I’d venture that those 500k jobs are mostly enshittified data annotations using contract work to avoid paying benefits, that people with comp sci diplomas have to take because India has already too many programmers for its local industry.
Game development has also been very badly hit since the pandemic but it also had a bubbly spurt of growth during and right after the pandemic. It’s been hell for the last 3 years, not just the job loss, but knowledge and know how, invidividual and team wise.
The AI bubble is making it worse while taking advantage of worsening financial results with the lure of more profits by replacing people with so called AI.
There is a trend in computing to increase performance by under-powering the GPU because it then stays cooler, and when its cooler the internal resistance is lower, so less current is required to power the device. The flip shows how performance is increased with under-powering: if the PC is over-powered then temperature rises and resistance rises too causing less current and slower data transfer. Thermal throttling may also occur as an internal system to prevent thermal damage, and reduces clock speeds and general activity.
The point is that you can ptobably produce the same system in 6nm and 12nm, and the smaller is a circuit that draws less power. The 6nm version could be more spread out, allowing heat to dissipate more between internal components. There is no reason why more transistors should be crammed onto the same chip just because its smaller and uses less power. The smaller chips are slower but generally more efficient.
The thermal runaway is another point to realize - resistance increases due to temperature, but to run the chip at higher resistance requires higher power, some gets dissipated as heat, and hence the effect continues.
More of each effect just cycles ….
Power→Temperature → Resistance → Power → Temperature ….
This is avoided with supercooling and thermal throttling, one costs power and resources, the other costs performance.
Theres another type of efficiency here …data power efficiency. You could measure it with this:
That doesn’t matter. My point is, whatever the efficiency, the energy will always be the bottleneck and “ai” overlords are insatiable and deluded. With more efficiency, they’d just scale up more. They wouldn’t altruistically back off in the name of saving the energy. Why would you assume otherwise, seeing how they’ve behaved so far and knowing what they were publicly saying. So more efficiency in this case doesn’t mean less energy spent in the grand scheme of things.
I might agree there. But it isnt only energy, there are numerous factors including space covered by the data center and alternative energy sources. If there was a 100 Watt solar panel per chip, 30 devoted to compute, 30 to cooling, and 30 watt sent back to the grid, that works but the spacial coverage is 1 panel every 0.5 square meter. For more compute capability the spacial coverage just has to increase. With the earths temperature and solar flux there is a limit to the amount of compute power that can be self sustaining. They could also build inside rivers for natural cooling, and hydro turbines could power them.
Its a shame that these relatively new energy revolutions are probably going to increase the proliferation and success of robots, who can also be viewed as an alien life form not requiring nature or oxygen or any of the things that humanity is symbiotic with.
Still, I really wasnt making any kind of point about the energy problem in the first place, but it was pointed out to me as a socio-economic factor that I chose to place in far lower priority because the jobs cost was greater.
Now theres an argument that jobs are actually being created, but I dont know whether to believe that because the argument may be biased and Jenson may be almost ironically making the point about re-industrialization being the net positive effect that boosts the economy. What people really want is a more comfortable and profitable job, but I think those types of position continue to shrink in numbers.
The jobs problem shows another shift in the class struggle, where unfortunately capitalism at an extreme of intelligence capability and manufacturing, ends up with a similar social structure to communism. Small elite and masses of lower classes, either working or sustained by the state.
having worked almost 3 years as a bicycle courier, full-time, i’m certain some people really like what they do/did. would still work in the field if i wasn’t being explored and had a decent payment (like being able to belong to the ~ middle class). sure cycling around the town is not cleaning floors of restaurants but maybe robots replacing cookers may be a problem to some?
either the way, that’s a good point. i’m not in the opposition of technology advancement but when it violates a bunch of rules already discussed, it’s impossible to cope with. copyright that lasts 100 years after the author’s death is an absurd but having rights while you are alive is completely sane, unless ‘money’ isn’t necessary anymore… supporting (the current state of) generative AI is plain wrong, no matter how pro-social your utopic visions wish to be