Hans C Nelson 🗽's banner
Hans C Nelson 🗽's profile picture

Hans C Nelson 🗽

@HansCNelson • 19,934 subscribers

Wrestling with: Jesus my flaws the beauty & tragedy of humanity engineering & technology

Videos

HansCNelson's profile picture

Google's recent Gemini 3 release has shocked the world with undeniable proof that the TPU is a powerhouse AI chip, leaving many to wonder what lays in store for companies like NVIDIA & xAI. Listen to Gavin Baker lay out the exact strategy that Jensen Huang & Elon Musk are going to roll out over the next 12 months to beat Google at AI. The world is in for a HUGE surprise. ---------- We'll see the first models trained on Blackwell in early 2026. I think the first Blackwell model will come from XAI. And the reason for that is just, according to Jensen, no one builds data centers faster than Elon. Jensen has said this on the record. ... So if you're Jensen or Nvidia, you need to get as many GPUs deployed in one data center as fast as possible in a coherent cluster so you can work out the bugs. And so this is what X AI effectively does for Nvidia because they build the data centers the fastest. They can deploy Blackwells that scale the fastest, and they can help work with Nvidia to work out the bugs for everyone else. So because they're the fastest, they will, they'll have the first Blackwell model. ... We know that these Blackwell models are gonna be really good. ... Then something even more important happens. So the GB 200 was really, really, it was really hard to get it going. The GB 300 is a great chip. It is drop in compatible in every way with those GB 200 racks. Now you're not gonna replace the GB 200s there. Just any data center that can handle those, you can slot in the GB 300s, and now everybody's good at making those racks and you know how to get the heat out. You know how to cool them. You're gonna put those GB 300s in and then the companies that use the GB 300s, they're going to be the low cost producer of tokens. Particularly if you're vertically integrated. If you're paying a margin to someone else to make those tokens, you're probably not gonna be. I think this has pretty profound implications because I think it has to change Google's strategic calculus. ---------- Listen to the full conversation between Gavin & Patrick OShaughnessy below 👇

Hans C Nelson 🗽

5,010,816 views • 9 months ago

HansCNelson's profile picture

In a world where FSD v14 exists, why is the take rate SO RIDICULOUSLY LOW? And why is the utilization rate low too, even for Tesla drivers who've paid for it? Here's one reason that no one is talking about: "You never get a second chance at a first impression," and Adaptive Cruise Control & Autosteer SUCK. They scare the 💩 out of people w/ phantom braking & poor perception all the time. Listen to this clip of Jason Cammisa & Derek Tam-Scott discussing the dramatic differences in how good FSD is compared to how horrendously awful Adaptive Cruise Control is (and Autosteer isn't much better, tbh). Now remember, these are the first (and many times only) impression that MOST Tesla drivers get of how well a Tesla can drive itself. If Adaptive Cruise scares them, why should they ever even try FSD? Won't that just give the dumb ass car even more opportunity to scare / kill them? If you're in the Tesla bubble like me, you know that FSD actually fixes all those nasty little quirks instead of amplifying them. But how are all the normie drivers who've never seen a video of a Whole Mars Catalog drive supposed to know that? I get that Adaptive Cruise & Autosteer don't drive revenue, that they're based on old versions of the tech, and that it's hard to prioritize updating them while FSD & Robotaxi are right on the cusp of getting to unsupervised, but if Tesla_AI wants to improve the take rate of FSD, they need to fix the user experience for these lower levels of ADAS ASAP. Elon Musk Ashok Elluswamy Yun-Ta Tsai Dhaval Shroff etc.

Hans C Nelson 🗽

42,411 views • 9 months ago