Richard Sutton just won the Turing Award for inventing reinforcement learning -- the training method that sits inside every system that learns by doing.So when he says today's large language models are a dead end, it is worth hearing exactly why:"I consider reinforcement learning to be basic AI. What is intelligence? The problem is to understand your world. Reinforcement learning is about understanding your world, whereas large language models are about mimicking people, doing what people say yo
1:31
Richard Sutton just won the Turing Award for inventing reinforcement learning -- the training method that sits inside every system that learns by doing.So when he says today's large language models are a dead end, it is worth hearing exactly why:"I consider reinforcement learning to be basic AI. What is intelligence? The problem is to understand your world. Reinforcement learning is about understanding your world, whereas large language models are about mimicking people, doing what people say yo
6.2K views2 weeks ago
x.comKarl Mehta
Today's robots are too unreliable and slow for real industrial work. Reinforcement learning changes that.Meet KinetIQ Ascend: our robots practice real production tasks and improve predictably towards industrial reliability standards.How it works 🧵
1:40
Today's robots are too unreliable and slow for real industrial work. Reinforcement learning changes that.Meet KinetIQ Ascend: our robots practice real production tasks and improve predictably towards industrial reliability standards.How it works 🧵
8.4K views3 weeks ago
x.comhr0nix
Reinforcement learning arm trained in MuJoCo and deployed onto esp32Inspo: @H0meMadeGarbage
0:28
Reinforcement learning arm trained in MuJoCo and deployed onto esp32Inspo: @H0meMadeGarbage
39.3K views2 weeks ago
x.comJack Melzer
Andrej Karpathy, OpenAI founding member, explains why the technique training every frontier model is "terrible" -- and the one thing all of them are still missing."Reinforcement learning is a lot worse than I think the average person thinks. Reinforcement learning is terrible. It just so happens that everything that we had before it is much worse.""You're given a problem, you try hundreds of different attempts. And then you check the back of the book, and you see this one and that one got the co
2:27
Andrej Karpathy, OpenAI founding member, explains why the technique training every frontier model is "terrible" -- and the one thing all of them are still missing."Reinforcement learning is a lot worse than I think the average person thinks. Reinforcement learning is terrible. It just so happens that everything that we had before it is much worse.""You're given a problem, you try hundreds of different attempts. And then you check the back of the book, and you see this one and that one got the co
18.2K views2 weeks ago
x.comKarl Mehta
Is this the first published demonstration of end-to-end, vision-based RL on production VLAs, trained on real bimanual humanoid hardware under true deployment conditions?What if robots could get better at their jobs the same way humans do?By practising and learning from their own mistakes on the actual factory floor?@TheHumanoidAI just dropped something new: their new reinforcement learning method (KinetIQ Ascend) that lets their humanoid robots improve through real-world trial and error instead
1:39
Is this the first published demonstration of end-to-end, vision-based RL on production VLAs, trained on real bimanual humanoid hardware under true deployment conditions?What if robots could get better at their jobs the same way humans do?By practising and learning from their own mistakes on the actual factory floor?@TheHumanoidAI just dropped something new: their new reinforcement learning method (KinetIQ Ascend) that lets their humanoid robots improve through real-world trial and error instead
5.9K views3 weeks ago
x.comIlir Aliu
Physics-based simulations for reinforcement learning are kind of mind-blowing.I was thinking about it today.It's like being Neo in The Matrix learning martial arts, but it's not sci-fi.Our AI is already experiencing it and it is strange to think that THIS could be its own physics-based simulation because we're not raised to think that way, yet at the same time, it feels so natural as a hypothesis for this world. The discomfort is in asking oneself questions that one can't answer.There's a Korean
0:50
Physics-based simulations for reinforcement learning are kind of mind-blowing.I was thinking about it today.It's like being Neo in The Matrix learning martial arts, but it's not sci-fi.Our AI is already experiencing it and it is strange to think that THIS could be its own physics-based simulation because we're not raised to think that way, yet at the same time, it feels so natural as a hypothesis for this world. The discomfort is in asking oneself questions that one can't answer.There's a Korean
474 views1 week ago
x.comLiora
What if robots could remember and learn from their own fastest success, even when it came from a “lucky” trial?People often treat efficiency as something to optimize after success. Our new work, Temporal Self-Imitation Learning (TSIL), takes a different view: fastest success itself can be a useful training signal in reinforcement learning.TSIL turns rare fast successes discovered during interaction into two learning signals: adaptive temporal targets that encourage faster completion in a self-pa
0:14
What if robots could remember and learn from their own fastest success, even when it came from a “lucky” trial?People often treat efficiency as something to optimize after success. Our new work, Temporal Self-Imitation Learning (TSIL), takes a different view: fastest success itself can be a useful training signal in reinforcement learning.TSIL turns rare fast successes discovered during interaction into two learning signals: adaptive temporal targets that encourage faster completion in a self-pa
7.3K views3 weeks ago
x.comYinsen Jia
4:59
$AMD| Enterprises are accelerating Agentic AI 🧵Not Financial Advice! DYOR!1. EPYC CPUs: The Orchestration and Control BackboneAgentic AI workloads differ from traditional training-focused AI. They involve complex, multi-step workflows: intent interpretation, planning, tool-calling (APIs, databases, code execution in sandboxes), multi-agent coordination, state management, policy enforcement, and observation/feedback loops. @AMD has explicitly highlighted that agentic AI is driving a “CPU renaiss
7.3K views2 weeks ago
x.comMike
0:29
Researchers at the Norwegian University of Science and Technology have developed Olympus, a four-legged robot designed for exploring the Moon and Mars. It can perform high, controlled jumps and stabilize itself in mid-air using reinforcement learning. Its spring-powered legs help it move across craters, steep slopes, and rocky terrain where wheeled rovers may struggle, and it was unveiled last year.
2.9K views3 weeks ago
x.comSpace and Technology
1:32
$AMD is heading to $3,000 much sooner 🧵Current CPU shortage is completely misunderstood! Not Financial Advice! DYOR! Research Purpose only! TLDR: Running more agents or RL workloads in real-world deployment could send demand for $AMD EPYC Venice to 100-120m units equivalent to rebalance in just 2023-2026. $3,000/share should hit when AMD gets to $70-$80B Revenue per quarter(15--18x P/S). EPYC Venice Flagship is estimated to sell in the $13,000-$20,000 depending on vol discount. EPYC Turn is ~$7
7.6K views4 weeks ago
x.comMike
0:25
I just spent my Sunday "Vibe Roboting". 5.6 Sol is an absolute beast in Blender compare to what was possible few months ago. For the first time, I feel like we can actually “vibe robot.”I sketched a robot on a paper, enhanced it with AI, asked GPT 5.6 Sol to generate all the parts for 3D printing, and then verify the design using reinforcement learning.Still a toy project but I feel like the future where everyone will be able to generate their own robot is closer than expected.What a time to be
4.3K views1 week ago
x.comDefend Intelligence (Anis Ayari)
2:17
Pet Care & Training: The Complete Beginner's Guide to Raising a Happy and Well-Behaved Pet.
9 views3 weeks ago
YouTubePets caring
0:57
$INTC $TSM $NVDA $AMD Outstanding interview of LBT. It gives me the conviction to stay long the GAI infrastructure trade even after this massive run up. I love his comment that he is targeting a 10x shareholder-return for INTC - he’s already making progress towards the goal.EXECUTIVE OVERVIEWNo Priors is an AI-focused podcast hosted by Sarah Guo and Elad Gil. Guo is the founder of Conviction and a former Greylock investor, while Gil is a serial entrepreneur and investor associated with Color Hea
102.5K views1 month ago
x.comTheValueist
2:43
A year ago, when I left Tesla Optimus to co-found Mondo, I made a promise to my son: I would build him a little robot buddy — something like the robot friends from our bedtime stories, but real.At the time, it was not really a product statement. It was a father’s promise to a little boy I love more than anything — a boy who loved stories about a child and his robot companion. Over the past year, that promise slowly became real — Beni the robot.Beni follows you, films you in 4K, moves across diff
18.4K views2 weeks ago
x.comShuo Yang
1:22
This is why I love Elon Musk, the richest man in the world, who just admitted he was wrong about his competitor Anthropic and said even my worst enemies can attack me on this platform."A major failure mode is when the ego-to-ability ratio becomes far greater than one. If your ego gets too high relative to your ability, you break the feedback loop with reality. In AI terms, you break your reinforcement learning loop.You want a strong learning loop, which means taking responsibility, minimizing eg
22.3K views1 week ago
x.comDoge Tipping
See more
Static thumbnail place holder