AI researchers seek to understand intelligence well enough to create beings of greater intelligence than current humans. Reaching this profound intellectual milestone will enrich our economies and challenge our societal institutions. It will be unprecedented and

More on LLMs, RL, and the bitter lesson, on the Derby Mill podcast.

@RichardSSutton @dwarkesh_sp no you didn't misspoke there, Richard. I miss quoted, the video and caption says " training " itself. Apologies, 🙏

This is a reasonable take on the podcast. One thing I would add is that people underestimate just how much babies learn as opposed to what they are born with. One of the big differences between us and other animals might just be that we rely much more on learning because we have

Still timely

This will age poorly. I largely have an optimistic view of LLMs; I use multiple LLM tools daily, and I don't think the LLM tech stack is a bubble—it will create a lot of value. I disagree that the length of tasks that LLMs can do has been doubling every 7 months. There are tasks

This is a thoughtful writeup, as I expect from Rod Brooks. I think he is right on the importance of input representation, touch, and physical safety in deployment. I also think he underestimates the potential for representation and subgoal discovery with reinforcement learning.

Two things can be true at the same time: 1. Without additional advances, LLMs won't get us to general intelligence. 2. Even without additional advances, LLMs will radically transform the economy.

💯

Dwarkesh and I had a frank exchange of views. I hope we moved the conversation forward. Dwarkesh is a true gentleman.

Mike is a powerful thinker and researcher. Very well deserved.

Is scale all you need? Or is there still a role for incorporating domain knowledge and inductive bias? While I was in Heidelberg, I took some time to write a short essay on this question called "The Bittersweet Lesson". https://t.co/DQEItqXomF #HLF25

Thanks to @the_logic for including Amii in this deep-dive into Edmonton's thriving AI ecosystem. The article highlights our world-class work in RL and features Cam Linke, @RichardSSutton, and many other brilliant minds. Read:

For those really into it, here are another 50 minutes of my views on planning and action selection in options-based AI agents (like in the Oak architecture). https://t.co/B2vqxKofDW

Have people seen this prescient 2001 post by @RichardSSutton on self-verification? "An AI system can create and maintain knowledge only to the extent that it can verify that knowledge itself". This sentiment underpins much LLM reasoning research today. https://t.co/gK3DOwqYm8

Designing a robot at Keen that is robust enough for online and continual reinforcement learning was fun. Robotics is so much easier when you take a scalable approach like learning instead of using human-designed sims or using data from human operation.

This is a big deal. It is the first large-scale demonstration of the advantage of real-time reinforcement learning. The recipe is scalable and requires no intervention in principle; the model can adapt forever as long as it is being used. There is no way to achieve similar

Evan Solomon is Canada’s new first minister of AI (and digital innovation)

Maximizing voluntary decision-making is the answer

The Pandemonium paper is seminal, but a little hard to find; here is a pdf: https://t.co/4wAm8jU0vt

Dwarkesh Patel is 100% right on this: AI's utility is very strongly dependent on continual learning. https://t.co/YR54QlaqZK