May 3, 2023 · 45m · no-priors
No Priors Ep. 3 | With Stability AI’s Emad Mostaque
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of No Priors, Stability AI founder Emad Mostaque discusses the transformative power of open-source artificial intelligence, arguing for modular architectures, public infrastructure, and decentralized models over centralized Big Tech monopolies.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 19.9% of the talking time here. How this is scored →
The hosts as informed peer 4.3 Guest teaching 2.8 Guest disagreement 1.0 The hosts pushing back 0.9
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 26:02 Dismissal of OpenAI's AGI focus
Mostaque flatly dismisses Sam Altman's stance on image generation and AGI, asserting that monolithic elder-god AGI architectures are the wrong target compared to intelligence augmentation.
Hardest push from the hosts ▶ 17:31 Historical pushback on medical AI adoption barriersGil directly challenges the timeline for healthcare AI by pointing out that Stanford's MYCIN outperformed human physicians in the 1970s yet remained unadopted decades later.
Biggest teaching moment ▶ 35:50 Debunking LLMs as factual databasesMostaque educates listeners on information theory, explaining that compressing terabytes of data into a few gigabytes inherently loses factual precision and turns models into fiction engines.
The host holds their own ▶ 17:31 Host demonstrates deep history of medical expert systemsGil demonstrates deep technical history by citing the specific 1970s MYCIN infectious disease project to ground a speculative medical AI discussion in historical reality.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Emad Mostaque's Journey from Finance to AI | 2 | 1 | 0 | 0 | Sarah Guo opens with a standard biographical question regarding Emad Mostaque's transition from hedge funds to AI. Mostaque delivers an uninterrupted introductory monologue detailing his son's diagnosis and his philanthropic initiatives. | |
| Origins of Stability AI and Open Source Communities | 3 | 2 | 0 | 0 | Guo sets context around scaling laws and open community efforts like EleutherAI. Mostaque explains his personal funding of early generative notebooks, his aphantasia, and the transition from a DAO concept to commercial open source. | |
| Open Source Infrastructure and Supercomputing Compute | 4 | 4 | 2 | 1 | Elad Gil probes how open-source foundation models will survive capital requirements versus closed ecosystems. Mostaque makes the bold claim that Stability has access to more compute than Google or Microsoft via national exascale supercomputers and outlines a multi-stage training paradigm. | |
| Multimodal Initiatives and Computational Biology at Stability | 5 | 3 | 0 | 0 | Gil asks for a breakdown of Stability's multimodal and computational biology initiatives. Mostaque details edge distillation, OpenFold, and BioLM while emphasizing focus on private regulated data. | |
| Global AI Deployment and Leapfrogging Western Systems | 7 | 3 | 1 | 4 | Gil offers strong domain context by citing the 1970s Stanford MYCIN expert system to question healthcare AI adoption friction. Guo supplements with an analogy on historical mobile tech leapfrogging in East Asia, which Mostaque validates using developing world deployment figures. | |
| The Future of Media Creation and Augmentation Versus AGI | 4 | 4 | 3 | 1 | Guo cites Sam Altman's view that image generation is not core to AGI. Mostaque directly dismisses OpenAI's monolithic AGI framing, arguing instead for human intelligence augmentation and modular federated models. | |
| Language Model Strategies and Commercial Architecture | 5 | 3 | 1 | 0 | Guo highlights specific training run parameters (150,000 A100 hours) to probe language model strategies. Mostaque details small-to-medium specialized models, RWKV attention-free architectures, and custom enterprise deployments. | |
| Digital Democracy, Governance, and Model Limitations | 5 | 3 | 1 | 0 | Gil references sci-fi literature (Lady of Mazes) regarding algorithmic governance representation. Mostaque discusses multi-agent consensus and explains why LLMs must be treated as creative lossy compression rather than factual databases. | |
| AI Safety, Regulation, and Content Attribution | 4 | 2 | 1 | 2 | Gil and Guo explore alignment, commercial ad manipulation, and data attribution regulation. Mostaque advocates for compute registration above specific FLOP thresholds and provenance metadata standards like Content Authenticity. |
Statements from this episode (24)
Democratizing AI Is a Moral Imperative as Models Reach Superhuman Level
“More on that, a bit of a moral imperative to make this technology available to everyone because we're now going to very narrow superhuman performance and everyone should have access to that.”
Emad Mostaque Personally Funded Early Open-Source AI and Midjourney Grants
“So I started supporting the whole space at Luther and beyond, giving jobs to the developers, compute for the model creators, funding the various notebooks from disco diffusion to these other things, you know, giving grants to people like mid journey that were …”
Stability AI Was Originally Conceived as a Decentralized Autonomous Organization
“And originally I thought you got communities, you gotta make them kind of coordinated. Could a DAO work or a DAO of DAOs? And that's how stability started. After about a week, I realized that was not gonna work and it was incredibly difficult. So then I figure…”
Stability AI Repositories Beat Ethereum and Bitcoin Combined in GitHub Stars
“You know, on GitHub stars, we've overtaken Ethereum and Bitcoin cumulatively. It took them 10 years. We got there in like three, four months.”
Stability AI Has More Compute Available Than Microsoft or Google
“The reality is we have more compute available to us than Microsoft or Google.”
Stability AI Secured a Seven Million Hour Grant on Summit Supercomputer
“So to give you an example, we just got a seven million hour grant on Summit, one of the fastest supercomputers in the US.”
All Base Foundation Models Will Ultimately Open Source Their Pre-Training Phase
“So my take is that foundation models will all be open source for the deep learning phase.”
Distilled Stable Diffusion Models Can Run Locally on iPhones in Seconds
“Now, as of the release of Distilled Stable Diffusion, it will run in a couple of seconds on an iPhone, and we have neural engine access, because our view of the future is creating models that aren't necessarily bigger, but that are customizable and editable.”
Computational Biology Will See Artificial Intelligence's Biggest Scientific Breakthroughs
“And then I think on a breakthrough side, computational biology will be the biggest one.”
Stability AI Plans to Release an Open Medical Foundation Model
“We will release an open MedPalm model. Well, med stable GPT.”
Stability AI Is Deploying Four Million Tablets to Children Across Malawi
“So right now we're deploying four million tablets to every child in Malawi.”
Emerging Market AI Systems Will Outperform the West Within Five Years
“I am certain that system will outperform anything in the West within five years.”
Movie Studios Are Already Saving Millions by Implementing Stable Diffusion Locally
“We are working with multiple movie studios at the moment who are saving millions of dollars, just implementing stable diffusion by itself, let alone these other technologies.”
OpenAI Is Incorrect That Image Generation Is Not Core to AGI
“I do believe that they are incorrect around multimodality being, or images being a core component of that.”
Generative Image Models Will Soon Achieve 25 Frames Per Second
“With the additional breakthroughs that are coming through, it'll be 25 frame images a second. That's a hundred times speed up over a hundred times just from getting it out there and people interested in doing that.”
Stability AI Is Exclusively Training Video Models on the Eros Bollywood Catalog
“So like now we're training on hundreds and soon thousands of Bollywood movies to create Bollywood video models with our partnership with Eros you know, and that is exclusively licensed.”
Only About 100 People Worldwide Possess Skills to Train Foundation Models
“There's only like a hundred people who can train models in the world. Like it's not really a science. It's more an art, like losses explode all over the place when you try to do something.”
Multi-Model Interactions Will Dominate the Future of AI Over Zero-Shot Models
“And again, I think this is the future, not just zero shot, multiple models interacting with each other is the way full stop.”
Large Language Models Are Creative Fiction Engines Rather Than Factual Databases
“I do say these large models as well should be viewed as fiction, creative models, not fact models, because otherwise we've created the most efficient compression in the world. Does it make sense you can take terabytes of data and compress it down to a few giga…”
Large AI Model Runs Should Be Registered Like Dual-Use Biological Weapons
“I think that a productive outcome of that regulation is anything above a certain level of flops needs to be registered similar to, well, bio weapons and things that have the potential for dual use.”
Governments Will Eventually Pour Hundreds of Billions Into AGI Model Scaling
“When governments decide, and I don't believe they've decided yet, that having an AGI is the number one thing, tens of billions, hundreds of billions will go into building bigger models.”
Malicious Actors Already Possess Advanced AI Models and Thousands of A100 GPUs
“I think the bad guys, I'm going to put that in quotes have access to these models already and thousands and thousands of A-one hundreds.”