One of many ways this could all play out from an incentive based perspective is that it starts with knowledge workers and coding. A next token prediction machine seems to excel at code generation for this reason. This has already begun causing problems in India with recent youth protests. It's not the entire reason for the protests but it is part of it. The indian grad to software developer pipeline is well established and under threat. The owner class wants to create the "value" from software without having to pay the high wages associated with development costs.
As AI advances, this spreads to other knowledge work that AI is capable of doing, which we might be seeing in mathematics if the OpenAI paper release stands up to scrutiny. Other kinds of data and analytics work will follow.
Then once the knowledge worker jobs are largely replaced, the middle management that is left will either be not needed or replaced.
By this time the only remaining jobs will be the manual labor jobs that can't be automated. Maybe as important as recursive self improving AI is going to be AI advances in robotics. Once this happens, the ownership class can finally start replacing the manual labor jobs that remain.
If AI can achieve RSI and be combined with robotics, maybe the prediction of just eventually having one company that does everything will become true.
One large AI company that does everything without any labor necessary, and everyone else I guess will just compete on OnlyFans. The only kind of work that neither AI or robotics can replace.
It’s also not guaranteed that owners are reliable narrators, when speaking about how great their business is doing. In other breaking news, worlds best cup of coffee found in truckstop diner!
Many of the bases were created to fight the "war on terror" and conduct a decades long massive regional regime change that cost the U.S. trillions of dollars. Anything that changes the course on this and prevents more global empire building is a good thing... Even at the cost of losing bases that never should have been built to begin with.
100% this. I've been vehemently opposed to the war in Iran, but only because I didn't expect it to be anywhere near as chronically catastrophic as it has been. I expected more of a long slow bleed like Iraq/Vietnam/Afghanistan 2.0. If I knew the outcome ahead of time, I would have been fully supportive of it. At the risk of sounding like 1984, losing this war will be the biggest victory in modern times for the US. Our hegemonic geopolitical strategy has been very damaging to both the US and to the entire world. We're not going to be king of the world, and we shouldn't endeavor to be such. Getting hided by a minor power is going to force us to rethink everything, and I think only good can come from this.
Even if the clowns in charge think the take-away is 'we need even more bombs, and bigger ones', they're going to learn that when we're expending this much on a relatively minor power like Iran, the economics/resources for this scale of war against a major power simply don't exist. We're looking at Proud Prophet 2.0 [1] which was also a huge victory in defeat for the US. It's unfortunate that in the past we were able to learn from wargames, whereas now we had to see things play out in real life. Perhaps next time when we do a wargame against Iran and lose [2], we shouldn't just change the rules so we can win.
It's also much more likely that Meta products will have much greater access to the Brazilian market if a U.S. friendly leader is elected. The alternative would push the alignment closer to using Chinese services. Brazil is kind of a battleground between Chinese and U.S. influence.
Not sure what greater access there is to have. Brazil is pretty much addicted to Facebook, Instagram and Whatsapp. But you can be sure that any plans of distancing themselves from these networks will be discarded if they end up going with Flávio.
The next token prediction machine is writing 90%+ code at multi-trillion $$$ companies, its already solved millennial problems -- but yes its definitely not living up to hype.
But it's creating new problems, and when the damage those new problems outweighs the value created, then that's when the bubble starts (or started) to form and we live in a delusion until the damage massively outweigh the benefits, is when the bubble popping starts being considered, and then when there's irreversible damage is when the bubble popping is marked for formal revision and then when ...
Photonic computing in general is not snake oil though. It's more real and practical than quantum computing especially for matrix multiplication which is the majority of model training compute usage.
Devices that may not be easily patched, especially industrial systems will not do as well in a world where LLMs are finding exploits. Atleast with other software the defending side can use LLMs and stay up to date in the arms race.
Anything that can be run offline should be airgapped and you still have to hope you dont get Stuxnetted.
Early copyright based legal arguments about the training of pirated information for AI didnt have much success. Im sure it will continue not being an issue for the frontier labs while copyright laws are enforced for everyone else.
It’s not about what the model was trained on (though it might be), it’s that fair use in reverse engineering relies on transformative, creative work. There are multiple court cases where RE is protected.
What’s not protected fair use is mechanical transformation (automated format conversion, etc.). I’d argue using AI for RE and publishing the result is not creative, and is mechanical transformation. That makes it not protected fair use and puts projects at risk. This also has court precedent (not necessarily the AI case, but mechanical vs creative transformation).
Misunderstanding this is going to ultimately bring harm to the RE communities where legitimate projects get wrapped up with AI projects in litigation and will chill the whole practice even though it’s legitimate fair use.
(US jurisdiction, I’m not a lawyer, I’m not your lawyer, but I’ve spent a reasonable time reading the landmark cases, EFF’s opinions, USPTO opinions, etc.)
As AI advances, this spreads to other knowledge work that AI is capable of doing, which we might be seeing in mathematics if the OpenAI paper release stands up to scrutiny. Other kinds of data and analytics work will follow.
Then once the knowledge worker jobs are largely replaced, the middle management that is left will either be not needed or replaced.
By this time the only remaining jobs will be the manual labor jobs that can't be automated. Maybe as important as recursive self improving AI is going to be AI advances in robotics. Once this happens, the ownership class can finally start replacing the manual labor jobs that remain.
If AI can achieve RSI and be combined with robotics, maybe the prediction of just eventually having one company that does everything will become true.
One large AI company that does everything without any labor necessary, and everyone else I guess will just compete on OnlyFans. The only kind of work that neither AI or robotics can replace.
reply