Subvolts take Reinforcement learning pioneer Richard Sutton argues LLMs are a dead end, advocating for continual learning architectures. Sep 26, 2025 - Developers
- Understand a foundational critique of LLMs and explore alternative research directions for agentic AI.
- Business
- Grasp the potential long-term limitations of current LLM investments and the need for future-proof AI architectures.
Features Richard Sutton, a Turing Award winner and father of RL, directly challenging the current LLM paradigm.
Richard Sutton – Father of RL thinks LLMs are a dead end
Audio streams from Dwarkesh Patel. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Richard Sutton, a pioneer in reinforcement learning and Turing Award winner, argues in this Dwarkesh Podcast episode that large language models (LLMs) are a dead end. He believes current LLM architectures cannot learn on the job and will be superseded by new architectures enabling continual, on-the-fly learning, similar to how humans and animals learn.
Imagine computers learning like babies, constantly picking up new skills from experience. Richard Sutton, a top AI researcher, says current AI models called LLMs can't do this well. He thinks we need a new kind of AI that learns continuously, like animals do, to truly advance.
This episode offers a critical perspective on LLMs from a foundational AI researcher. Listeners will learn about Sutton's argument for a paradigm shift beyond current LLM scaling, focusing on continual learning. It's for AI researchers and enthusiasts questioning the long-term viability of LLMs. This challenges the current trajectory of labs like Google Gemini and OpenAI, pushing for new architectural approaches.
- HostDwarkesh Patel
- GuestRichard SuttonFather of Reinforcement Learning, Turing Award winner
- Guest: Richard Sutton
- Guest Credentials: Father of Reinforcement Learning, 2024 Turing Award winner
- Episode Title: Richard Sutton – Father of RL thinks LLMs are a dead end
- Show: Dwarkesh Podcast
- Length: 66 minutes
- Sutton's Claim: LLMs aren’t capable of learning on-the-job and will be rendered obsolete by new architectures.
Richard Sutton is the father of reinforcement learning, winner of the 2024 Turing Award, and author of The Bitter Lesson. And he thinks LLMs are a dead end.
After interviewing him, my steel man of Richard’s position is this: LLMs aren’t capable of learning on-the-job, so no matter how much we scale, we’ll need some new architecture to enable continual learning.
And once we have it, we won’t need a special training phase — the agent will just learn on-the-fly, like all humans, and indeed, like all animals.
This new paradigm will render our current approach with LLMs obsolete.
In our interview, I did my best to represent the view that LLMs might function as the foundation on which experiential learning can happen… Some sparks flew.
A big thanks to the Alberta Machine Intelligence Institute for inviting me up to Edmonton and for letting me use their studio and equipment.
Enjoy!
Watch on YouTube; listen on Apple Podcasts or Spotify.
Sponsors
* Labelbox makes it possible to train AI agents in hyperrealistic RL environments. With an experienced team of applied researchers and a massive network of subject-matter experts, Labelbox ensures your training reflects important, real-world nuance. Turn your demo projects into working systems at labelbox.com/dwarkesh
* Gemini Deep Research is designed for thorough exploration of hard topics. For this episode, it helped me trace reinforcement learning from early policy gradients up to current-day methods, combining clear explanations with curated examples. Try it out yourself at gemini.google.com
* Hudson River Trading doesn’t silo their teams. Instead, HRT researchers openly trade ideas and share strategy code in a mono-repo. This means you’re able to learn at incredible speed and your contributions have impact across the entire firm. Find open roles at hudsonrivertrading.com/dwarkesh
Timestamps
(00:00:00) – Are LLMs a dead end?
(00:13:04) – Do humans do imitation learning?
(00:23:10) – The Era of Experience
(00:33:39) – Current architectures generalize poorly out of distribution
(00:41:29) – Surprises in the AI field
(00:46:41) – Will The Bitter Lesson still apply post AGI?
(00:53:48) – Succession to AIs
This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.dwarkesh.com
- Reinforcement Learning (RL)
- Richard Sutton
- A way for AI to learn by trial and error, getting rewards for good actions and penalties for bad ones.
- LLMs
- Richard Sutton
- Large Language Models, the AI systems behind tools like ChatGPT that are good at understanding and generating text.
- Architecture
- Richard Sutton
- The underlying design or structure of an AI model.
- Continual Learning
- Richard Sutton
- The ability of an AI to keep learning and adapting over time, without forgetting what it already knows.
1 h 371 h 201 h 37AI researchers debate how close we are to recursive self-improvement
WithJohn Schulman, Beren Millidge, Charlie O’Neill
2 h 21
1 h 10Sonnet 4.5 & the AI Plateau Myth — Sholto Douglas (Anthropic)
WithSholto Douglas
47 minThe good, the bad, and the future of AI agents
WithDavid Hershey
1 h 17OpenAI Sora 2 Ignites Likeness Debate
WithTilly Norwood
1 h 05The Decentralized Future of Private AI with Illia Polosukhin - #749
WithIllia Polosukhin
Questions people ask
- What is Richard Sutton's main argument about LLMs?
- Sutton believes LLMs are a dead end because they cannot learn on the job. He argues new architectures are needed for continual learning.
- What does Sutton propose as the future of AI learning?
- He suggests a paradigm where agents learn on the fly, like humans and animals, without a separate training phase, rendering current LLM approaches obsolete.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Dwarkesh Patel.