Why Diffusion Will Win AI Inference with Inception Co-Founder and CEO Stefano Ermon
No Priors: Artificial Intelligence | Technology | StartupsShow Notes
As generative AI hits hardware and latency bottlenecks, Stanford professor, diffusion pioneer, and Inception co-founder and CEO Stefano Ermon is betting on a radical new architecture. Stefano joins Sarah Guo to talk about Inception, and how his team is applying diffusion architecture beyond images and video into discrete text and code generation. Stefano explains the limitations of autoregressive LLMs, as well as why parallel token generation in diffusion models offers superior inference scaling and hardware utilization on standard GPUs. He also shares details about Inception’s Mercury models, real-world voice agent applications, the software stack required to serve diffusion-based models at scale, academia’s role at the frontier of AI innovations, and why the next era of AI competition will be defined by efficiency.
Sign up for new podcasts every week. Email feedback to [email protected]
Follow us on Twitter: @NoPriorsPod | @Saranormous | @EladGil | @StefanoErmon | @_inception_ai
Chapters:
00:00 – Stefano Ermon Introduction
00:35 – Research Background
02:54 – Starting Inception
05:59 – Why Diffusion Beats Autoregressive
11:10 – Discrete vs. Continuous Modalities
13:19 – Inception Today
16:45 – Where Speed Wins
17:31 – Inception Customer Base
18:49 – Interaction with Hardware Landscape
19:34 – Inception and the Broader Industry
21:41 – Data Compression and Structure
24:45 – Controllability of Diffusion Modeles
27:25 – Emergent Capabilities at Scale
29:02 – Future Workload Split Between Diffusion vs. Traditional
30:03 – Adoption Challenges
31:44 – Hiring and Team Organization
32:50 – Recursive Self Improvement
34:02 – Resource Allocation
35:10 – Impact of Academia
38:13 – Conclusion
More from No Priors: Artificial Intelligence | Technology | Startups
In the transition to the agentic web, security has become an afterthought. Now, organizations are racing to figure out what AI security means, and how to keep their data safe. Live from Cloudflare Connect, Sarah Guo sits down with Cloudflare CTO Dane Knecht to discuss the changing attacker landscape, and how his company is allocating its engineering resources to meet this emerging security challenge. Dane explains how enterprises that have moved away from castle-and-moat infrastructure to more of a zero trust infrastructure for security are in better positions than those who haven’t. He also discusses the challenges of patching 70 years of software and why the software development lifecycle will need to change for security for the agentic web, why he sees token spend in the next 18 months primarily going towards security rather than engineering, how the current web is in a ‘messy middle’ state where humans and agents are riding the same ‘rails’ on the internet, Cloudflare’s conference monetization gateway release as well as its Signals project with OpenAI, the rationale behind the release of Cloudflare OS, and the thinking behind a recent company restructuring. Plus, why Cloudflare hired one thousand interns, and why curiosity about AI is required to be successful at Cloudflare.
Adversaries are already running AI agents, and the fastest attacks now play out in seconds. Sarah Guo and Elad Gil sit down with CrowdStrike president and CEO George Kurtz on what AI-equipped attackers mean for enterprise security. George explains how agents are flattening the adversary pyramid by putting advanced tradecraft in the hands of low-skill attackers, and why CrowdStrike and Nvidia are building a frontier-style lab of their own to train security models. They also get into how open-weight and closed models compare on security, where CrowdStrike is directing its platform investment, lessons from the recent Hugging Face incident, and why George sees AI, like every platform shift before it, as a net creator of opportunity.
Technology has always been the fundamental driver of improvement for human quality of life. And with the recent rollout of advanced frontier models, like OpenAI’s GPT-6, humanity can now massively accelerate scientific discovery for its benefit. Sarah Guo and Elad Gil are joined by OpenAI President and co-founder Greg Brockman to discuss the company’s rapid progress across frontier models, cybersecurity, and scientific discovery. Greg shares how nearly 400 century-old math problems have been solved using its AI agents, and how the technology has the potential to transform other fields like biology in the realm of drug discovery. He also talks about how proactive, always-on AI agents like Dots impact modern productivity, why he believes the best UI is no UI at all, how cybersecurity is in a critical window of time to use peak intelligence to find and fix vulnerabilities in systems, and why OpenAI remains committed to keeping humans in the driver’s seat even as models surpass us in capability.
Is the future of frontier AI open or closed? ReflectionAI co-founder and CEO Misha Laskin joins Sarah Guo and Elad Gil to discuss the launch of Beam, the company’s 500 billion parameter open-weight reasoning model. Misha breaks down the pre-training and reinforcement learning required to produce Beam’s reasoning efficiency, the shift in enterprise compute from renting to owning intelligence, and how he believes that open models will capture the majority of global token demand. He also talks about the open model ecosystem in China and why competition with China’s models is a good thing, safety considerations around open models, and how frontier-level open models may accelerate the pace of scientific discovery.
Founder and CEO of full-stack AI chip company Fractile, Walter Goodwin, joins Sarah Guo to discuss the bets he’s made on the future of the chip market as other major players like Broadcom, NVIDIA, and AMD try to accelerate their workloads. They discuss the difference in Fractile’s newer approach on model architecture with a full-stack team in the current chip landscape and the technical bets they’re making in that direction. Walter also talks about compressing the gap between the chip design cycle and its payoff period, and making a generational leap in AI inference to realize the bet in volume against the value to be captured.