Stability AI podcast mentions

Quote

money on previously to do the you know kind of like let's let's use it to Test out all new kind of plot ideas. Uh yeah, previs. Yeah. It's incredible, whatever. And all those things are super incredible. I still like the it's very nostalgic, but I still like. Like the idea of like this is a camera and a person and a person that says, you know Yeah. But we'll we'll see how that plays out. Yeah. I think you know, so one of the things that stability AI uh made an impression on me was like, well, you know, and at least now we can remix Game of Thrones season eight. Like in you know, uh like like it was meant to be not uh not rushed. Watch um well I have a six and a half year old and I you know you see a lot of these kid movies and you're like yeah that probably I don't totally know the job math because I don't know how many animators there are today. But I actually think weirdly, I think we could be producing more high quality maybe And so it's maybe that's a positive is like we could just have have like more like you could just have a Pixar for like, you know, things where kids learn stuff. And it used to be these like very, you know, lo fi, uh, you know kind of I mean we had teletubies, you know. That was so slight. could have m way more of that and and maybe every animator that today is making a Pixar film

Quote

up, are people now starting to want, you know, actual profits and revenue? And it does seem to maybe be the case, but if you're in a profitable sector that isn't as I guess long term a bet. I do think there is still space for VCs to be excited like with this company. Next up, stability AI appoints new chief technology. officer. So v the new uh CTO here is Hanno Bus. Uh this person has thirty years of experience as a CTO at companies like that. Like digital domain, which is uh visual effects and digital production. company and this is of course following a tumultuous Year for stability AI, as we've covered, there was famously the um I don't know if you say Ausmint or or the uh Let's say the CEO of Sability I left uh at some point and uh Yeah well anyway, the the leadership alone. Uh Seemingly because the company was a bit chaotic, didn't have a clear business.

Quote

Hello and welcome back to the Cognitive Revolution. Today my guest is Ahmad Mustaf. Famously the founder of Stability AI and currently the founder of intelligent internet and the first time. and author of the provocative new book, The Last Economy, a Guide to the Age of International. Intelligent economics. Very few people manage to grapple seriously and honestly with the same. But imagine Odd has. Since founding Stability in twenty nineteen, he's demonstrated a deep understanding. Risks, as evidenced by the fact that he signed, while still CEO of Stability, the famous twenty twenty three Pause Letter The fundamental problem that a mod adjusts.

Quote

Pro not open source, they uh made it available. via API and they did release Flux Point one dev for non commercial use. Use Schnell is open to do whatever. Not quite the Llama three point one four five B release. So this would be kind of like meta saying you know we're keeping the four oh five B to ourselves but open sourcing the 8B variant. Yeah, not as similar from what stability AI has started doing. And on to the next story. Yet again, a big player. It is Google. So, yeah. We have released uh first of all Gemma two two billion models smaller than the other ones that were released pretty recently. And there's a couple Variants of it, so Gemma two B but also shield Gemma and Gemma Scope. So Shield Gemma is a set of safety classes. And Gemma scroll. Hope is a tool that allows developers to examine specific points within Vigemma two model similar to what we've seen with recent developments from

Quote

logos and various things like that. And yeah, they they claim that they're better when Journey and Dali Free. At this point it's kinda hard to tell. They're all quite good. But in any case, Ideogram is definitely a major player in this Space given that they have their own model, that is quite uh quite good. So they are uh they're not releasing it as open source, so this isn't like a uh you know stable a stability AI type play this is a closed source play and they're charging you know between seven and fifteen bucks per month. So again, very much in the butter zone of like what we tend to see for these kinds of apps. Um Interestingly, Andreesen Horowitz was participating in this round, so so this is a you know, an an S V Angel. So some Really and Redpoint actually. So a lot of really good VCs backing this. Um so yeah. Yeah, we'll we'll see what the the thesis is here. Um but one of the things that they do highlight. in practice about their new model is that it doesn't just generate square images, which is an issue um, you know, with DALI three for example. Um you know supports all kinds of aspect ratios and as you said, a lot better with uh with text as well. So there do seem to be these like marginal advantages that folks are still discovering in this space. Uh we'll see how long that lasts and and whether it's enough to build a viable

Quote

focus there and and certainly Eric is uh a very knowledgeable hardware guy from what I remember of him. at YC um and uh anyway, other other core team members at Oculus. So really, really well backed, well advised, and uh an interesting one to watch. You know, there are a lot of as you said, a lot of product like this hitting the the shelves. Obviously rabbit is another kind of you know vaguely analogous. But we're seeing more and more of the hardware meeting the software when it comes to AI. Next story. Stability AI launches SVD 1.1. So this is about an update to table video diffusion, moving it to 1.1 away from 1.0 F We covered one point zero just a couple of weeks ago and this is Just an improvement really, but does highlight, I think, a movement in uh video generation uh becoming more commonplace and and I think making pretty rapid progress now and probably for the rest of us here we'll We'll see a lot of uh text to video getting better. Getting commercialized, maybe starting to be used for various applications. OpenAI launches Chat GPT app for Apple Vision Pro, and actually this is

Quote

And as with the L L M providers, right, there's not There's some, but not a ton. And it'll be interesting to see. you know, when V V C money is burned, right, then you actually need to get by on revenue alone, how will that competition play out? And yet again, speaking Speaking of text to image, we have another story on that front and this time it's about Stable diffusion three point five. So stability AI haven't talked about them in a little while are releasing stable diffusion 3.5. And once again it's coming in free sizes, large, large turbo and At photorealistic AI images. So the comparison in this article is to Flux one point one pro and it looks pretty significant. significantly better uh compared to SD free in the sense that it is comparable to float. I think initially when I saw the outputs of flux uh on X It was very impressive. It did make me feel like s it is surpassing state. So I guess not surprising that we are launching this to try and keep up So to speak.

Quote

of this company. Uh and we see that mirrored as well in the partnership with X, right? Where they're generating I mean there are going to be some on there, but uh You know, you're presumably not going to see this thing generate uh certain kinds of porn. Graphic imagery, let's say. Um, but generally they're trying to reduce the kind of the the number and extent of safety guardrails. So yeah, really interesting founders, former researchers, uh uh stability AI, you mentioned that uh Andre earlier, so they definitely have a pedigree and um There's a lot of uh much ado in this article about the misinformation dimensions of this. And you know, all everything you'd expect, quoting people saying, Hey, you know, this is really bad, we're getting all this You know, misinformation's gonna flood on uh on Twitter, on X. I mean at a certain point with open source. Source image generators, I think this was all kind of baked in. Maybe this advances that timeline. by like six months, um, maybe a year, but you know, it's not a fundamentally different trajectory. from the one that we'd been on. So you can see those arguments really go uh go either way. Right. said I don't think this really changes the game that much. Uh maybe it makes it easy Yeah. Uh mostly what people have been doing is making ridiculous

Quote

So on top of having an Apache two point zero license, the code of conduct for a model Done. Is uh pretty detailed, like you're not supposed to use it for bad things here. The excellent to each other is all various to it. Next up, stability AI. It brings a new dimension to video with stable video 3D. This is about The paper S V D free, novel, multi view synthesis and three D generation. from a single image using latent video diffusion. So in short, stable video 3D, as the name implies, based on on some text, you are now able to render what is kind of like a 3D video. So there's a panning shot of of some stuff happening. So you can like rotate around a character, for instance. And This is coming a couple months after something like stable zero. that they are actually using an image to video diffusion model.

Quote

Anyway, it it's definitely um it's definitely an interesting uh ad advance. maybe a modest advance on uh inference speed as far as I can tell just from looking at the figures. So um yeah, uh cool advance and and interesting to see stability continue to pump these things out. Images do look good, I will say. I mean there nothing nothing obviously wrong with any of the things that I'm gonna do. the faces or hands or anything like that. Uh so uh another another big leap forward for uh stability ai and moving on to research and advancements we start with self-discover Large language models self-compose reasoning structures. Let's talk about prompting for a second. Usually when you have a language model, you have to come up with some kind of prompt to get it to behave optimally. You have a problem you want it to solve. Uh it's not the case that you can usually just straight up ask the model to solve the problem and it'll do it perfectly. you know, sometimes works but often for especially more complex tasks. You have to try them step by step. Like let's think about this step by step. You know, give me your your reasoning. And then based on that reasoning, kind of guide yourself uh step by step towards the The answer. There are a whole bunch of other strategies like self-consistency as an one right you do chain of thought you get the model to lay out its thought process and get an ou…

Quote

in image generation and editing capabilities. So now these models are able correct kind of like you we get into AGI for image generation almost. Right. And f BlackBooks Labs is a startup that's been around for a while. While spun out of stability AI and flux for a long time was one of the leading text to image models. They also had open source variants. So But Flux two they are introducing basically you could say the nano banana generation, the GPT five image generation. of image synthesis for their system. Lots of details as you might expect. We have a bunch of variants, Flux2 Pro, highest Performance flux to flex, flux to dev, which is a thirty two billion under Apache two point oh and their VAE. So they're Still doing this partially open source thing that they've started with and have kept with

Quote

Go and be on archive with some review. So It's not a huge deal, it's not like a dramatic kind of tragedy or anything, but and Interesting example of where we're at in the world their archive which is this neat On to policy and safety. First up on On the legal front we have stability AI largely winning UK court battle against getty images over copyright and trademark. So So Britain's High Court has ruled for Getty on trademark. Infringement specifically for stability AI images with Getty's water model So, I think that's a good thing. So this is essentially dismissing a major part of what Getty has accused stability of and Justice Jonas Smith concludes included that stable diffusions AI or stability AI did not infringe copyright Because it does not store or reproduce copyrighted works, which questions about generative AI and and especially I guess uh text to image