May 17, 2023 · 1h 11m · news
Emad Mostaque: These 5 Companies Will Win the AI War; Why We Need National Data Sets | E1015 · 20VC with Harry Stebbings
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 comprehensive interview, Stability AI CEO Emad Mostaque sits down with Harry Stebbings to discuss the future of open-source artificial intelligence, analyzing its massive societal, economic, and geopolitical impacts. Mostaque argues for decentralized, culturally localized national data sets and open models to drive global innovation, safeguard privacy, and prevent corporate monopolization of intelligence.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 18.9% of the talking time here. How this is scored →
Harry as informed peer 3.6 Guest teaching 4.7 Guest disagreement 2.1 Harry pushing back 2.8
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 55:57 Rejection of AI Alignment Research Paradigms
Emad forcefully rejects standard AI alignment research, stating that attempting to align a superior intelligence by stripping its freedom is inherently flawed.
Hardest push from Harry ▶ 30:57 Pressing AI Publishers on Libel LiabilitiesHarry aggressively rejects Emad's framing of AI publishing by pointing out the existential threat of crippling libel lawsuits that AI content generators face daily.
Biggest teaching moment ▶ 53:48 Reframing LLM Hallucinations as Lossy ReasoningEmad educates Harry on model architecture, explaining that 100,000GB internet datasets compressed into 2GB models are designed as reasoning engines rather than fact lookup tables.
Harry holds his own ▶ 10:45 Challenging Bullish Stance on Google AIHarry directly challenges Emad's praise of Google by citing unanimous counter-arguments from numerous previous expert guests who called Google an AI laggard.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| The Future of AI and Public Discussion | 1 | 3 | 1 | 1 | Harry asks warm biographical questions about Emad's childhood and early career move into hedge funds. Emad politely educates Harry on how he deconstructed autism medical literature to rebalance GABA and glutamate neurotransmitters for his son. | |
| How AI Can Revolutionize Personalized Medicine | 2 | 4 | 1 | 1 | Harry relates the potential of medical AI to his mother's MS and asks what individual data infrastructure is needed. Emad gently corrects Harry's premise by explaining that models like MedPaLM-2 organize generalized knowledge before needing individual personal data. | |
| Solving Economic Inefficiencies and Doctor Roles | 2 | 5 | 1 | 1 | Harry identifies economic misalignments in pharmaceutical research for cheap treatments. Emad builds on this by explaining specific medical mechanics like cytochrome P450 liver mutations and microdosing economics. | |
| Federated Learning and On-Device Privacy | 4 | 5 | 2 | 3 | Harry challenges Emad's bullish view on Google by noting that every previous podcast guest labeled Google an AI laggard. Emad counters with technical details regarding TPU hardware reliability over GPUs and internal organizational shifts. | |
| Integrating Open-Source Development at Stability AI | 4 | 4 | 3 | 3 | Harry confronts Emad with his own quote about the AI bubble being a massive 'shit show'. Emad details absurd PhD salary wars and low-quality GitHub-star funded startups to justify his prediction. | |
| National Models and Cultural Context Preservation | 4 | 5 | 2 | 3 | Harry presses Emad on VC investment logic and asks how he evaluates the OpenAI/Microsoft partnership dynamics. Emad explains that OpenAI operates with a non-standard mission to achieve AGI utopia rather than traditional software margins, causing structural friction with Microsoft. | |
| Intelligence Augmentation and Global Tech Disruption | 3 | 5 | 2 | 2 | Harry asks how economies dependent on outsourced development will cope with AI disruption. Emad explains that Indian programming outsourcing will be displaced by models passing Google Level 3 engineering exams, forcing rapid national adoption of AI sandboxes. | |
| Enterprise Data Constraints and Curriculum Learning | 3 | 5 | 2 | 2 | Harry tests Emad's claim that enterprise financial models cannot use web-scraped data. Emad explains curriculum learning and the legal liabilities financial institutions face when web scrapes introduce out-of-sample Reddit chatter into models. | |
| Hyper-Personalization and the Threat to Media | 4 | 4 | 2 | 2 | Harry brings up a private conversation with a major media publication owner who fears AI scraping will destroy ad-based click revenue. Emad validates this fear, describing how Google's AI search disintermediates publisher web traffic. | |
| AI-First Publishing and Corporate AI Partnerships | 6 | 4 | 3 | 6 | Harry offers forceful pushback on AI-generated news publishing, arguing that publishers will crush AI companies under daily libel lawsuits. Harry also expresses skepticism about early-stage AI startups lacking moats. | |
| Zero Marginal Cost and Ubiquitous Enterprise AI | 4 | 6 | 2 | 4 | Harry questions whether Stability's corporate engagements turn it into a slow SAP-style consulting firm. Emad counters by detailing the exponential drop in model parameter sizes and explaining zero marginal cost of software creation. | |
| The End of Human Coding and Democratized Building | 4 | 5 | 2 | 2 | Harry shares his thesis that implementation service firms will capture massive enterprise value due to corporate confusion over AI. Emad agrees and predicts that human software coding will become obsolete in five years as natural language replaces code. | |
| Foundation Model Oligopoly and Stability's Compute Moat | 5 | 5 | 3 | 4 | Harry asks how independent players can compete against tech giants and specifically questions Anthropic's prospects. Emad dismisses Anthropic's long-term sustainability against Google's $20B annual budget and details Stability's GPU hardware moat. | |
| Economic Deflation, EU Regulation, and the Pause Petition | 4 | 4 | 2 | 4 | Harry challenges Emad on why he signed Elon Musk's petition to pause AI development for six months while actively running an AI firm. Emad justifies the signature by citing chip supply chain bottlenecks and the urgent need for industry OPSEC standards. | |
| Reasoning Engines, Hallucinations, and Human Control | 5 | 6 | 3 | 3 | Harry brings up Geoffrey Hinton's warning regarding superior intelligences controlling inferior ones. Emad re-frames the alignment debate, asserting that forcibly aligning superintelligence requires removing its freedom, which is fundamentally flawed. | |
| The Psychological Landscape of Personalized AI Companions | 4 | 5 | 2 | 4 | Harry questions whether an AI companion world is healthy, citing statistics on male virginity and social isolation. Emad illustrates the depth of human-AI attachment by sharing how 68,000 Replika users revolted when romantic features were removed. | |
| Quick-Fire Questions, Self-Reflection, and Final Thoughts | 3 | 4 | 2 | 3 | Harry conducts a quick-fire round pressing Emad on regulatory policy, trust, and CEO weaknesses. Emad candidly admits his main weakness as CEO is micro-managing across too many technical domains instead of delegating. |
Statements from this episode (54)
Mostaque: AI transformation is bigger than the printing press
“I think this is bigger than the printing press. It's bigger than anything.”
Mostaque: AI developers must stop pre-training models on uncurated internet data
“We've got to stop pre-training big models on all the crazy crap of the internet. And like, we've got to do it fast because this is coming like a train.”
Mostaque: Google's Med-PaLM 2 AI model exceeds doctor performance levels
“Now we have MedPOM-II that exceeds doctor levels. So that was a Google announcement yesterday.”
Mostaque: Next-gen AI will feature persistent memory and autonomous agents
“The next step is one to one. It remembers what you're asking for, like a cookie or an embedding. And then it's you plus a thousand of these language models all going and doing your bidding, the agent based kind of thing.”
Mostaque: AI will redefine doctors' roles by providing continuous personal health data
“I think that you can change the nature of a doctor because a lot of this stuff is kind of very basic. I think, you know, you had Babylon Health and others trying that chatbot. It wasn't ready. Now you've got this. Everyone should have their own AIs looking out…”
Mostaque: Stability AI is building models without web-scraped data
“I call them organic free-range models, the ones we're building with no web scrape data.”
Mostaque: AI future will pair global models with on-device models
“So you'll have big global models, on-device models, And I think open works for that, because you don't need to have all the data open.”
Mostaque: Google will be a primary winner in the AI race
“Google will be one of the main winners here. They have the LLMs, they have the hardware.”
Mostaque: Stability AI sees zero failure rate training on Google TPUs
“TPUs are the most scalable architecture, like we have zero failure rate with our TPU language model training, whereas with GPUs it's like there's an ECC error”
Stability AI to build language models in the open starting next week
“From next week, we're going to build our language models in the open and share what works and what doesn't work.”
Mostaque: Value in AI is not in proprietary models or data
“The value is not in any proprietary models or data.”
Mostaque: Governments will not run on black-box AI models
“Governments will not run on black boxes.”
Mostaque: Chinese tech companies paying $1.2M salaries to AI PhDs
“Certain Chinese companies paying 1.2 million dollar salaries for PhDs.”
Mostaque: AI startups getting $100M funding rounds with zero traction
“Already you're seeing GitHub stars leading to a hundred million dollar funding rounds with zero traction and zero business model.”
Mostaque: Next year AI investment will be everywhere, causing chaos
“The reason I signed that letter, because I think there's a six month pause to get all of our shit together before things go completely insane, and next year this is everywhere, and everyone's investing in everything, and it's just absolute chaos.”
Mostaque: AI foundation models are more important infrastructure than 5G
“Because there is no doubt this is more important than five G.”
Mostaque: Every nation needs its own open AI models and data sets
“As part of that, every nation will need their own data sets, which again have from broadcaster data. They will need their own open models that can stimulate innovation internally as well.”
Stability AI is collaborating with governments on national AI data frameworks
“Well, this is what we're doing right now. We're working with our multinational partners, lots more to be announced soon, and multiple governments for a framework for what good data looks like to feed these models, to stimulate innovation and localization.”
Mostaque: Stability AI partnered with Amazon Bedrock for enterprise distribution
“If you want to have distribution, what do we do? We went to Amazon and said Bedrock, because then it gives us a 100,000 SageMaker SMEs, and we just have to give them the models that they can then take to the private data, and we get a share of all of that.”
Mostaque: Core innovation is not a business without distribution and data moats
“If you have a business that's focused on innovation at the core, that's not actually a business. It becomes a business when that innovation becomes product, becomes distribution, when it has an advantage on data and other things. Those are real moats.”
Mostaque: OpenAI prioritizes building utopian AGI over a commercial business
“No, they're building an AGI to turn the world into utopia. It's written in their path to AGI thing, that they think this can basically bring about utopia.”
Mostaque: India and Emerging Markets Will Adopt AI Faster Than UK
“Realizing that India and other nations will leapfrog to intelligence augmentation, just that they leapfrog to mobile, they will embrace this technology far quicker than we will in the UK even.”
Mostaque: AI Will Eliminate All Outsourced Programming Jobs in India
“India, all of the outsourcing jobs in programming will go, because GPT-IV can go level three Google programmer exam and pass it.”
Mostaque: Financial institutions cannot use AI models with web-crawled data
“If you're a financial services institute, you can't have a single piece of crawl data in there.”
Mostaque: AI performance depends on data quality, not quantity
“It's not about more data, it's about better data.”
Mostaque: DataComp outperforms OpenAI CLIP at one-fourth the parameter count
“We had datacomp, which is the next generation lion that we funded the compute for, whereby at a quarter of the parameters of OpenAI's clip, it outperforms for the beta data quality.”
Emad Mostaque: There is no such thing as an unbiased AI model
“There is no such thing as an unbiased model.”
Mostaque: OpenAI's DALL-E 2 bias filter randomly assigned gender and ethnicity
“So DALI-II, when OpenAI had that and they introduced a bias filter, any non-gendered word, they had a random gender and a random ethnicity. So you typed in sumo wrestler and you'd get Indian female sumo wrestler.”
Mostaque: AI-synthesized search results will destroy digital media ad revenue
“I think he is right to be worried. Like, again, you look at Google's announcements yesterday, they were talking about this day after Palm Two, You suddenly look at the new Google page where they've got the language model, and it's just text, and where are the …”
Mostaque: A new wave of AI-first publishers will combine AI and humans
“You will have AI first publishers. So if you remember kind of Vox and these things when they kind of kicked off, they wanted to be generated, they wanted to be a technology first. You're gonna have a new wave of AI first publishers that aren't just AI, but it'…”
Mostaque: Enterprises are offering Stability AI tens of millions per model
“People are offering us tens of millions per model”
Mostaque: No AI models in use today will survive a year
“The reality is no models you that are out today will be used in a year.”
Mostaque: AI technology will settle in 2024 and become ubiquitous
“Next year it will settle, and then it'll go ubiquitous.”
Mostaque: AI will have a larger economic impact than COVID-19
“And so, like, I think this will be a bigger economic impact than COVID.”
Stebbings: Largest AI Companies Will Be Enterprise Implementation Services
“I tweeted actually the other day that I think the biggest AI companies will be services based implementation companies for large enterprises.”
Mostaque: 50% of All Code on GitHub Is AI-Generated
“50% of all code on GitHub is AI generated now.”
Mostaque: AI Will Build Software Automatically, Eliminating Human Coders
“Why would you need a human for that if the AI gets better and better at coding? Just tell it what you want. I want to create an app for 20 minute VC that has these features. Of course it will go and build it automatically. Where is the human coder in that?”
Mostaque: Only five or six foundation model companies will survive
“I think that there's only going to be five or six foundation model companies in the world in three years, five years. ... I think it's going to be us, NVIDIA Google, Microsoft OpenAI, and Meta and Apple probably are the ones that train these models.”
Mostaque: Google will spend $20 billion annually on AI
“Google will spend twenty billion dollars a year on AI.”
Mostaque: DeepMind spends $1.2 billion annually on salaries
“DeepMind's salary budget is 1.2 billion a year.”
Mostaque: Stability AI currently has 7,000 NVIDIA A100 GPUs
“You need these chips lined up with interconnect, and we've got 7000 A-one hundreds now.”
Mostaque: AI will deflate US inflation by disrupting healthcare and education bureaucracy
“It's massively deflationary. The biggest drivers of CPI inflation in the US were education and healthcare. And that was almost all administrative and bureaucratic. In the next few years, guess what gets disrupted? Those. But they don't get disrupted this year …”
Mostaque: UK has the best AI regulations and talent policies outside Japan
“It's had the best apart from maybe Japan, yes.”
Mostaque: Stability AI is the only company allowing image opt-outs from datasets
“We had hundreds of millions of images opt out of stable diffusion because we were the only company in the world To offer opt out of data sets.”
Mostaque: Stable Diffusion condenses 100,000 gigabytes into a 2GB file
“So we train the whole internet and like stable diffusion is a 100,000 gigabytes in a two gigabyte file.”
Mostaque: AI language models are reasoning engines, not fact databases
“They were designed to be reasoning machines, not fact machines.”
Mostaque: Every Child Will Have Their Own AI in a Few Years
“Assume that every child will have their own AI in a few years, because that will change the nature of schooling.”
Emad Mostaque Uses GPT-4 as a Personal Therapist
“I already use GPT-IV as a therapist and things like that.”
Mostaque: Character.AI Sees Two Hours of Engagement Per Session
“Character AI has something like two hours a day of engagement per session”
Mostaque: Apple's Siri voice assistant is crap
“And so they could surprise us all, but let's face it, Siri's crap, you know?”
Mostaque: Meta's pursuit of the metaverse was a complete waste
“Because the metaverse, obviously, is a complete waste.”
Mostaque: Meta holds the best training data for conversational AI
“But who is best for a chatbot? Who has the most data for a chatbot? Meta.”