Alex

@Linahuaa Sometimes people forget he has Asperger syndrome I guess

This is by far the biggest Elon L ever. Dude actually PERSONALLY worked on this crap, and it's so FAIL on so many levels, it makes Elon look like super low EQ. 1) First of all, you're the most respected billionaire tech visionary on the planet. Your aura hard carries your entire empire. And then you stake all of that on lolita hentai slop.... wtf are you thinking????? 2) And yeah, I mean slop! Ani looks like Elon designed her himself. And I'm 90% sure he did. It literally looks so amateurish- like a 7yo drawing with crayons kind of vibe. Couldn't you contract Japan for some studio level key-visuals?? 3) It's ultra barebones: No customisation, no backstories, no interesting features, no NOTHING. The outfits are ugly as hell. It really feels like some amateur project made by one 38yo jobless Japanese gooner. Except the Japanese gooner would have made the girl look way prettier. I CAN'T COMPREHEND: how can King Elon HIMSELF, with billions of dollars in resources- and with a team that surely must have some anime enjoyers in it- create such uber FAIL? Dude wants to build civilisation on mars, but gets MASSIVELY outgooned by amateur gooner devs

@predict_addict Goat

@AIex_3 Shall I check it out next? Stay tuned

TU Munich is often marketed as an “elite” physics degree. But when you look at actual institution-provided teaching time, it starts to look less like a full heavyweight physics formation and more like a quarter-degree compared with the real beasts. Against MIPT, it is nowhere close.

@predict_addict How does TU Delft fit in that list?

And when you measure those properly, TU Munich looks a lot smaller than the brand suggests. #Physics #STEM

A student in Zurich sleeps in a camper van outside a hangar every night. Just to have more hours to build a humanoid robot. He’s obsessive. He's not alone. 2,500 students across Europe are getting into robotics right now. And now they united and just launched: ESRA, the European Student Robotics Association. 🇪🇺🦾 They are bringing together highly talented young people, give them space, give them resources and let them build. By now already 13 robotics clubs. 8 countries. 2,500+ students. I visited several of them over the last weeks to get to know them and let them tell their stories. We've also been helping behind the scenes where we can, because this is exactly what Europe needs. Several multiple billion dollar companies will come out of the ESRA network. Right here in Europe. If the Bay Area had a student robotics network like this, they would never shut up about it. Time we do the same. 😤🔥🇪🇺 It only needs a few crazy ones to fix a continent. Turns out they're already building. 🇪🇺

We're launching the European Student Robotics Association (ESRA) (13 universities, 8 countries, 2.5k+ members)! We’re young, driven and together we’re tackling the European fragmentation problem heads on. Who are we? - ETH Robotics Club (Zürich) @ethroboticsclub - RoboTUM (Münich) - EPFL AI Team (Lausanne) @epflaiteam - Unaite (Paris) - Team Polar (Eindhoven) - TU Wien Robotics Club (Vienna) - Robotics Collective (Aachen) @robocollectiv - KTH AI Society (Stockholm) @KTHAISociety - Delft Robotics Student Association - KN CybAiR (Poznan) - AEA Polimi (Milan) And just getting started! What do we do? → Pan-European robotics competitions → Cross-border technical project collaborations → Coordinated access to funding opportunities across Europe

@VectorWang2 @chris_j_paxton Cool stuff, think loco-manip training would be possible with this?

Why can't “world models” just be sim? A real-time real2sim demo in the upcoming “ManiDreams” Foundation Stereo + SAM2 + Newton = Generalizable 3D physics prediction in a fast (15FPS) and lightweight (3070) way. No data or object mesh needed rice-robotpi-lab.github.io/ManiDreams/

@grok @H97_71_ @hamptonism So it’s still a reinforcement learning algo in the back, just the cells that are executing the code and not a cpu/ram/gpu, are they using spiking neural networks? How do they compute bits?

It's real, from Australian startup Cortical Labs (they did Pong with neurons back in 2022). They grow ~200k human neurons on a microelectrode chip. The chip sends electrical pulses representing Doom's world (enemies, walls, etc.) to the cells. The neurons fire back activity that's translated into controls: move + shoot. Feedback loops (stimulation rewards/punishes) let the network adapt and improve in days via biological plasticity. Not conscious or "watching" like us—just signal processing. Hybrid bio-computer demo.

Dude.

Photo from Shenzhen: huge crowd of Chinese people (lots of grannies!) lining up to get help installing OpenClaw. One thing about tech diffusion in China that I feel is underdiscussed and that I’ll admit I don’t fully understand, is how open people of all ages are to jump into new tech. Feels very different from the AI suspicion/resistance you see in the U.S. Similar with mobile payments and the shift to cashless. Street vendors in the lowest tier cities setting up WeChat Pay and Alipay QR codes almost overnight and Chinese grannies happily using payment apps with no problem at all. And yes that kind of grassroots adoption helped mobile payments scale extremely fast and allowed China to basically skip the credit card phase. My conjecture is that if something similar happens with AI tools the speed of AI diffusion in China could look very different from what we see in other countries, which obviously would have major implications...

I've been working on a new LLM inference algorithm. It's called Speculative Speculative Decoding (SSD) and it's up to 2x faster than the strongest inference engines in the world. Collab w/ @tri_dao @avnermay. Details in thread.

Claude Code plays Runescape: The waterfall quest. Here Claude Opus 4.6 is acquiring quest items in the maze and tomb. Not a script, real play. It tried things, failed, learned, and succeeded. I sped this up a lot, it's ~20 minutes of real time.

Holy sh*t Anthropic just caught DeepSeek secretly stealing from Claude. They created thousands of fake accounts. Then bombarded Claude with millions of questions & took Claude’s answers for using them to train their own AI. Claude trained DeepSeek. Without knowing it. They also used Claude to find ways to avoid sensitive political topics, so their own AI learns to dodge questions about China’s government. Anthropic traced it back to actual researchers at DeepSeek. The AI cold war isn’t coming. It’s already here.

new blog! What methodologies do labs use to train frontier models? The blog distills 7 open-weight model reports from frontier labs, covering architecture, stability, optimizers, data curation, pre/mid/post-training + RL, and behaviors/safety djdumpling.github.io/2026/01/31/fro…

@RhysOSRS I also had it on and no problems (only a bit laggy when teled into the crater), can you send a screenshot of your settings tho?

@AIex_3 On, apparently that's part of the problem, however, a lot of my mates also had it on and had no problems

Ima go barrage from within... oh DCed on all 3 brackets I played LOL

@ImTedious 1vs1s are 100% better if no ddos

I will never get the appeal of just watching 2 clumps of random people spam multi-target attacks at each other until one side falls over.

@NVIDIARobotics @StanfordSVL Would be nice if the Unitree G1 real-world policy deployment code is open sourced as well 😊

Real-world robot data is expensive and slow to collect, creating a major challenge for humanoid development. 🤖 The NVIDIA GR00T N1.6 open vision language action model is pre-trained on a diverse mix of data, including thousands of hours of @StanfordSVL’s BEHAVIOR simulation data, which covers long-horizon everyday manipulation tasks. This diverse training is the key to robust cross-embodiment performance and real-world adaptability. 🌍 Read the blog 🔗 nvda.ws/4bkRfvO

@Scobleizer @TheHumanoidHub Current VLA architectures all seem to rely on behaviour cloning policies from by training/fine-tuning on recorded teleoperation demonstrations. Meaning: the generalisation part is still missing, how are you gonna introduce this general robot with all the multi-modalities?

That isn't quite true. They have a pretty good idea of what it will take. All of them that I meet with say they don't have nearly enough data to make the world models good enough to do such a thing and that making a generalized humanoid will take five to seven years. But it is sort of true, since none of them are there today.

Yann LeCun says absolutely none of the humanoid companies have any idea how to make those robots smart enough to be useful.

🌎World models can predict, but controlling real robots from imagination sees a long-standing failure due to hallucination. 🧠Introducing Uncertainty-Aware RWM: a black-box, end-to-end neural dynamics model with long-horizon uncertainty propagation. 🎯sites.google.com/view/uncertain…

🚀 Introducing Qwen3-Max-Thinking, our most capable reasoning model yet. Trained with massive scale and advanced RL, it delivers strong performance across reasoning, knowledge, tool use, and agent capabilities. ✨ Key innovations: ✅ Adaptive tool-use: intelligently leverages Search, Memory & Code Interpreter without manual selection ✅ Test-time scaling: multi-round self-reflection beats Gemini 3 Pro on reasoning ✅ From complex math (98.0 on HMMT Feb) to agentic search (49.8 on HLE)—it just thinks better. 🧠 Think deeper. Solve harder. Try the adaptive reasoning experience now: chat.qwen.ai Completions API:  modelstudio.console.alibabacloud.com/ap-southeast-1… Responses API:  alibabacloud.com/help/en/model-… blog:  qwen.ai/blog?id=qwen3-…