Then Stability AI raised money from investors in October 2022.
Are you really telling me these investors didn't even do any sort of due diligence (not even "barely any" but literally any at all), didn't realized that Stable Diffusion was the work of many, and checked the license of the model and code?
That's a bit far fetched, to be honest.
Supposedly, this "leaked pitch deck" is supposed to show us that they lied to investors, but where is the pitch deck itself? Seemingly, only one of the slides is in the article, how could anyone reach any sort of conclusion based on just one slide?
Edit: The article seems to be some sort of clickbait trash that is so rampant around the web... One selected part:
> The Stable Diffusion code was released by researchers at LMU Munich in April 2022.
Links to https://arxiv.org/abs/2112.10752 which is the Version 2 of the paper, indeed released in April 2022. However, that paper is about, and links to https://github.com/CompVis/latent-diffusion which is not Stable Diffusion, it's Latent Diffusion. Stable Diffusion was released in August 2022, and is at https://github.com/CompVis/stable-diffusion
Or maybe growth is accelerating, and those who passed on series-A are upset?
Or maybe the lizard men are involved?
On this front. SD1.4/1.5 is extremely successful, but there has been no followup. SD2.0 was a total disaster that shocked the community, SD2.1 is still no better than SD1.5. StableLM currently is still a joke in performance. Like, they don't seem to even test their models before releasing them.
Compare their releases, to the polished and immediately dominant releases from OpenAI, and you have to question their researching chops. Did they only succeed with SD1.4 because of collaboration with other companies?
And why are you putting leaked pitch deck in quotes?
Is there examples of investors investing in a project without even visiting the homepage of the project or trying to understand even 0.1% of what the project is about? I don't think it's super common, but maybe I'm not hanging around the right/wrong circles in SV.
> And why are you putting leaked pitch deck in quotes?
AFAIK, this is the only article that mentions a leaked pitch deck from Stability AI and it only includes one image, which puts into question if there really is a leaked pitch deck, or if it's just one image that got sent to them.
Exactly. This reminds me of Fast.co [0] who recently shutdown after attempting to raise more VC capital when it was revealed that they only made 600K and with a very high burn rate.
I don't see how it is so difficult for many commenters here to see that investors were mislead by the claims made by Stability.ai's IP, and now they're also are burning cash with questionable revenues being generated whilst begging to VCs again to raise at a higher valuation.
Some of these investors just run into hype and get burned repeatedly and never learn from FOMO.
[0] https://techcrunch.com/2022/04/05/fast-shuts-doors-after-slo...
They had an issue with Runway with a model takedown or something but I think that was sorted out later on.
I don't know why the article seems desparately trying to show Stability in bad light while they have in fact funded a lot of AI model development and helped release it with permissive licence
Emad had to have funded the training of SD even before DALLE-2, investing the cash in exchange for branding rights to a free model.
Its obviously an extremely good bet in retrospect (RunwayML is probably bleeding with regret), giving stability herculean name recognition despite not having done any high profile research themselves. But when the bet was made, it was quite an insane bet requiring a lot of vision.
Naming rights on the model were no part of the compute grant, we give them incredibly freely and also support. Naming was suggested by the researchers in this case.
We don't just put out compute, but made sure to clear up everything from Stable Diffusion 2 onwards 100% trained by Robin and team: https://github.com/Stability-AI/stablediffusion
Actually the _only_ thing in contention per the article.
https://i.imgur.com/5XAndrR.jpg
https://i.imgur.com/OAZeZEl.jpg
So for whatever that’s worth, here’s a claim that predates this article by a week that emad was lying about stability origins to investors. I’ve trimmed the name off the DM for obvious reasons, but it’s from someone who seems credible (though I don’t know them).
I don’t like posting random unsubstantiated accusations without context. But this article just provided both.
As for what he lied about (or stretched the truth), I don’t know. I don’t even know if it matters. But this is why I never lie to investors, and why I’ve always really felt negatively towards colleagues who stretch the truth. It’s all kinds of misleading, and it’s a kind of misleading that tends to matter.
It would also make me upset as an investor if I found out that someone had fooled me. So even if it’d be to my advantage to lie for money, I can’t imagine doing it. I’d rather live in relative obscurity.
But this doesn’t seem like that. It seems like any disappointed investors just didn’t do the bare minimum you’d expect with an investment of $100m.
So while it's possible investors just don't want to admit they didn't look too deep, it's also possible everyone relevant knew exactly what was being offered: a fast-moving team of opportunists, adapting core tech originating elsewhere. That alone is plenty valuable in a hot area. 'Fast-followers', rollups, & integrators can do great in a tech boom.
So, this story might mainly be resentful competitors/employees taking potshots. (Or maybe even: potential next-round investors trying to win for themselves a lower-valuation via media spin.)
They certainly did that better than the way OpenAI did, or Midjourney did.
Where is the evidence of this? The article certainly doesn't contain anything that proves they claimed the owned the IP.
Stability has done it better since they purposely acknowledge the researchers instead of the way the public domain research is just used to make their own model with no acknowledgements (no co created mention anywhere). OpenAI would say 'co created by Google' if they did.
I think an ecosystem of users coming back to perform (and pay for) all kinds of generative AI tasks is totally worth the investment.
I can't figure out how Stability hasn't been able to make revenue though. Are they just not charging enough? Are they subsidizing everyone's compute?
That's wild. Handing out a bunch of cash for nothing except vague association. No control? No equity? No IP?
Reminds me of marketing. Cola puts logos on t-shirts of players and proclaims to have sponsored them, but I have yet to see cola claim to have "co-scored" a goal.
I understand having critiques about the post, but stories about major AI companies fundraising seems on-topic.
I recently saw a satirical post about vector databases get to top of HN and then drop to place 60 rapidly (no flagging in that instance).
I am trying to maintain an assumption of good faith with regards to moderation but lack of explanation on suspicious actions when it comes to stories with clear conflicts of interest between YCombinator/investors and the hn community don’t inspire confidence.
And if it is indeed open source with a completely permissive license, then what material difference does it make who originally developed it?
Its just sad to see such a negative article.
There are, however, fair questions about valuation and (lack of) moat.
The Runway takedown or something your refer to was Stability sending a takedown request to HuggingFace claiming an IP leak over the publication of version 1.5 by their academic partners Runway. That was sorted out, if you want to call it that, after massive public outcry and Emad claiming it was just confusion on HuggingFace's part. HuggingFace then confirmed Stability's legal team contacted them to retract the request afterward.
Then there was the incident where they contacted Discord to take over the main StableDifussion channel without informing the owner. There was also the incident where they convinced the owner of the main stable diffusion subreddit to give them mod rights and removed all other mods as well as any info and mentions related to one of the most widely used StableDiffusion tool (Automatic1111).
Stability AI lawyers never contacted HuggingFace requesting an IP takedown. You may ask HuggingFace themselves.
RunwayML agreed not to release 1.5 for safety reasons in writing and then chose to anyone presumably leading to confusion by our tech team who let HF know it couldn't be them.
You can ask them why they did that, Runway being a private company https://www.forbes.com/sites/kenrickcai/2022/12/05/runway-ml...
Three of the 5 authors of latent diffusion including the two lead authors are at Stability AI.
On the Discord it was Discord themselves that did it to the puzzlement of Kaarsetun and I, we are good with each other.
On the Subreddit that was our fault completely, for which I did a public apology around the issues with Automatic1111.
This article makes no such claim. It is about whether or not investors knew.
We employed Eleuther team members as Stability AI employees/contractors and incubated them until the 501(c)3 was set up and we managed to bring in other funders too: https://techcrunch.com/2023/03/02/stability-ai-hugging-face-...
I am on the board and delighted to continue to support their work as an independent organisation for LM evaluation, alignment and interpretability which is much needed.
Indeed though our approach was handing out significant compute for no control, no equity, no IP.
Anyone who has received Stability AI grants will be able to attest to this with multiple breakthroughs as a result, for example funding https://github.com/BlinkDL/RWKV-LM, the work of https://github.com/lucidrains and others.
Similarly we funded the beta of MidJourney with a cash grant for compute without ever even floating asking for equity etc as it is a market-creating innovation.
At the time MidJourney was using cc12m_1, a model developed by one of our lead (employed) generative AI developers Katherine Crownson / RiversHaveWings (https://github.com/crowsonkb/v-diffusion-pytorch)
Our model is simply to take open innovation and create commercial variants of that (our stable series models) from scratch and on our own, plus variants of that for private data - https://twitter.com/EMostaque/status/1649152422634221593?s=2...
This means we can be hands off versus other funders and trust researchers and help them succeed, something others do not.
GPT-J and co are GPT-2 equivalents, that are not actually useful in any way. So its more just building up the open source ecosystem.
Stability did get the branding rights to SD1.4, they spent a few million to get incredible name recognition out of nowhere, in the hottest and most competitive tech space ever. Those sponsored alphago tournaments costed way more and had less reputational impact than the millions using SD every day.
A model is not a paper and the transformer paper isn't what keeps the lights on in Mountain view. (Nor alphago for that matter.)
Agree on the incredible name recognition, but that's a mighty fragile thing if not built on sound foundation. Even a slight breeze like this article can make it wobble
It'll probably work given how everyone is falling over each other trying to give AI-anything money, but still gives me strong dot com era vibes. Appearance first, substance perhaps later.
We fund dozens of models and collaborate on them too, the ones we assert IP over are those in our GitHub as benchmark models based on open research coded and trained from scratch.
It gets really complex otherwise when you have lots of collaborators eg on the LAION models.
the correct analogy here is, how many in the crypto space invested in crypto companies and didn't realize that the companies didn't actually have any original IP or product, and the answer is a ton of them
"Company StabilityAI has requested a takedown of this published model characterizing it as a leak of their IP
While we are awaiting for a formal legal request, and even though Hugging Face is not knowledgeable of the IP agreements (if any) between this repo owner (RunwayML) and StabilityAI, we are flagging this repository as having potential/disputed IP rights. [...]"
They didn't hint at an account hijacking by a third party if that's what your team intended to convey to them however.
To add to that, your CIO addressed the community at the time and characterized RunwayML in the following terms: "[...] We also won't stand by quietly when other groups leak the model in order to draw some quick press to themselves while trying to wash their hands of responsibility."
Don't say buyer beware.
Which claims, precisely? Could you quote these claims that were made by Stability/Emad that makes it clear they were misleading the investors?
Because the article doesn't make those clear, these claims you talk about.
Assuming you have read the article (and its links), there is a video podcast which he claimed that 'his team' originally developed Stable Diffusion under Stability.ai without mentioning the original researchers that created the model, until they raised money which one of the researchers clearly mentions this in the article:
> “Once he had this money he became a bit clearer who actually developed Stable Diffusion and that his company did not own the IP,” he says. “Now we come to the interesting thing — if you’re running a company that’s raising a Series A and pushing for a lot of money and your VC guys realise that you don’t own any IP, that probably makes it a bit tricky.”
So according to the leaked Telegram message, the claim of millions of people or '5M' users using Stable Diffusion (for free) and DreamStudio which that itself is a paid SaaS and now they are running out of money [0] doesn't add up or seem to be very promising to Stability either way.
Just other typical VC cash in and burn as much money as possible for inflated numbers.
[0] https://www.semafor.com/article/04/07/2023/stability-ai-is-o...
This leaked deck was before stable diffusion release and the number refers to other open source models built by stability ai employees.
Our model is not SaaS and has never been.
The worst is when your solid business suddenly gets a competitor flush with capital which decides to start a price war, which, depending on the amount of capital they've got they may well win before they themselves go out of business.
Even though the troubles of FTX were hidden behind financials, someone who would look deeper into it would spot issues for sure. But maybe not from a brief look, while in this case with Stability AI, no one didn't even do a brief look?
I think it unreasonable to compare at this stage, we are still early as a company and organisation.
We have revenue (to be announced) but release models differently as we are not proprietary.
The numbers in that deck were as reported on models developed by team members (we chose not to assert IP rights) in use by various applications.
We were exploring co-creation full stack at that point, now we are focusing on base open models with commercially licensed variants.
They do, for the products they have/will have. Right now, I think it's only DreamStudio, but like any SaaS product, they can and surely do measure the amount of users, and trivial for them to see the revenue.
What you're talking about, the models they help release, are not products themselves, more like research output they give away for free. Probably in order to drive users to their paid products.
99.9% of SD usage happens outside of Dreamstudio. The power of SD comes from using fine-tuned models, and dreamstudio doesn't allow that (For copyright reasons presumably). There are websites like happyaccidents that do allow using fine-tunes, because they are willing to shoulder the massive legal risk as smaller startups.
SD's primary business model was always intended to sign up corporate clients in a semi-LLM-consulting/support role for their open sourced models. Like Redhat in a way. And to succeed in that space, you need massive name recognition.
Clipdrop, also, at least.
Do you have any more details on this?
We will be open sourcing DreamStudio shortly having refactored it.
