Your curated collection of saved posts and media
We just released an hf CLI extension to detect the best model/quant for a user's hardware and then spins up a local coding agent. Time to go local/private/free/fast for your agents thanks to open-source! https://t.co/LcVJzGCqWx
Alibaba just open-sourced OpenSandbox ( a general-purpose execution environment ) to give AI agents an isolated environment to run code safely. 8k+ Github stars βοΈ This stops your AI Agent based applications from accessing your actual host infrastructure. By removing the hardest security roadblock, this release will massively accelerate how fast developers can build autonomous Agent based tools. OpenSandbox puts the agent inside isolated runtimes like gVisor or Firecracker. You can run it locally using Docker or scale it up using Kubernetes. The system includes a code interpreter and a file system that the agent uses to complete tasks. It also manages network traffic so you control exactly what the agent accesses online. I think this will become the standard infrastructure for autonomous systems because building custom sandboxes is too dangerous for most teams.
this is what reading RL papers feels like https://t.co/PEEnvTaOe4
this is what reading RL papers feels like https://t.co/PEEnvTaOe4
Yeah, this does not sound like imminent recursive self-improvement to me. https://t.co/FhCOy4zM38
Yeah, this does not sound like imminent recursive self-improvement to me. https://t.co/FhCOy4zM38
@joelgrus https://t.co/JTgYtV8JTk https://t.co/z6LfUEoGky
.@nodejs has always been about I/O. Streams, buffers, sockets, files. But there's a gap that has bugged me for years: you can't virtualize the filesystem. You can't import a module that only exists in memory. You can't bundle assets into a Single Executable without patching half the standard library. That changes now π
Happy St. Patrick's Day, New York. https://t.co/DPCGPlFbE2
Happy St. Patrick's Day, New York. https://t.co/DPCGPlFbE2
Oh wow, Mamba-3 is here! For me, the most interesting use case of Mamba and Mamba-likes are the recent transformer attention hybrid architectures (Qwen3.5, Kimi Linear, etc.) Would be interesting to swap Gated DeltaNet with Mamba-3 (which now also has RoPE) in next gen hybrids. https://t.co/wcTg0uZ3gg
The newest model in the Mamba series is finally here π Hybrid models have become increasingly popular, raising the importance of designing the next generation of linear models. We've introduced several SSM-centric ideas to significantly increase Mamba-2's modeling capabilities
6 months of free Codex. I will use this to make Athas better. Huge thanks to everyone who made this happen β€οΈ https://t.co/vszTEpp7MU
We're shipping a new feature in Claude Cowork as a research preview that I'm excited about: Dispatch! One persistent conversation with Claude that runs on your computer. Message it from your phone. Come back to finished work. To try it out, download Claude Desktop, then pair your phone.
I've had ChatGPT-5.4 Pro working away at a project I always wondered about: how lucky are you to be alive right now? Of all the ~117B humans who ever lived, only about 1.5% had a lifestyle roughly equivalently to a middle-class person in a middle-income country today, or better. https://t.co/HFZ5XmBnC6
Petrol prices through the roof Inflation going crazy But I just signed up for Costco and got this pizza slice for 2.99 https://t.co/JpgaonMjVx
Someone built an old windows 3.1 OS inside the perplexity computer, so we can @AskPerplexity π https://t.co/UA0Vd0uPmb
@HamelHusain https://t.co/98ty7jGws1
We hosted an executive dinner at NVIDIA GTC with @Modular. 600+ people signed up, and we had to turn away more than 500. We had a packed house π₯ This is one of the only dinners where we discuss the impact of AI agents across the entire stack - from infrastructure/systems to context and agentic engineering. Some of the topics we discussed: - Is the future composed of general models or specialized models? - Are agents a systems problem or model problem? - How are non-technical users adopting agent tooling like Claude Cowork? - Where do humans spend the most time on repetitive document work? Massive thank you to the Modular team for partnering with @llama_index on this, and to every single person who showed up. To the 500+ on the waitlist - next time we're getting a bigger venue.
I thought I knew everything about autonomous vehicles, but, nope. I almost got in an argument with cofounder @dfergusonnz of @nuro about LiDAR at NVIDIA GTC during his session. He laid out that LiDAR shows safety improvements. But then said that LiDAR can see a kid all in black at night lying on the road around a turn. My answer? Matrix headlights see all that. Truth be told, his session at GTC was awesome and Nuro has amazing tech that can generalize to any vehicle. He showed it off driving in Japan for the first time driving on a left hand road it had never seen before. But seriously, been doing a ton of work today to understand the autonomous vehicle industry. Talked with NVIDIA's team, met people from General Motors, Uber, and others, attended a class by Nuro's founder, and on Friday NVIDIA is taking me for a ride. The thing that stunned me was NVIDIA's simulator. You give it a picture of a road and a prompt and it creates a simulated road for you. Then you can make it snow, rain, or any number of situations. Put pedestrians on. And all created with a fairly small model. Now I understand why NVIDIA is in love with simulators. I asked Jensen at his press conference about it. And what it means for glasses coming soon. And he was bullish on the future of glasses to show both ground truth (reality) and whatever you want the simulator to show you (the Holodeck). GTC 2030 is gonna be insane. :-) Videos will come someday. I have a ton of them and no time to edit or upload. Packed the week with too many meetings. Be back over the weekend. Or I might steal one of those simulators. The whole thing was running on a box under the demo station. You think @AlexFinn has cool inferencing computers under his desk? A Mac can't simulate any road, with any condition, and his Mac doesn't have a steering wheel. New goals. :-) Oh, and the conclusion (yes I'm burying the lede, as we say in journalism circles): Pretty much car company will have self driving cars within 18 months. Which is awesome. Fewer lives lost. The real question is who can do Robotaxi. Nuro is only planning on a few tens of thousands of vehicles. That's not nearly aggressive enough for me. Guy who works for GM was sitting next to me and he said "the companies can't figure out the liability and regulation issues yet to really scale up." Back to the argument, he said LiDARs cost in "low thousands" now. In the car industry that's a huge amount, as one analyst that I didn't catch name of, pointed out in the audience. Everyone I'm talking to says Tesla drives smoother than the others. Will see on Friday when I get my ride in NVIDIA's. My money is still on Tesla. They have all the pieces to do Robotaxi at scale and become the "Starbucks" of Robotaxis. But this market is just starting, so the next 18 months are gonna be highly entertaining. More arguments ahead. So honored to be reporting for CNQQ. Chinese Tech ETF. Investors in China are highly interested in autonomous vehicles. https://t.co/2BjIVhqvR4
π€ From perception to action. Exciting progress in embodied navigation. π Codeveloped by @LimX_Dynamics and @RealSenseai VSLAM, in collaboration with @nvidia cuVSLAM #Robotics #EmbodiedAI #AutonomousNavigation https://t.co/7gH2jtEnct
World models can predict the next frame. They can't predict the next touch. That's the gap visuo-tactile world models will close. Is the robot gripping hard enough? Is the surface rigid or soft? When exactly does contact begin and end? Vision doesn't know. Tactile does. We built @OpenGraph_Labs to capture what cameras miss. Egocentric RGB Γ 5-finger multi-taxel tactile gloves. Frame-synced. Calibrated. In-the-wild. No lab setups. No scripted pick-and-place. Just humans doing real tasks in real stores. Watch the exact moment contact happens. The pressure map lights up in sync. Every touch. Every frame. π
Holyβ¦π€― One person controlling a whole fleet of robotic hands. @HexaCercle demonstrated a system where an operator simply moves their own arms, and the motion is replicated 1:1 across multiple robotic hands in real time. The specs are wild: β’ 0.001Β° motion capture precision β’ ~3 ms transmission latency β’ 25 sensor nodes β’ 30 m wireless range, 8-hour battery β’ 3-minute quick setup But the real implication is bigger: One skilled worker can teach a whole group of robots at once. From assembly and warehouse picking to hazardous operations,human skills can now be scaled across many robots/humanoids simultaneously.
Today we're starting to test an early version of our V8 model with our community. It's much better at following prompts, 5x faster, has native 2K modes, improved text rendering and the best personalization, sref, and moodboard performance ever. Have fun! https://t.co/bc54Iod3nv
(1/N) We're launching Dreamverse. Most AI video models take minutes to generate a 5 s 1080p clip. In 4.5 seconds, we can generate 30 s 1080p clips on a single GPU. Our videos generate faster than you can watch them: stop waiting on prompts and start directing scenes live. πΉοΈDemo: https://t.co/Os8KAxJCld π Blog: https://t.co/ZJxhWmwZpV Welcome to the era of vibe-directing π
Risks and Limitations of AI in the Life Sciences. Through concrete cases studies, I discuss how data quality, proxy measures, & missing domain expertise can undermine AI science results in this video and transcript from a recent talk: https://t.co/FpEsoHsdyY
MrBeast has been laying off employees across his companies, with Viewstats seeing a 65% drop in headcount. Feastables has also seen stagnation and some decline, currently sitting at about 154 employees. This is based on LinkedIn data. https://t.co/sy82zSnB7u
Jeremy's (@jeremyphoward) and Terence's (@the_antlr_guy) backprop guide is sooo well written that 3 years later as I'm reviewing the guide, everything is just coming back so fluidly. 11/10 would read again https://t.co/9qB3RbJKVI
HSImul3R bridges the gap between perception and simulation for human-scene interactions By introducing a physics-in-the-loop bi-directional optimization pipeline, it reconstructs stable, simulation-ready 3D scenes from casual videos that can be directly deployed to humanoid robots.
https://t.co/cDAhsJrDgB
https://t.co/cDAhsJrDgB
Warwick Davis is taught how to use a bong for Leprechaun: Back 2 tha Hood (2003) https://t.co/Pc4M0mHfI3
Warwick Davis is taught how to use a bong for Leprechaun: Back 2 tha Hood (2003) https://t.co/Pc4M0mHfI3
