commandar

joined 2 years ago
[–] commandar 4 points 1 month ago (3 children)

Sure, but those are largely the big tech companies you’re talking about, and research tends to come from universities and private orgs.

Well, that's because the hyperscalers are the only ones who can afford it at this point. Altman has said ChatGPT 4 training cost in the neighborhood of $100M (largely subsidized by Microsoft). The scale of capital being set on fire in the pursuit of LLMs is just staggering. That's why I think the failure of LLMs will have serious knock-on effects with AI research generally.

To be clear: I don't disagree with you re: the fact that AI research will continue and will eventually recover. I just think that if the LLM bubble pops, it's going to set things back for years because it will be much more difficult for researchers to get funded for a long time going forward. It won't be "LLMs fail and everyone else continues on as normal," it's going to be "LLMs fail and have significant collateral damage on the research community."

[–] commandar 5 points 1 month ago* (last edited 1 month ago) (6 children)

There is real risk that the hype cycle around LLMs will smother other research in the cradle when the bubble pops.

The hyperscalers are dumping tens of billions of dollars into infrastructure investment every single quarter right now on the promise of LLMs. If LLMs don't turn into something with a tangible ROI, the term AI will become every bit as radioactive to investors in the future as it is lucrative right now.

Viable paths of research will become much harder to fund if investors get burned because the business model they're funding right now doesn't solidify beyond "trust us bro."

[–] commandar 17 points 1 month ago* (last edited 1 month ago)

This is yet another example of our system fundamentally being incapable of dealing with someone like Trump willing to deviate from all established norms.

Legally, POTUS is the classifying authority. They can give clearance to whomever they want.

That's worked mostly fine since the classification system was established in the early 1950s because the assumption has always been that POTUS isn't wildly compromised and completely surrounded by compromised individuals.

Oops.

[–] commandar 11 points 1 month ago* (last edited 1 month ago) (2 children)

Some decent comics pop up on Kill Tony on occasion, but, every time I've watched it, Tony himself has been deeply unfunny.

I didn't know much about him beyond that other than the association with Rogan, but this.... does not surprise me in the least.

[–] commandar 19 points 1 month ago

On a long enough timeline.

In reality, they lost to TSMC as much as anything. That's the real meat of why this case was so important: AMD not being able to gain marketshare meant they couldn't afford to reinvest into R&D. AMD falling behind in fab tech and having to spin off Global Foundries to stay afloat was a near-direct result of Intel's anti-competitive moves back then.

That lag in process tech had ripple effects for years. AMD didn't really start to gain serious marketshare until the one-two punch of Zen and Intel hitting a process wall (while TSMC kept moving).

[–] commandar 8 points 2 months ago

Hardware like that has been and is still being donated through third parties daily.

It's more in Ukraine's interest to limit the use of Starlink to only those terminals that have been vetted through official channels than to allow blanket use and try to filter out things through other means due to... the exact kinds of situations this article is talking about.

but that would require the CEO of the company to actually want to help honestly.

Sure. And part of the reason we know Starlink is entirely capable of geofencing is because Elon's done it explicitly to stop Ukraine from being able to operate near Crimea. That whole kerfuffle lead to military usage being pushed over to Starshield and a contract with the US government that gives them explicit say on when and where Starlink works in Ukraine.

Elon is dumber than a bag of hammers but it'd be next level stupid even for him to willingly break a DOD contract, especially when people were already floating the idea of invoking the Defense Production Act last time around.

[–] commandar 1 points 2 months ago* (last edited 2 months ago)

The old FourSevens Quarks used to rotate the tailcap the switch between modes. I've got one of the older QT2L-Xs that's probably my favorite light ever for that reason. That and it's the perfect size for pocket carry while still being decently bright. The newer models since they got bought out ditched it which sucks because it was such a simple interface.

I got a couple of Fenix lights recently that I don't hate. They still do the "cycle through modes with a button" thing, but it's at least a dedicated button separate from the tailcap switch.

[–] commandar 41 points 2 months ago* (last edited 2 months ago) (8 children)

There are definitely solutions but they all involve giving a Russian agent direct knowledge of troop movements.

Starlink terminals are activated using a unique identifier. It's how billing works.

SpaceX knows which terminals have been provided to Ukraine. We know they can geofence service. Geofencing the Ukrainian theater to terminals that were provided to Ukraine shouldn't be a massive technical leap and doesn't provide any information they don't already have.

[–] commandar 2 points 2 months ago* (last edited 2 months ago)

As long as you don't need particularly tight tolerances or fine details, it works perfectly fine. The setup really isn't anymore complicated than I described. I have done it just because I wanted to see how difficult the process is. It's around $100 in startup costs assuming you have access to a printer. After that it's mostly just waiting and occasionally measuring cut progress.

Check out the Rack Robotics Powercore as well. It's a low cost wire EDM system that uses cheap 3D printers as a motion platform. It uses a very similar principle to cut metal using wire as the cutting tool. May or may not be more suitable depending on your exact use case. Still pretty rough around the edges though; SendCutSend makes more sense for most people that need things cut from flat stock for the most part.

[–] commandar 3 points 2 months ago* (last edited 2 months ago)

The majority of components in these things typically are printed. It's just the pressure bearing components like the barrel where ECM is being used.

The FGC-9 discussed in the article is notable because it was explicitly designed to be manufactured using components that are not regulated as firearms parts under European law. So there are non-printed parts in it, but everything is dual-purpose and can be acquired through unregulated channels, e.g., the hydraulic pipe that gets used as a barrel.

[–] commandar 100 points 2 months ago

Because Google holds a monopoly position and Epic doesn't.

That said, the irony didn't escape me either.

[–] commandar 2 points 2 months ago* (last edited 2 months ago)

The reality is this is one of a handful of emerging technologies that are going to reshape a lot of things about the world in the future in ways I don't think society, as a whole, is cogniscent of, let alone prepared for.

This is one of them. The battlefield use of small drones is another.

I tend to say that the world we're living in now is one where gun control is increasingly obsolete. That's not a moral judgment. It's not a statement on whether that's a good or bad thing. It's just what I think we're going to increasingly find to be the new reality: the rise of small scale, low cost, divertible manufacturing technologies is going to make traditional supply-side approaches to regulation untenable. That genie is out.

(Drones are in a similar, if distinct space: low cost, commodity, and divertible from low/no regulation supply chains in a way that makes it nearly impossible to cut off supply without shutting down other legitimate economic activity).

I don't know what the right answer is. I do think it's going to take a pretty fundamental rethink of how we approach these problems. I don't think the full ramifications of these types of technology have really reached the wider zeitgeist, and, frankly, I kind of worry about how people will react. There are a lot of pretty scary paths this could take, both in terms of how the technology gets used and in terms of what attempts to curb them could look like if they're not carefully thought through.

view more: ‹ prev next ›