Nvidia Halos
nvidia.com
nvidia.com
The real technical challenge is rappresented by edge cases: a software that is excellent 99.9% of the time can still be unacceptable if the remaining 0.1% contains rare but catastrophic scenarios. And that's why we still don't see many self-driving vehicles on the roads today.
However, NVIDIA has a credible shot because it controls much of the loop - hardware, training infrastructure and simulation environment. If it works they will impose a huge vendor lock-in, difficult to replicate for other competitors.
The new game is finding a single sentence with the most instances of "safe" or "safety". My current high score is 4..
Meta Horizons World probably puts up similar numbers if you sum up the hardware/software tech stack to get this number.
Here's Archive.org's copy of the page from 2025 in September, their earliest copy:
https://web.archive.org/web/20250920031549/https://www.nvidi...
"Safety transistors safety assessed" exists in this version too
It's also hard to blame Nvidia for the pivot, from where I'm standing. Their proprietary middleware like PhysX, DLSS and RTX has been memed to death by PC gamers, while high-margin edge and datacenter customers are chomping at the bit for CUDA compute. Nvidia's raster stack is more-or-less complete, the things that PC gamers are asking from them are not realistic or fairly priced at this moment in time.
That was my point. It's not even a pivot! They're still making consumer cards! They've even product-differentiated enough that the consumer cards are still on the shelves at close-to-MSRP, despite world-historic demand for adjacent parts of the lineup.
Being _mad_ at Nvidia in this setting is weirdly possessive - a business that was 90% gaming is now 10x larger and 9% gaming[1]. You haven't lost ground!
[1]: numbers made up but you get the point
gamers always want cheaper and faster gpus. its the same thing data centers want but they care about ray tracing and fragment shaders instead of raw number crunching power. its only "not realistic" in the sense that nvidia alone cant do much to lower prices when theres a memory shortage and they depend on tsmc.
> 18,600+ Engineering years invested in vehicle safety to date
What does this even mean?
> 7,000,000 Lines of safety-assessed code
Are we seriously using LoC as a measure of productivity again?
Not to mention the em-dashes
If it means what I think it means, you take every engineer working on it (and maybe the years of research involved) and add it all up. Say you have a room with about 10 engineers with 10 years of experience per developer, you can claim there's 100 years of developer experience between all of them (maybe the overlaps not unique enough and its more like 30 to 50 years? but in this case I think they're rounding up, and I assume it means thousands of engineers involved in the project) that's how I took it.
My first interview in tech I was asked what the heck I was even doing with the D programming language, followed by the remark that in the next room (where all the devs were) there was at least 100 years of experience between everybody there, and not a single one knew what D was, my manager clearly did, which cracked me up.
It means over 18,600 engineering hours have been spent working on vehicle safety. This is a pretty common metric.
Yes, sadly. Because its how everyone justifies LLMs. "Look at how much code it writes!" is the only measure they can come up with to sell its usefulness, completely forgetting that it'll be more useful if we started talking about how much code they remove.
I don't even know what that was supposed to mean. Hopefully all the safety transistors in the safety graphics card of my safety-PC were safety-assessed, too /s
Hot take here, but personally I feel they should safety assess the danger transistors, reducing the need for so many safety transistors.