That's mostly accounting for the resolution and motion sensitivity in different parts of the eye. With enough cameras a car should be able too "see" more than we could at any one time.
I think car automation peaked at adaptive cruise control. It's a simple tractable problem that's generally well confined and improves the drivers ability to concentrate on other road risks.
I can see the argument that visible light should be enough given we humans can drive with just two eyes and a few mirrors. However that argument probably misses the millions of years of evolution of our neural networks have gone through while hunting and tracking threats that happens to make predicting where other cars might be mostly fine.
I have a feeling regulators aren't going to be happy with a claim of driving better than the average human. FSD should be aiming to be at least 10x better than the best human drivers and we're a long way off from that.
For portability Vulkan is the way (it also gets you GPU compute for free without needing vendor libraries). That said the ruttabaga encapsulation is useful for things like Wayland over virtio-gpu which is useful for some use cases.
I should note for even closer to native performance you want virtio-gpu with native context. Patches for that are currently being reviewed on the mailing list: https://patchew.org/QEMU/20241024233355.136867-1-dmitry.osipenko@collabora.com/
I'm not sure how assaulting children is ever going to build an effective relationship between kids and their parents. Parents should represent safety and unconditional love because then the educational message will have an easier time being accepted by the kids.
Very binary, much wow.
Buy games from indie developers on platforms like itch.io. You may have a negative view of the other people involved in funding and marketing a triple AAA game but they all contribute and get a share of the retail price. You don't get to pick and choose who deserves to get their slice.
Don't be too hard on Collin. Looking back on the threads it's fairly clear he's been the victim of a social engineering attack on an overworked maintainer. People were pressuring him to hand over maintainership while expressing disappointment at the slow pace of development. The off-list contact by Jia must have seemed like a helpful enthusiastic solution to a burnt out developer.
It's looking more like a long game to compromise an upstream.
Yes training is the most expensive but it's still an additional trillion or so floating point operations per generated token of output. That's not nothing computationally.
So the entire article basically comes down to democracy is messy and with PR you can't necessarily predict who you are going to get in coalitions.