There's a thread of thought that pops up in pro-AI posters from time to time: technology can't go backwards. The implication being that the current state of AI can only improve, and is here to stay.
This is wrong. Companies are spending multitudes of piles of cash to make AI work, and they could easily take their ball and go home. Extending copyright over the training data would likely trigger that, by the industry's own admission.
No, self-hosted models are not going to change this. A bunch of people running around with their own little agents aren't going to sustain a mass market phenomenon. You're not going to have integration in Windows or VisualStudio or the top of Google search results. You're not going to have people posting many pics on Facebook of Godzilla doing silly things.
The tech can go backwards, and we're likely to see it.
What's amazing about this strategy is that it wasn't secret. They published it in the Wall Street Journal in 1974. It still worked. Similar to the "flood the zone" strategy; Steve Bannon straight up talked about it to the press in 2018. That also still works.
There's just one problem: you can't keep burning down the house around you and then blaming the people trying to stop it.
Right from one of those articles ("It's Time to Cut Taxes" by Jude Wanniski):
"The level of U.S. taxes has become a drag on economic growth in the United States," [Professor Mundell] says. "The national economy is being choked by taxes — asphyxiated. Taxes have increased even while output has fallen, because of the inflation. The unemployment has created vast segments of excess capacity greater than the size of the entire Belgian economy. If you could put that sub-economy to work, you would not only eliminate the social and economic costs of unemployment, you would increase aggregate supply sufficiently to reduce inflation. It is simply absurd to argue that increasing unemployment will stop inflation. To stop inflation you need more goods, not less."
Which is interesting, because the ultimate solution to stagflation--which was a problem that reared its ugly head in the few years before the article was published--was to do the "simply absurd". Paul Volcker as Fed Chair would eventually say fuck it, we're sending interest rates to the moon. That caused a spike in unemployment, but it brought inflation under control. Then you bring interest rates back down and unemployment sorts itself out.
It's harsh medicine, but it works. Kept capitalism going for several decades more. Of all the possible solutions to stagflation, this remains the only one that's been tested to work.
These people have been wrong for decades and fought against strategies that save their own economic system.
There's a historical reason for it, but much of the financial press isn't thinking about how it's different this time.
A bunch of debt rating outlets downgraded the US in the wake of the 2008 financial crisis. Everyone said "eh" for a simple reason: where else were you going to go? Everyone's economy was slagging. Many countries tried austerity measures, particularly in the EU, but that didn't work. The US was the best of limited options for a low-risk/low-returns portfolio piece (which is what bonds are), and the downgrade just didn't matter.
Trouble this time is that it's obvious to everyone that the US has created the current situation for itself. Things weren't great for average people before, but you could gussy up the numbers and say things are fine if you didn't look too hard at the details. It now threw away the ability to do that. Nobody else is doing that, because it's dumb as fuck.
So now you get a bond rating downgrade when nobody else has that problem. Suddenly, it matters. The financial press isn't used to this situation and hasn't had to deal with the obvious yet.
Interesting to see that take. When it was new, Redditters criticized him for not giving a better response. You know, the response they came up with when they're sitting comfy in their house, and not actually there in the moment.
Which is to say that your take is correct when you have reasonable expectations of ability to improv a response.
There's nothing about being larger that makes access speed inherently slower. We just have to use cheaper technologies to improve density. CPU cache is usually SRAM, which is less dense than DRAM, but faster. 1GB of SRAM would be god tier. Even the Ryzen X3D chips only have 96MB of L3 cache, all SRAM, and those are sick.
Last part that I need is for SSDs to come down in price to where ~80TB isn't too ridiculous (that's 40TB usable space with RAID1). Cut the price per TB in half two more times to make it there. Otherwise, spinning platters are the bottleneck with my 10Gb network.
Which probably would have happened in the next few years if not for tariffs.
A lot of those modules would work fine if the companies didn't fuck with their drivers.
The Linux ixgbe driver (for Intel 82598 and 82599 chipsets) was submitted with a whitelist for Intel SFP+ adapters. Linux devs added a module option to shut off the whitelist, and tons of stuff is perfectly compatible.
I'm an enthusiast with no kids (and a vasectomy to make sure it stays that way) and I've driven both around a track. Glad to have done it once, but I'm not in any hurry to do it again. Wouldn't buy either one over my Miata.
I also hate the way "algorithm" has taken over the public consciousness. You can find people unironically saying "I don't want any algorithm in my social media feed", which is a nonsensical statement.
From old electrical connections that weren't designed for the different rates of expansion of aluminum and copper. Today, most of them are.