a bunch of risky debt that he'd be liable for.
A shareholder cannot and should not ever be responsible for the debts of the company imho. As for time costs, the shareholder could just not attend or do anything, if they perceive the work to be of worthless value, and thus, should therefore have zero as the floor, rather than negative!
It's one of the most blatant attempts to pit workers against workers and undermine unions that I've ever seen.
And of course you may say that all CEOs are liars. And I would agree. But the smart CEO’s know not to lie about trivial things like education credentials.
I don’t think this is a hit piece. I think the incompetency of the CEO is finally bubbling up to the surface and rotting the company from the inside. And now the original believers are starting to turn on their former saviour.
Maybe not really having the degree but it mostly being satisfied is a positive with venture capitalist after the richest man in the world had problems with his physics degree.
https://twitter.com/emostaque/status/1682091613278072832?s=4...
Told them that and more https://emad.posthaven.com/on-setting-the-record-straight
On the cofounder share sale thing it was the Ron Wayne effect with massive sour grapes and him straight up lying
https://twitter.com/emostaque/status/1680774535342358528?s=4...
If it's the latter, then that's deliberately misleading to keep referencing it as if it were a normal MA. I do appreciate your efforts with the open source models though.
I don't get the issue when this is a standard thing, was actually at Oxford same time as Mustafa Suleyman who dropped out and is doing just fine, as is Sam at OpenAI who also dropped out of Stanford.
So its not a qualification thing, nor is it a deliberate thing to mislead when that's how you're meant to identify your degree (still weird maths and computer science is an arts degree tbh there).
The issue is that describing it as an MA implies that it is equivalent to a normal Master of Arts, when in fact you did not do any further work beyond the BA. Other Oxford graduates may "standardly" benefit from this "confusion", but it's not fair to all of the millions who have worked hard for years to achieve a real graduate degree.
Whether some other famous people dropped out of an undergraduate or graduate degree program or not does not mean that it is okay to pretend that you have completed a master's degree program of work. I do believe qualifications like that are given more weight than warranted, but that doesn't mean that it's okay to be deliberately misleading. Even having lots of other people from the same university doing it doesn't make it okay.
I didn't complete any degree at all. I don't go around telling people (in so many words) that I attended UCSD. Even though I did. Because just saying that without clarification would imply that I graduated.
@emadm the wikipedia article you link to explains why it's misleading.
I think the thing here is also intention given this is a weird thing, it would make 0 difference in anything I do for me to have tried to mislead folk that I did a masters degree.
I worked as an enterprise dev in my gap year for Metaswitch doing low level programming <= this for example is far more relevant experience.
For the role of an AI CEO typical qualification is actually dropping out tbh
My comment was meant to convey that I think this convention is misleading in general. If I was reviewing a stack of resumes from not otherwise noteworthy or qualified individuals I would have assumed the MA candidate had graduate level education as I'm unfamiliar with this convention, the Wikipedia article has some survey data regarding this.
I also agree it's irrelevant in your specific case, I honestly didn't even know if you had graduated or from where before this post and this detail doesn't change my opinion of your qualifications.
It is extremely unusual, for multiple media outlets to repeatedly write an article against a small startup, that's completely unrelated to customer harm or say employee abuse. Wow, the founder fudged his VC presentations, wow, the masters degree wasn't legit, like WHO CARES?
The only missing piece was the motivation, who exactly holds such a grudge against Emad really?
Then the missing piece finally came to light: https://www.reuters.com/technology/stability-ai-is-sued-by-c...
Basically, this "Cyrus Hodes" guy, thought stability was worthless before StableDiffusion, sold out his stake for $100. That is a fair assessment, because Stability before SD was completely aimless and going nowhere, and definitely not worth anything.
However, Emad spotted an incredible bet, and obtained the naming rights to a latent diffusion model for cheap. And the titanic success of SD instantly propelled Stability to a serious player on the AI world. Hence the 100 mil raise afterwards.
Needless to say, this Cyrus guy is extremely mad. If you check his credentials, he's not a technical guy, but actually extremely political, full of adviser roles to OECD or governmental organizations. Therefore its unsurprising that he:
1. Is able to organize the multi-month media effort in secret, to give journalists the angle to write against Stability.
2. Is ultra bitter about being kicked out of stability (But I doubt he would have been a good leader anyways at Stability, since AI transitioned from being a hype-based industry to a real product driven one after SD.)
In any case, Stability has had some hiccups, but they appear to be recovering very well in the recent months. They've had 3 separate product launches.
1. SDXL, the first successor to SD that finally has community acceptance (So the actual useful fine-tunes and ecosystems can grow around it)
2. Stablebeluga, a fast-follower high quality fine-tune for Llama2 with decent acceptance
3. Stablecode, a open source coding-specific LLM
It shows their R&D team is once again working, and gaining the confidence of the community again, after the two disasters that was SD2.0 and stableLM
> 1. Is able to organize the multi-month media effort in secret, to give journalists the angle to write against Stability.
This is cryptocurrency levels of conspiracy theory. Do you have any evidence? Or do you have an existing track record to back up these claims at least?
And then it devolves into worshipping the founder/priest/oracle. Usually by making claims of their supposed genius and foresight. Blah so boring. You can see it all on display in this comment. Emad the genius, yet he lies about trivial things. Yeah who cares that he lied about his master's degree? I mean I agree, who gives a shit about a master's degree. But then why feel the need to lie about it? You don't think that reflects on his character?
I gotta stop coming here. I just have to. Board the hype train and proudly ignore the red flags.
Choo choo, or something. I swear to God reading the comments here, on average, is making me dumber lately.
Can we fork the community? The MBA, PM, hype beast clout chasers can go to one side, and the folks pushing to GH 10 times a day and actually building stuff can have another?
I remember the one time talked about how he could scrape data behind paywalls and about deliberately bypassing anti-scraping techniques on Twitter. I don't think it was a very good PR move.
In the back of my mind I always had this thought that Stability could be some kind of AI accelerationist fund - hype up a shining future, raise enough capital to clear the bar for model training, release their model unencumbered and gather a fervent following of gracious users, and... not have a fully realized plan for what comes after.
But I have a feeling that even if they go under people will still be talking about Stability and their influence on generative art for a long while, at least until another freely available model can oust Stable Diffusion.
I noted that the most interesting data is behind firewalls as in private data as our entire model is open models to private data => not using that data to train our models
You can also license that data smartly, with royalty schemes and more, three no reason to get past paywalls etc nor do we do so (others do of course)
We are the only AI company to even offer opt outs, first models soon on that
People are creating all sorts of cool LoRas.
https://www.bloomberg.com/news/articles/2023-08-08/stability...
With that said 8.8.8.8 is still throwing me into an infinite Cloudflare captcha loop.
Edit: Cloudflare is doing the right thing, archive is being unreasonable - https://news.ycombinator.com/item?id=19828702
But the link works fine on Safari.
Does anyone understand what is meant by "a very definitive view of the future" in this context?
It even has a tvtropes page, which does seem to largely align with the article.
Interestingly, a different Bloomberg article from June has a slightly different version of the Mostaque quote, was also questioned by a commenter on HN, and "emadm" (I'm assuming that's him) responded: https://news.ycombinator.com/item?id=36512994
Edit: There's a third article with the same quote too: https://www.bloomberg.com/news/articles/2023-06-28/stability...
That's three articles within 6 weeks by Rachel Metz and Mark Bergen using the same quote. I found it interesting too but that's a bit much (and lazy).
https://www.reuters.com/legal/getty-images-lawsuit-says-stab...
Stable Diffusion and other AI models however create unauthorised derivative works.
This is a conclusion of law for which neither relevant facts nor legal analysis reaching the conclusion from the facts is presented. Its not even clear what the alleged “unauthorized derivative works” would be; it clearly can’t be images created by the model through prompting alone, since the copyright office determination that they lack the requisite human creativity to be works within the scope of copyright law necessarily means they cannot be derivative works.
(Let me remind you that what the act states is that fair use explicitly applies to, but is not limited to: "criticism, news reporting, teaching [emphasis mine], scholarship, or research purposes.")
Or perhaps you can point to settled case law on the subject?
That's US law, of course. EU law (which Stable Diffusion 1.x was created under) simply explicitly says ML model training is legal for research purposes, so the question is more about providing it as a service.
So under U.S. law, the more "transformative" something is (as opposed to "derivative"), the more likely it is to be deemed fair use. The line between derivative and transformative is fuzzy but generally, something like a movie adaptation of a book is derivative whereas parody is transformative.
Given that, suppose I have a cat pic. Google creates a thumbnail of that cat pic, which is, by itself, obviously derivative. But Google includes my cat pic alongside many other cat pics in response to a query for cats, the thing as a whole starts looking more transformative.
Now suppose my cat pic gets sucked into a generative AI. I suppose it could be used to create a merely derivative copy of the original cat pic, albeit with reduced quality, like a thumbnail. But the whole point of these models is to recombine features from millions of images to create something unique. That is, if I tell the model to draw a cat, it combines features from thousands of other cat pics. Which seems at least as transformative as simply showing the same cat pics in a grid.
From an ethics standpoint, the main difference between Image Search and a generative AI is attribution. Google Image Search is just links whereas Stable Diffusion is sort of opaque about its sources. But attribution isn't one of the factors of fair use -- a parody which makes fun of the original without ever directly identifying the original is still parody.
That said, I suppose you could argue it affects the economic impact of the copying, one of the other fair use factors -- it seems plausible to me that AI generated images impacts the market for, say, Getty images in a way that does image search does not. But it's anyone's guess how a court would balance those two things -- courts very often pretend all of the fair use point one way to discussing how factors are balanced.
Some parodies has been found as non-fair use when they supplant or becomes a replacement for the original work. People often get upset when that happens with head lines like "they are outlawing parody!", but from the laws perspective the outcome of a situation is actually very important when determining fair use. In the general case, parodies do not replace original works so the law works generally fine. I guess one could also view it as a transformative work in general do not supplant the original work, while a derivative usually do. A generative AI version of grumpy cat might behave more like a derivative than a generative AI version of a generic cat, even if both are using the same technology.
A sufficiently precise index is the same thing as the original work. See: https://en.wikipedia.org/wiki/The_Library_of_Babel
Many years ago, we licensed some pics from a small studio. Getty bought them. Claimed the pics were unlicensed (they apparently didn't keep the records of the studio), refused to believe our receipts, and threatened to sue us. Our lawyer said that the costs of defense would exceed the amount they wanted, so we should just pay.
Turns out we were an early victim in a major campaign throughout Europe. They never took anyone to court - just blackmailed a lot of little companies for what they could get.
Evil.
> Earlier this summer, Forbes published an exposé that highlighted his "history of exaggeration," and in its opening lines notes that Mostaque's claim that he has a master's degree from Oxford didn't hold up to scrutiny.
Those statements are very precise and can really only be true or false. I can’t verify the veracity, but if they’re false, he is lying, not embellishing or “exaggerating”.
There is a forgiving attitude towards compulsive liars in tech startups, and perhaps business in general, but that’s a cultural decease. There is no reason for journalists to play along in shifting the Overton window way into this kafkaesque relativistic worldview dictated by a small group of sociopaths and their enablers.
https://twitter.com/emostaque/status/1689376585047003136?s=4...
Masters degree was because I didn’t pay for postage for it to be sent which they acknowledged but spun, got it now
https://twitter.com/emostaque/status/1682124779426553856?s=4...
https://www.new.ox.ac.uk/oxford-ma-0
The only question is whether Emad applied and was awarded the MA.
> In Oxford (as in Cambridge), the status of Master of Arts is a mark of seniority within the University which may be conferred 21 terms after matriculation.
For the community, open-source AI is a blessing, no question.
I would say it makes sense to build a business on running open-source AI models.
I’m a hacker with side-projects. The outputs of ML models are useful to my side projects. But there is a world of difference between whipping up a tiny bit of Python code, or even using a web interface to query an open source model online, than running that same model locally.
For starters, I would need to invest a few hundred bucks in hard disk alone. My 3090 is no good to run a bunch of models, so I would have to invest in a more beefy setup. And, when the hardware is home and properly installed, I would have to spend tons and tons of hours fiddling with the code and its setup.
I rather put all those resources to the direct function of my side-projects, and not to something that, say, Stability AI can offer me for cents and in seconds, at least for the volume of services that I need.
That’s not to say there wouldn’t be scenarios where the financial equation goes the other way, but for me at least that hasn’t been the case so far.
Take OpenAI for example, even with their enormous popularity, I believe they are still making huge losses. The only way they can survive (right now) is with the huge financial injection from Microsoft. Besides, you can run your AI models on GCP or AWS and other providers easily, no need to invest locally. For a pay-on-hours tariff, it’ll be affordable for experiments. Huggingface is yes taking care of all the “logistics” for you already.
Had a record month
It’s known as an open core business model
https://twitter.com/emostaque/status/1649152422634221593?s=4...
The models are pretty tiny as well, a few GB. I think you could get pretty far with your setup if you wanted too
Operating a Google Cloud VM with a good GPU continuously is expensive for a bootstrapped startup. Sure, they opened themselves up to competition by making it open source. For example, I have also used Replicate quite a bit.
Maybe a lot of their income comes from contracts to fine tune models. For which open source makes things smoother.
Overall it seems doubtful that most of these companies will really make back all of the massive investments. But I certainly appreciate their (apparently subsidized) products.
Tunnels WireGuard over a websocket