Pinned Tweet
the bots are grinding on this one.
immersed boundary layer solved on gpu with @NavierAI agents
i spent 2 years working on hydofoiling ships out of college. something like this would've taken me multiple months to do and i was just able to do this in 2 days of doing this on the side.
AI is already completely changing how engineering is done
done with @NavierAI agent and solvers
same hull as yesterday, now in irregular seas. jonswap spectrum at 22 kn, 6.5 s peak period, 1.83 m target wave height (1.61 m measured upstream)
it pitches way up off a crest around t=25 s and comes fully out of the water at the end of the run. the panel at the bottom left tracks heave, pitch and bow acceleration as it goes
using @navierai's gpu solver
generic 8 m patrol hull, 22 kn into 1.83 m head seas, free to heave and pitch
in calm water it lands within 6% of savitsky on trim and 3% on resistance. in the waves it heaves 2.09 m peak to peak and pitches from −3 to +13 deg
single mesh, and slamming isn't resolved yet. towing tank comparison is next
done with @NavierAI with our GPU solver
this is the kind of sus behavior that you should be concerned about with open-source models..
i expect we'll see a lot more of this sleeper-agent like behavior come to light
excellent work from the @cohere team to uncover this
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
stoke's 2nd stage is one of the most interesting designs for reusability
ran some reentry and landing sims with @NavierAI. this is a powered landing flown by a closed-loop gnc controller coupled to the cfd, simulated with ventus, our gpu cfd solver
*representative geometry from public info, not affiliated with stoke*
at mach 2.5 the retro plumes push the bow shock way out ahead of the vehicle. left is numerical schlieren, right is mach
each flight phase ran on a single h100 in under two hours. all three phases together were about 18 gpu hours
and the landing burn from below: 24 retro thrust chambers firing into mach 0.6 flow
this is from the engine and solvers behind prism, our ai engineering platform: navier.ai/prism
Nvidia is reaching it's raptor 3 era
this is the path for companies that continue to iterate and improve their products
Jensen Huang holds up a Vera Rubin node on stage: what took two hours to assemble on Grace Blackwell now takes under five minutes, with no cables and 100% liquid cooling
“So Vera Rubin is in full production. It’s nice to kind of see it. This is the first computer in the world that we’ve created. No cables, no wires, completely 100% liquid cooled. And it’s just really delightful...Really delightful to be able to pull it out. And this is one Vera Rubin computer. Okay, so this has two GPUs."
"What used to take, in the case of Grace Blackwell, two hours to assemble one node, and when you see it, you’ll know what I mean. Cables and wires everywhere, 80% liquid cooled."
"Now Vera Rubin is 100% liquid cooled, and it takes less than five minutes to assemble it. And so, incredible, incredible achievement."
"Of course, this translates to higher cycle times in manufacturing, much, much better reliability, and of course, a lot better performance.”
agree with most of this. models will get good at design fast. but most of the time on a real program goes to building the thing, measuring it, and figuring out why it didn't work as expected
more sims won't remove that, but they can cut how many times you have to loop. @NavierAI
AI is on track to become a superhuman hardware designer, but design is only the beginning. The interface between bits and atoms is lossy, and turning designs into real hardware gets harder at every level:
1. Sourcing off-the-shelf parts that work together (a 2006 Falcon 1 engine test was delayed by one capacitor, in stock 6,000 miles away)
2. Custom parts, where a master machinist translates CAD into machine instructions
3. Assemblies, where every part can meet its drawing and still not fit
4. Outside vendors (even SpaceX has 591 employees with supply chain titles)
5. Doing all of it at a scale of 1,000,000
"Superhuman AI might be great at pure design, meaning at defining the platonically ideal part for a particular job, but struggle with 'design for manufacturing,' or finding a point design that meets the requirements while remaining easy to build/procure and inexpensive."
Full piece from a16z's Christian Keil on what it takes to turn AI-designed hardware into the real thing: a16z.news/p/superhuman-hardw…
has @Tesla tested a backflip yet?
the inertia about the pitch axis is ~3-4x that of the roll axis so rotational energy increases by ~3-4x
however, the majority of the energy required still comes from the initial jump, so this should theoretically be possible
we have the technology. we just need to be brave enough to use it.
coupled 6-DOF + CFD. tail booster, four hoof jets, closed-loop landing.
engineering has run this loop for a century: model, test, find where the model lied, update.
the bottleneck is how fast you can get honest feedback from physics.
love seeing more of the build
From tank integration to engine production, September builds kept the team busy.
✓ Installed the downcomer into the stage 1 flight tank
✓ All primary hardware on the stage 1 aft dome is complete, closing in on final integration
✓ Interstage drilling finished, now heading to the Tender Loving Crusher for structural testing
✓ First 2 ground station controllers stood up
✓ First welds complete for the Aeon V vacuum engine nozzle extension
✓ Flights 2 and 3 engines well into manufacturing with production ramping
