Speaker
Robin Rombach
Appearances over time
1 episodes
Episodes
1
Podcasts
Quotes & moments
Robin Rombach and his co-founders invented the latent diffusion algorithm — the foundational method behind all deployed image, video, and physical AI generative models.
Martin Scorsese worked directly with Robin Rombach to use Black Forest Labs' generative models to visualize pre-production scene concepts for a potential new film.
Black Forest Labs recently crossed 100 employees, with offices in Freiburg, Germany and San Francisco, and is actively hiring researchers and engineers.
Black Forest Labs' visual understanding models require only a few hours of fine-tuning data to adapt to a specific robotic task, dramatically reducing deployment friction.
By any definition anyone would have offered 10, 20, or 50 years ago — including the Turing test — we have already hit AGI. The goalposts moved because the reality arrived faster than our imagination could keep up.
Aristotle tutored Alexander the Great one-on-one. We've known for millennia that personalized adaptive teaching produces better outcomes. We chose factory-model classrooms anyway. AI agents that adapt to each child's learning style aren't a new idea — they're a 2,000-year-old idea we can finally afford to execute.
Cerebras is sitting on $25 billion in backlog, and every hyperscaler from OpenAI to AWS faces the same problem: demand is fully booked, and they're racing to keep customers from leaving — not chasing speculative future adoption.
The AI infrastructure buildout is unlike anything in modern history. Individual data centers draw more power than mid-sized cities, and nations from Kazakhstan to France are racing to participate. This is the Great Wall of China moment of our era.
Every chip before Cerebras followed Moore's Law — doubling performance every 18 months. Cerebras broke that curve with a fundamentally new architecture, and expects to far exceed 2x gains in the next 18 months. New architectures have room to optimize that mature 20-year-old GPU designs simply can't access.
Modern reasoning AI consumes enormous numbers of tokens internally — essentially thinking out loud before responding. That internal computation is inference, and speed directly translates to more reasoning cycles per dollar. Run Cerebras for 24 hours and you get the equivalent of weeks of AI thinking.
The open source AI landscape has quietly become a geopolitical flashpoint. Outside of OpenAI's OSS model, most available open source options are Chinese. Regulated industries in finance and healthcare that need on-premise, sovereignty-friendly AI have almost nowhere to turn for domestic alternatives.
Early AI adoption looked like everyone grabbing unlimited tokens with no strategy — exactly like someone wandering every aisle at Costco and walking out with $200 of impulse buys. Enterprises are now learning which AI models to use for which tasks, and routing intelligently between frontier and open source.
Nikesh Arora of Palo Alto Networks told Jason Calacanis that testing a frontier AI model against their own systems was devastating — it found critical bugs in hours that their own team had missed. They halted everything for six weeks of emergency patching. That's the case for government red teaming in a single anecdote.
Human learning moves at generational speed — like elephants, one cycle per 15–20 years. Geneticists study fruit flies because they produce two generations a day, compressing evolution into observable timeframes. AI is doing the equivalent: compressing thousands of generations of learning into near real-time iteration.
Every major generative AI model for images, video, and physical AI runs on latent diffusion — invented by Robin Rombach and his co-founders as PhD students in Munich. The insight: compress natural data like images and video into efficient representations, then train a transformer on that. JPEG and MP3 principles, applied to generative intelligence.
Robin Rombach sat with Martin Scorsese multiple times to demonstrate Black Forest Labs' generative tools. What captivated Scorsese wasn't automation — it was the ability to take a visual scene living in his imagination and externalize it for his team to iterate on. Language is lossy. Images are not.
The most underappreciated insight in AI right now: a single multimodal generative model can produce a movie and act as the perception and action brain for a physical robot. Pre-training on video gives implicit understanding of real-world physics — which transfers directly into robotic action prediction.
A Bitcoin movie starring Gal Gadot was filmed entirely on a sound stage, with all scenery generated by AI in post. The result: a $30M production that would have cost $150M with traditional set builds. It never would have been greenlit at $150M — generative AI didn't just cut costs, it made the film possible.
Star Wars AI fan films are already pulling millions of views per video on YouTube, with channels like 'Star Wars Stories Untold' leading the way. Jason Calacanis argues the smarter play for studios is to build licensing models that let fans be creative with beloved IP — and take a cut of the output.
Analysis
What they talk about
- Technology 67%
- TV & Film 33%
Connections
Shows they appear on and people they share episodes with. Drag to explore.
