Cursor can now show your agent's context usage as an interactive report in a canvas. The context explorer breaks down where tokens go across the system prompt, tool definitions, rules, skills, and more.

A 2005 state-designed worm designed to corrupt physics simulations sat undetected on VirusTotal for nearly a decade. Fast16, intercepted executable files at the kernel level and silently rewrote floating-point calculations to make them produce slightly wrong answers. Targets: high-precision engineering suites used for structural analysis, crash simulations, and physical process modeling, including LS-DYNA, a tool cited in reports on Iran's nuclear weapons research. The sabotage vector relied on deployment of the driver across a network via worm, corrupting calculations on every machine, and eliminating the possibility of cross-checking results against a clean system. Stuxnet got the documentary. Fast16 got twenty years of nothing. sentinelone.com/labs/fast16-…

Europe’s share of newly created unicorns 🦄 is rising to currently 23% 🇪🇺 Asia 🇨🇳 🇯🇵 🇰🇷 🇸🇬 is at 18% and shrinking and U.S. 🇺🇸 still leading at 58% More of this

"Most people don't realise that Europe is the second largest manufacturing hub, that we're world leading in automation, precision manufacturing and in robotics." Andreas Klinger (@andreasklinger) on going behind the scenes of hardware-heavy European tech companies and the legit claim it makes more sense sometimes to start a robotics company here.

Spicy take. I think humanoids are a terrible idea until at least one of these criteria is met. ≥ human level intelligence ≥ human level dexterity If we’re not at parity on at least one, it’s not going to pencil out. Full stop. It’s just not economically viable. How could it be, when you can literally hire a person? Humans are SO absurdly good. For $15–$75/hr I can have a system with the following specs: • 20 W multimodal self-training 100 petaflop supercomputer • 18 years supervised unsupervised pretraining (publicly funded) • ATP-based biochemical energy system (accepts donuts) • 50 MP stereo vision on 3-DOF gimbal with dynamic range adaptation • 244-DOF compliant actuator network with haptic feedback • Self-healing, self-repairing, self-assembling biological substrate • ~5 million analog force-feedback sensors across digits and dermis • Autonomous recharge via oxidative metabolism (~16 h duty cycle) • Recursive genome-encoded firmware with continuous self-updates • Natural language interface with context-adaptive inference • Multi-objective optimizer (survival, novelty, dopamine gradients) • Emotionally fault-tolerant stochastic control architecture • Multi-sensor fusion: proprioception, vestibular, nociceptive arrays • Adaptive gait control with terrain classification and self-righting • Wideband acoustic output for communication and threat signaling • Olfactory chemical sensing array (volatile organic detection) • Load-bearing skeletal truss with self-lubricating joints • Hydrophobic dermal coating with cellular self-regeneration • Distributed thermal management via vasodilation and evaporative sweating • Predictive maintenance through pain and fatigue heuristics This is how I look at it: can I make money on it versus hiring a person? If you’re not viewing automation through a brutalist economic lens, you’re out of your mind. I’ve said this for years. In the year of our Lord 2025, I’m extremely pro-automation for deterministic problems. Completely against it for nonlinear, non-deterministic, non-closed-form ones. I’ve spent more time telling people what not to automate than what to. This isn’t armchair talk. I’ve built some of the largest manufacturing robots around. 64 axes of synchronized high-speed motion. Hundreds of I/O channels. Written vel/acc/jerk trajectory planners from scratch. Closed MIMO loops with vision and thermal sensors. 5-axis toolpath generation, laser point-cloud fusion — all the “hard” stuff. Except it’s not. Those are easy because they’re closed-form. The real world isn’t. What blows my mind is how few people grasp the scope difference. This isn’t a gap of degree. It’s a gap of kind. I can look at a millisecond-timed multi-axis servo platform and feel actual joy. Then I watch a vision-guided arm try to unload a dishwasher and I want to scream. A decade ago I said: forget human-replacement automation until AGI is real. I still stand by it. Humans are so good, so brutally efficient, that parity isn’t enough. You need super-parity.

BREAKING: Danish media reveals that a Russian Navy Ropucha-class landing ship, Aleksandr Shabalin has been spotted by helicopter lurking ~12 km off southern Langeland/Lolland in Denmark with its AIS transponder switched off.

Sharing some acceptable examples of attire for the @eutncom launch party below:

we’re building the TBPN for Europe. introducing @eutncom

My friend @danwwang’s book Breakneck is up for pre-order! So proud of him. This book gave me a new framework for understanding China-US relations. Anyone who cares to understand this critical geopolitical rivalry should read it. amzn.to/4lHPb2P

Watch Helix's neural network do 60 minutes of uninterrupted logistics work Helix now incorporates touch and short-term memory and it's performance continuously improves over time