Rupom Ghosh

Autonomy in Design and Prototyping The platform functions as an agentic workspace where you can input text descriptions or reference images to build out polished visual artifacts. Whether you need an interactive user interface, a vertical timeline for a change log, or a full slide deck, the agent autonomously plans, writes, and self-checks its work. It comes pre-loaded with twelve specialized design skills, including dashboards, landing pages, and editorial typography, allowing it to handle complex layouts with hover states and functional tabs already wired up.

Even the most advanced models are still struggling with basic human intuition. While the top AI agents hover around 30% performance, the novice human baseline remains significantly higher at 64%. This performance gap is most visible in tasks requiring precise timing, spatial navigation, and long-horizon coordination in open-world environments. Despite these challenges, general chat models are successfully analyzing visual screens and pressing keys to interact with environments they were never specifically trained to play.

@beffjezos Imagine not doing any work from here

Imagine vibe coding from here

@CognyAI Ngl pretty cheap

Claude runs your SEO. $9/mo.

@SawyerMerritt This is will acc be efficient for freight operations

Walk-around of the redesigned Tesla Semi in Las Vegas.

@paulg Thank god i have garage 📝💯🤣

It could actually be a significant problem that Europe doesn't have enough garages. This sounds like a joke, but I'm serious. Garages let you work on stuff that doesn't matter yet, which is how big things often start. The outliers of ideas need the outliers of space.

@MarioNawfal @muskonomy Musk doing everything to get those big tech companies go brokeee

🇺🇸 xAI is only using 11% of its insane 550,000 NVIDIA GPUs right now (just ~60k active). Meta & Google are hitting 43-46%. Scaling these monster clusters is still wild; even the best are figuring it out. Colossus is sleeping… for now. Source: @muskonomy

@gdb I think you really need to make this open-source

codex for helping you to ship fast

Adobe is cooking again, but don't expect to run this on your local rig yet. The technical paper for TokenLight shows state-of-the-art performance in relighting everything from skin to transparent objects. But as with most Adobe research, this is currently a closed-door project. While other labs are open-sourcing their code, Adobe is likely refining this for a future Firefly or Photoshop update. At Grivus, we’re tracking if these attribute tokens will eventually be accessible to developers. For now, it’s a masterclass in what’s possible when transformers actually understand light.

The speed of 10 m/s is the new world record for humanoids. Unitree just pushed their flagship H1 to 36 km/h. This isn’t just about raw power, it’s a masterclass in whole-body control systems and high-frequency stability. The Specs: Weight: 62 kg ⚖️ Leg Length: 0.8 m 🦵 Velocity: 10 m/s ⚡ We are seeing the gap between human-like and super-human motion close in real-time. Dynamic motion is no longer a bottleneck for robotics.

@elonmusk @PeterDiamandis Time to bust 🤣

@PeterDiamandis $10T or bust

.@elonmusk just crossed $800 billion — roughly 2.7% of the entire US GDP. The last person to hold that much of the American economy? John D. Rockefeller in 1913. It took a century for anyone to match him. Rockefeller had oil. Musk has the future.

@zavxai ThinkPad is the real OG

ThinkPad Prove me wrong.

@gdb Let me know when this is open sourced....

codex for startup ideas

Interviewer : "Where do you see yourself in 5 years" Me:

😂

Alibaba’s ATH Lab is officially cooking the competition. First, they dominated the leaderboards with Happy Horse (now ranked #1, beating Seedance 2.0). Now, they’ve released Happy Oyster. While everyone is focused on 2D video, ATH Lab is solving interactive 3D world models. This is the Google Genie 3 competitor we’ve been waiting for. If you’re a developer and you aren't on this waitlist, you’re missing the next architectural shift in AI.

Stop renting the cloud to build 3D worlds. Nvidia just dropped Lyra 2, and the file size is the real story: 131 Megabytes. It turns raw video into interactive 3D point clouds (Gaussian Splatting) that you can run on almost any consumer GPU. The breakthrough: "Long Horizon" consistency. You can walk away for an hour, come back, and the scene is identical. No drift. No flickering. High-fidelity 3D environments used to require gigabytes of data. Now, they fit on a thumb drive. The barrier to "Spatial AI" just evaporated.

The biggest AI skill isn’t prompting. It’s thinking clearly. AI just amplifies how well you already think.

Everyone is talking about AI tools. Almost nobody is talking about AI systems. Tools save minutes. Systems make money.

Product integrity is the final boss of AI video. OmniShow just beat it. We’ve seen digital avatars before, but they usually "melt" the product they’re holding. OmniShow uses a new reference-guidance system that keeps both the person AND the product pixel-perfect. - Pose Control: Input a skeleton video to dictate exact hand/finger movements. - Voice Sync: Zero-shot audio-to-video cloning. - Zero Bleed: The product remains "brand-safe" and recognizable. This is how you scale high-converting video ads without a production crew.