I own a Tesla and am renting a car that only has CarPlay. It is miserable; maps are significantly worse, integration is half-assed and clunky. It is amazing to me that a Tesla going on 7 years old is so much better than a new Ford...

As for the cost premium, hard to see where vehicle RAM requirements will get close to $1k premium.

Is using BFI novel? I hadn't seen that before personally.

I doubt the 2,000 trainees (likely from a pool of 20,000 gamers) will result in more than 500 people that make it through two years.

The US has around 14,000 controllers so that puts them close to a nominal replacement rate. It is not a magic game changer, but if they do manage to retain more or expand the recruiting class size it will help over ~5 years.

This has been a slow moving train wreck for a year. Another year of production is fully sold out (likely the next 5). For someone to have not put plans in place last year to add capacity *somehow* is a market failure. The risk at current RAM prices is minimal, and dumping is impossible if inventory is sold out.

The key is to keep the additional capacity secret for as long as possible, until it is actually online.

No, that isn't what I am saying-- humans cannot accurately judge distance based on our "binocular" vision [at highway distances].

My point is that cameras can do better with strategic placement, without LIDAR. It will still have failure cases, but it can be pretty robust.

Humans don't have binocular vision at highway distances; the eyes are too close together. We use other visual cues to establish distance. A forward facing binocular vision system for a car would need cameras in each A-Pillar, or at least center and driver side with the wipers.

More importantly, cameras can be better because they can have real binocular vision, no blind spots, and plenty of redundancy. No, current systems don't have that, and I am happy my Tesla still has/uses ultrasonic sensors for better distance estimation close-in. Ultimately by the point cameras are completely useless, so are humans. Get off the road and wait for better conditions!

SpaceX has 1.4GW of (nameplate) compute infrastructure, and they are not the biggest player. >1GW load, with plans to double it within 6 months, is a real concern for infrastructure.

California has a peak electrical demand around 35-40GW as a frame of reference. Load growth on the peak of more than ~300MW is a very big deal. When you have 5-10 players collectively looking at adding >20MW over the next two years there are no cheap and easy solutions for additional capacity.

But does a border search of a US Citizen have such standing as a court order or warrant? What was the basis of the request to inspect the device-- was there information that he could be a child pornographer, or was he just on a watch list because of his occupation?

Can a prosecutor prove beyond a reasonable doubt that it was wiped with a duress password with the intent of hiding criminal activity, or maybe it was wiped in advance? Can the prosecution get a court warrant to search the individual's home based solely on the wiped device?

I think there are a few improtant subtleties to it. Namely that Nvidia is providing a [loan] guarantee and not an actual loan. It is also unclear if this guarantee is on the data center, or for the power purchase agreement. So, when OpenAI can't pay for whatever it is, Nvidia is responsible for payments. Depending on the terms of the deal, the risk could actually be quite low for Nvidia. Something similar goes for financing chips with a >300% margin-- payments after the a year or so are pure profit.

It sounds like the guarantee is on a PPA and not data center build based on the summary. That is a pretty important distinction as a PPA is a very different kind of risk than a compute facility (especially given grid shortfalls for projected growth).

So as I understand it, current AI models need about two orders of magnitude efficiency improvement to be sustainable as a business. You burst the bubble if usage growth doesn't exceed roughly infrastructure growth divided by efficiency improvement-- flat usage means a 10x growth in infrastructure requires a 10x efficiency improvement. There is some fudge factor to cover sunk costs, but I'm not sure how meaningful that is as you talk about orders of magnitude.

Most of what AI is useful for today could be done more efficiently with small models, and the idea of AGI is a long long time away despite what true believers hold as gospel. That, from my limited understanding is where most of the token spend is today-- the money could easily be starved in that context.

Personally I would not want to live within 3 miles of the perimeter of a data center, but the summary really mixes up issues. On-site generation, presumably around a >300MW scale, should be regulated as a power plant from an environmental perspective. Diesel generators are hopefully just a backup, and at 80 hours of operation per year I would not stress about them myself. It really sounds like they shopped regulators in this case.

Baggage automation isn't too hard with random sized/shaped objects; you will have a breakage rate higher than humans though. Really though, containerized bag holds are an easier solution that doesn't require (but permits) automation.

The modern carry-on problem is that you can't check things with batteries in them.

A friend was a flight attendant on the recent suicide-by-aircraft-engine-injestion incident. People don't listen; they take videos, they grab their rollerboards and everything else, they do stupid things (and apparently some flight attendants freeze up and forget to open their door). Just to throw it out there, the real reason you should not take a bag with you is it will greatly increase the risk of injury as you evacuate. Someone at the bottom of the slide is supposed to help you stop-- if you take them out with your baggage that is much less likely to happen. If it is important, have it on your body at least for taxi, takeoff, and landing.

Safety briefings really need to change. They need to be compact, concise, and focused on the high risk events. They don't need to be comedy, they can't be advertisements, and they need to have a system to make it permanent memory for people that do fly a lot, kind of like education.