This is an interesting insight into the challenges of porting a TSMC fab from Taiwan to Arizona.
There are hundreds of variables that differ, even just based on geography, any one of which can kill nano-scale wafer yields and mean the difference between 90% ("printing money") and 30% ("bleeding cash") yields.
For example, the harder stone bedrock in Arizona has a higher natural resonant frequency than in Taiwan, requiring the fab design to be adjusted to that variable.
Other variables are local air, water, power, suppliers, and culture, all of which differ from Taiwan and require modifications to the original fab design.
> and they didn't have to spend barely anything on marketing.
To be fair, their marketing was all that gaming related BS you listed, plus developing and maintaining CUDA. Winning gaming got them the winning AI architecture and mindshare around it, and CUDA got them the winning developer interface to it and locked in that mindshare.
I think part of it was that they were testing a new aerodynamic design that eliminates or minimizes sonic boom, so they can go supersonic over land almost immediately after takeoff, and operate over populated land routes. It makes sense to test that kind of thing with the smallest possible model first, then see if you can scale it up to passenger size without losing that quiet acceleration. Their timeline for doing that may be optimistic, but what they're doing makes sense.
The XB-1 doesn't have any boom reduction shaping. That's the NASA X-59, though that aircraft is pretty much a dead-end in that it's not scalable to a passenger configuration.
The XB-1 made use of an atmospheric trick to minimise boom propagation to ground level on one test flight, so well-known in fact that Concorde sometimes used it to accelerate as it coasted-out without an audible ground-level boom. Unfortunately that trick runs out at about M1.17.
That may be true for domestic coast-to-coast flights, but not for transoceanic ones across the Atlantic, or especially the Pacific, or north-south across hemispheres, that can take 8+ hours. Flight time is a higher portion of the total travel time in those cases, and seems like the main market for Boom, especially if they initially target Business Class flyers who do those kinds of trips regularly.
Boom XB-1 did 750 mph air speed. If I've got an 8 hour flight at 561 mph in an A380 that's a reduction to 5.984 hours when I move to the Boom XB-1. Who cares about saving 1.1 hours on a transatlantic flight. There is a reason why Concorde's cruise speed was 1,341 mph.
So when Boom makes a commercial airliner that hits 1000+ mph with the same availability and turnaround time as a typical passenger plane then I'll pay attention. Until then, it's for rich people who can buy their own plane.
Just put a QR code on the front that transmits a vCard. Or a way to make the LEDs on the back display a QR code. Then you can still show people your digital business card, even let them hold it and play with it, but it's still obvious the idea is for them to scan the QR code and hand it back.
> But I think we're not even on the path to creating AGI.
It seems like the LLM model will be component of an eventual AGI, it's voice per se, but not its mind. The mind still requires another innovation or breakthrough we haven't seen yet.
The inflection point is recursive self-improvement. Once an AI achieves that, and I mean really achieves it - where it can start developing and deploying novel solutions to deep problems that currently bottleneck its own capabilities - that's where one would suddenly leap out in front of the pack and then begin extending its lead. Nobody's there yet though, so their performance is clustering around an asymptotic limit of what LLMs are capable of.
reply