Really interesting and apparently effective ideas here.

Critics of renewables often point to land use as a reason not to invest in wind & solar. Land use IS a real issue for all energy assets incl renewables (unless on-site). But what critics conveniently forget is the huge land impact of fossil fuels. Aerial view of fracking sites in Wickett, Texas. 👇

Tesla influencers tried Elon Musk’s coast-to-coast self-driving, crashed before 60 miles

"global private sector investments last year of more than $2tn in green energy & low-carbon industry, dwarfing the $1tn poured into fossil fuels" Green is the way forward and ... ... "Gov who choose to ignore this, financial advisers and economists who chose to ignore this, do so at their peril.”

The myth of optimality in human movement science www.sciencedirect.com/science/arti...

This is absolutely true -- this is a superb and much-needed consolidation of so much of modern RL. Kevin, inquiring minds want to understand the process you use to put this artwork together! @sirbayes.bsky.social Perhaps this is also the ultimate benchmark for Gemini Deep Research reports. ;-p

Congratulations to UBC’s student hacking club Maple Bacon for making history at hacking convention DEF CON 2025! Read more about their win in our new article: www.cs.ubc.ca/news/2025/08...

I don’t need a self driving car. I need a vehicle that runs between major areas and can transport large numbers of people at once to reduce environmental impact…wait it’s a train. I’m describing trains. Build more goddamned TRAINS, America.

Where do some of Reinforcement Learning's great thinkers stand today? Find out! Keynotes of the RL Conference are online: www.youtube.com/playlist?lis... Wanting vs liking, Agent factories, Theoretical limit of LLMs, Pluralist value, RL teachers, Knowledge flywheels (guess who talked about which!)

WATCH & SHARE: This is very good. A prominent youtuber is actually giddy (but also very thorough in covering the joys, advantages, issues and challenges) about how trying biking for transportation in London has changed his life. Easy to watch & more convincing probably than 99% of public campaigns.

Really like this Dale Schuurmans talk at RLC 2025, touching on what LLMs can and cannot do, from first principles. youtu.be/-2URALk7IfQ?...

Excellent talk by Joëlle Pineau at the Reinforcement Learning Conference (RLC 2025) youtu.be/PZcaMKxljxc?...

My daughter knew I would enjoy the evolutionary dynamics (an evolutionary stable state / Nash equilibrium) described in this 3 minute video from KQED’s excellent Deep Look team: These Lizards Have Been Playing Rock-Paper-Scissors for 15 Million Years #evolution #nature

Excellent and fascinating Veritaseum video about the history and science of rubber. youtu.be/AFXLZ7FEJc4?...

Table For One Please.

Box, run, crash: China’s humanoid robot games show advances and limitations

2nd worst forest fire season thus far in Canada: 70,000 km^2 burned: the area of New Brunswick or Ireland. Action on reducing emissions, please.

@mark-carney.bsky.social Time to create a car market that supports EVs, i.e., similar to Europe and Australia. Would love to see Renault 5 EVs here. Keep the current EV targets. The country (and the world) is on fire. And the era of the intertwined Canadian-American car manufacturing is over.

The latest must-watch from @weareyellowdot youtube.com/shorts/pLTkO...

Nature never fails to amaze!

I very much like this idea. Consider this vision:

This is insane. youtu.be/hCQCP-5g5bo?...

RL either going OOD or robo-breakdancing

The new Unitree R1 starts at $5900, contains built in multimodal ai integration, and is only 25kg.

Kids playing Ozzy's Crazy Train on xylophones

Complex benchmarks for zero-sum games that come with exact scores? Check. Fairly surprising result that policy gradients do well on these benchmarks? Also check.

This is how it looks when a Sea Anemone runs away from a Starfish.

Humans are crazy good at this control-of-movement stuff!

🔥 Presenting Relative Entropy Pathwise Policy Optimization #REPPO 🔥 Off-policy #RL (eg #TD3) trains by differentiating a critic, while on-policy #RL (eg #PPO) uses Monte-Carlo gradients. But is that necessary? Turns out: No! We show how to get critic gradients on-policy. arxiv.org/abs/2507.11019

very worthwhile read -- David Suzuki interview: "We have to get climate change and species extinction, all these other problems, out of the political realm and make it something that confronts all of us." www.cbc.ca/radio/frontb...

Brutal heat wave here in Europe currently. Let's all make sure that the ever-growing impacts of climate change are part of daily conversations. Electrify everything for a better future for everyone.

In our latest paper, we discovered a surprising result: training LLMs with self-play reinforcement learning on zero-sum games (like poker) significantly improves performance on math and reasoning benchmarks, zero-shot. Whaaat? How does this work?

This was a great long-term effort from @martinklissarov.bsky.social, Akhil Bagaria, and @ray-luo.bsky.social, and it led to a great overview of the ideas behind leveraging temporal abstractions in AI. If anything, I think this is very useful resource for anyone interested in this field!

Wow!!

More Veo 3 magic, check this out!! 🙀 😂🏅 youtu.be/77xxXfKUBfc?...

Area burned by fire in Canada this summer are off to crazy fast start (black line below), second only to 2023 (blue line below), when about 5% of the entire forest area of Canada burned.

A wonderfully ambitious direction. Intuitively it would seem difficult to discover the best algorithms and architectures, eg, Quicksort, transformers, in code space. But intuitions can so easily be wrong.

Spectacular dynamic quadruped motions from Jemin Hwangbo's RaiLab at KAIST. Random Sampling MPC-style footstep planner + foostep-conditioned tracking policy. All done on the RaiLab quadruped. Amazing. (www.youtube.com/watch?v=EZbM...

📢 I'm very excited to release AgarCL, a new evaluation platform for research in continual reinforcement learning‼️ Repo: github.com/machado-rese... Website: agarcl.github.io Preprint: arxiv.org/abs/2505.18347 Details below 👇

Here's the third and final part of Slater Stich's "History of diffusion" interview series! The other two interviewees' research played a pivotal role in the rise of diffusion models, whereas I just like to yap about them 😬 this was a wonderful opportunity to do exactly that!

I have cleaned a bit my lecture notes on Optimal Transport for Machine Learners arxiv.org/abs/2505.06589

I'm looking for good set of slides that document the growing energy demands of compute (AI/datacentres/crypto) and why CS students should care, e.g., incentives to maintain fossil-fuel generation to meet expanding demand and connections to climate change. Any ideas?

The new foundation model from physical intelligence - pi-0.5 - works zero-shot in multiple homes. They tested at airbnbs around san fransisco. I also really love the robot design. blog: www.physicalintelligence.company/blog/pi05

Great overview of the importance of RL in training language models.

Because it's been a shit week, here's some good news to cheer you up. A thread with 10 cool stories from around the world: 1. China now has more #wind and #solar than thermal generating capacity and is installing 1 million solar panels EVERY SINGLE DAY. theprogressplaybook.com/2025/03/25/i...

Can anyone recommend good material (slides or material) that covers the basic arc of GANs / VAEs / diffusion models, at a high level? This would be for a CS graphics/animation class, to provide a one-lecture overview of the basics of generative models. Looking to avoid reinventing the wheel on this.

Excellent talk!

I find it interesting that RL-based tracking seems to have generally won out for bipedal/humanoid motions, while "pure objective"-based RL has largely won out for dynamically agile quadruped skills. Maybe due to both data availability and exploration difficulty, for these two cases.

Love the smooth, dynamic motions here. Looks like RL-based motion tracking (in the style of DeepMimic, DeepLoco, Drecon, SuperTrack, UHC, PHC and so many others) has been shown to work so well on so many bipedal robots. www.youtube.com/watch?v=I44_...

It is an honor to receive a Killam Accelerator Research Fellowship. Thank you to everyone involved!