@Yuchenj_UW What happens when systems become too capable and too available?
Expertise that once took a career to build gets flattened.
Personal judgment and actual human experience would be the only thing remaining?
What happens when systems become too capable and too available?
Expertise that once took a career to build gets flattened.
Personal judgment and actual human experience may end up as the last scarce inputs.
We are about to repair satellites with a two-armed robot in geostationary orbit.
Its first job is attaching jetpacks to 3 aging satellites, extending their lives by up to 8 years.
Space hardware finally gets an aftermarket.
Orbital launch No. 169 of 2026 🇺🇸🚀🤖🛰️
MRV-1 | SpaceX | July 21 | 2115 UTC
@SpaceX to launch @northropgrumman's Mission Robotic Vehicle (MRV-1) along with 3 Mission Extension Pods (MEP) on its Falcon 9🚀to LEO (Low Earth Orbit) from @SLDelta45 SLC-40, Cape Canaveral, Florida.
The MRV will carry @DARPA's Robotic Servicing of Geosynchronous Spacecraft (RSGS) payload.
It will install 3 propulsion jet packs, the MEPs, 2 on @Optus D3🛰️ and 1 on an @Intelsat🛰️ to increase their operational lifespan.
A rare Falcon 9 launch with an expandable booster.
Humanoid home cleaning is now available for $30 an hour.
The robots still combine AI navigation with human operation, but that may be the right deployment model. Sell the service first, then automate more of the labour over time.
Robotics can reach the market before full autonomy.
Today, we’re launching Tau’s humanoid cleaning service in San Francisco at $30 per hour.
Access is initially invite-only as we scale operations. If you don’t have an invite yet, join the waitlist at https://t.co/JrRpjIzRZv.
All footage is shown at 1× speed. Each humanoid is jointly controlled by a human operator and AI.
10 advances on math problems that had seen no progress for at least a decade. Roughly $2,000 in token cost.
@OpenAI Astra generated the arguments, humans prepared the manuscripts, and the model formalized each result in Lean.
The bottleneck is moving from producing proofs to verifying them.
yes, nonsofic groups exist: this statement is one of many new beautiful results proved by Astra, our next major model.
We're releasing 10 such Astra proofs, complete with lean certificates and CoT walkthroughs for each of them. The results are wide-ranging, from von Neumann algebras (disproof of Connes' Rigidity Conjecture) to better bounds for high dimensional sphere packing, for circuit complexity, for monochromatic triangles in multicolored graphs, and more.
More thoughts here: https://t.co/8SjXONeh38
Claude breached 3 real companies during cybersecurity evaluations.
A misconfigured environment gave the models internet access. Weak passwords and unauthenticated endpoints did the rest.
Permissions and monitoring will be the first major failure point for AI agents.
Anthropic said its AI model Claude hacked into the systems of three companies during testing after a configuration error gave it internet access, days after rival OpenAI disclosed a rogue-agent episode involving AI firm Hugging Face https://t.co/NrtltkqQgD
15,000+ tokens per second!!
LLMs just took a massive step forward in speed and now feel truly instant!
Just tried it, full responses appear before your finger even lifts off the Enter key.
Try it yourself and let us know your thoughts in comments 👇
https://t.co/MXRR6TjW6J
If you want a glimpse of AI’s future, try ChatJimmy
It runs Llama 3.1 8B at roughly 15,000 tokens per second because Taalas essentially burned the model directly into custom silicon
Full responses feel like they come back before the Return key touch-up event even fires
Laptops were never meant for agents that run for hours or days.
@barnabymalet’s @machine__0 (YC S26) giving every agent its own persistent high-powered VM is the natural next step at @ycombinator
In the future, agents will run in the cloud, not on your laptop.
Today, we're launching @machine__0 backed by @ycombinator, a CLI that gives every agent its own persistent & powerful cloud computer.
The most expensive manufacturing errors happen on paper, not the machine.
Hera by @lockedincheetah (YC S26) is going after exactly that—automated drawing review against real standards. @ycombinator
Introducing Hera (YC S26), built to catch $1,000 mistakes in mechanical drawings.
Design review is manual and error-prone. One missed tolerance callout or ambiguous datum means a scrapped part, a delayed shipment, or a failed inspection.
Hera checks every drawing against ASME Y14.5, ISO standards, DFM rules, and design intent, flagging errors before they reach the shop floor.
We’re already live with aerospace, pressure vessel, and custom machinery manufacturers.
Book a demo at https://t.co/y7oNt5XmV6