Subvolts take Diffusion models promise faster, cheaper LLMs for real-time AI applications. Jan 4, 2026 - Developers
- Learn about diffusion models and how they can improve real-time AI applications like coding and voice assistants.
- Business
- Understand how new LLM architectures could lead to more cost-effective and scalable AI solutions for enterprise use.
This interview explores a promising new AI architecture that could significantly improve LLM performance and applications.
#310 Stefano Ermon: Why Diffusion Language Models Will Define the Next Generation of LLMs
Audio streams from Craig S. Smith. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
This episode of Eye on AI features Stefano Ermon discussing diffusion language models as the next generation of LLMs. He explains how their parallel, inference-first approach differs from current token-by-token generation, potentially improving speed, cost, and scalability for real-time AI applications like coding and voice systems.
Current AI text generators work one word at a time, which is slow. Diffusion models work differently, generating text in parallel, making them much faster and cheaper. This could power instant AI assistants and better coding tools.
This episode offers a deep dive into AI architecture, explaining how diffusion models could overcome limitations of current LLMs. It's for AI researchers, developers, and product managers interested in the future of generative intelligence. The discussion on inference efficiency and real-time applications is highly relevant as labs like Google Gemini and OpenAI push for more scalable and responsive AI systems.
- HostCraig S. Smith
- GuestStefano Ermon
- Guest: Stefano Ermon
- Show: Eye on AI
- Length: 52 minutes
- Topic: Diffusion Language Models vs. Autoregressive LLMs
- Key Benefit: Potential for improved speed, cost, and scalability
This episode is sponsored by AGNTCY. Unlock agents at scale with an open Internet of Agents.
Visit https://agntcy.org/ and add your support. Most large language models today generate text one token at a time. That design choice creates a hard limit on speed, cost, and scalability. In this episode of Eye on AI, Stefano Ermon breaks down diffusion language models and why a parallel, inference-first approach could define the next generation of LLMs. We explore how diffusion models differ from autoregressive systems, why inference efficiency matters more than training scale, and what this shift means for real-time AI applications like code generation, agents, and voice systems. This conversation goes deep into AI architecture, model controllability, latency, cost trade-offs, and the future of generative intelligence as AI moves from demos to production-scale systems. Stay Updated: Craig Smith on X: https://x.com/craigssEye on A.I. on X: https://x.com/EyeOn_AI (00:00) Autoregressive vs Diffusion LLMs (02:12) Why Build Diffusion LLMs (05:51) Context Window Limits (08:39) How Diffusion Works (11:58) Global vs Token Prediction (17:19) Model Control and Safety (19:48) Training and RLHF (22:35) Evaluating Diffusion Models (24:18) Diffusion LLM Competition (30:09) Why Start With Code (32:04) Enterprise Fine-Tuning (33:16) Speed vs Accuracy Tradeoffs (35:34) Diffusion vs Autoregressive Future (38:18) Coding Workflows in Practice (43:07) Voice and Real-Time Agents (44:59) Reasoning Diffusion Models (46:39) Multimodal AI Direction (50:10) Handling Hallucinations
- Autoregressive
- Stefano Ermon
- Generating text one piece at a time, like writing a sentence word by word.
- Diffusion language models
- Stefano Ermon
- A new type of AI model that generates text in parallel, potentially much faster.
- Inference
- Stefano Ermon
- The process of using a trained AI model to generate output, like text or images.
- RLHF
- Stefano Ermon
- Reinforcement Learning from Human Feedback, a method used to train AI models.
55 min42 min51 min34 minUkraine's Secret Weapon: The Points System Winning the Drone War | Andrii Hrytseniuk
WithAndrii Hrytseniuk
Same guests, other shows
38 min
1 h 03The Race to Production-Grade Diffusion LLMs with Stefano Ermon - #764
WithStefano Ermon, Stefano Ermon
Questions people ask
- What is the main limitation of current LLMs?
- Most current large language models generate text one token at a time, which limits their speed, cost, and scalability.
- How do diffusion language models differ?
- Diffusion language models use a parallel, inference-first approach, contrasting with the sequential token-by-token generation of autoregressive systems.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Craig S. Smith.