Okay, and in 1969 the Navy installed an LM2500 with zero real world use. What makes you think that companies won't do that today?

I think you're imagining that large projects like this are like consumer products. They're absolutely not. Buying a system that is brand new, or very limited run is completely normal when you're buying things this large. You're not going to find one that's produced in the thousands.

To give you an example - Google are even planning on building these with CFS's fusion power plants. That's not just a completely untested design of an existing well understood mechanism, it's a completely untested design of a completely untested way of generating power, with a near century long history of "don't worry, we'll have it working in 20 years".

Gas turbines are extremely efficient. They are of the order of 60% thermally efficient, which if you're going to burn fossil fuels, is about as good as you can get. Of course, renewables would be better, but if you need power in a specific place, right now, gas turbines are about as good as you can do.

Right - "pivots" seems like the wrong way to frame this. They're not (yet) going "we're not going to build supersonic jets, we're going to build power plants", they're going "we need to develop a highly efficient, powerful jet engine, we can fund that by building power plants, and use that as a stepping stone to a supersonic jet".

I'm guessing you paid significantly more for those computers than if Sam Altman had not been involved. For many people, his buying power is enough to simply disallow buying a computer at all.

https://www.pcmag.com/news/sk-...

Yes, but there was an implied "can stick parts of cars together in an order that mostly makes sense" as the third condition here. Once you have that, you have the obvious

1. Stick car parts together
2. Check if it's a car
3. If it's not a car, go to 1.

The better it gets at step 1, the faster the process works, but it works no matter what. In practice, AI is pretty good at step 1 already.

Because the algorithm for producing flawless code is simple:

1. Write code
2. Check the code for flaws
3. If the code has flaws, go to 1.

It might take a *really* long time if the AI is crap, but in practice, AI these days can write secure code much better than humans in most cases. It doesn't sit there accidentally slipping in memory corruption bugs, or buffer overruns, ... It gets those things right, basically 100% of the time.

AI psychosis is where you talk to an AI about your weird beliefs, and it reinforces them generating paranoid delusions.

This is just him being more bullish on AI than his employees. Frankly, his employees need to figure their shit out, because their concern about security is bullshit. We can't have "AI is able to detect incredibly obscure security vulnerabilities and exploit them as zero days", and "AI is incapable of producing secure code" at the same time. All you need to do is make sure you actually prompt it to find the vulnerabilities and fix them (and actually review the code, which, contrary to most people's weird beliefs, is perfectly readable in general).

While that's true,

1) We're not talking about an asteroid-sized object
2) We're not talking about one object occulting things
3) We're not talking about one object being occulted.

While you're right, that through pure chance, the likelihood of taking a shot of one particular asteroid occulting one particular star if you take a shot once every 3 nights is near zero. However... one shot from vera rubin captures on average 10,000,000 stars, and around 1000 potential occulters. It's estimated that an occultation will occur in roughly 1 in 10 to 1 in 100 exposures from vera rubin.

Up until very recently weâ(TM)ve not really been looking at large areas of the sky repeatedly every night. There were likely thousands of occultations and we just didnâ(TM)t see them because we werenâ(TM)t looking at that particular bit of the sky when they happened.

Vera Rubin came online a few months ago, and has precisely that job - to take images of a large portion of the sky with a very wide angle very high resolution camera, and compare them every few nights. The result⦠oh look - thereâ(TM)s a dwarf planet, not just occulting things, but reflecting light.

What, you think that classified military industrial complex can achieve the kinds of economies of scale that Chinese toy makers can? Dude, give me a break. China will be making these things for $100 while we make them for $1,000,000.

Ukraine has even been involved in training exercises with NATO troops. Their drone battalions have been regularly wiping out American and British infantry units in exercises, because the Americans and British just don't understand how to fight a war with drones in.

The good news on the running out of people front is that Russia is running out of people 10 times faster. The bad news is, Russia has 4 times as many people. That said, those numbers still lie in Ukraine's favour.

The equipment can be carried by other drones. Though certainly there's all kinds of risks here. A wounded soldier who will die if not extracted has a very different risk profile they'll accept than a healthy soldier going into an attack. You don't want to deliver ammo and supplies to your enemies if the supplies make it, but the soldiers don't. etc etc.

That said, Ukraine has already used a variant on this - they've used these heavy lift drones to deliver land drones that then carry out an attack. That was one of the things that made their recent advances near Lyman possible.