The problem is that a majority of the supply was bought out by AI companies who couldn’t compete by offering a better product, instead needing to make them be the only ones with hardware to run models. It isn’t a problem inherent to the technology, but one manufactured by American billionaires riding on over speculation while it’s still hot.
If anything, these companies are holding back machine learning from reaching its true potential by wasting all the hardware on overgeneralized slop. The most productive uses of machine learning aren’t most of what these fuckheads use it for or force down our throats. The increases in productivity for most uses of LLMs are marginal at best, while more specialized models can save orders of magnitude more time by doing very specific things. However, all that hardware being dedicated to corpo models makes it more expensive for people to run what they actually need.
If silicon was manufactured like any other technology, by thousands of companies all around the world, rather than by a handful of companies who all work together. Or even, by nationalized government branches, after all computers are a key piece of what makes modern life possible (not just for entertainment but to provide the services for a healthy and sanitary life). Then this whole fiasco wouldn’t be feasible or even possible. The silicon supply chain is fragile and there is little being done to fix that.
The problem is that a majority of the supply was bought out by AI companies who couldn’t compete by offering a better product, instead needing to make them be the only ones with hardware to run models. It isn’t a problem inherent to the technology, but one manufactured by American billionaires riding on over speculation while it’s still hot.
If anything, these companies are holding back machine learning from reaching its true potential by wasting all the hardware on overgeneralized slop. The most productive uses of machine learning aren’t most of what these fuckheads use it for or force down our throats. The increases in productivity for most uses of LLMs are marginal at best, while more specialized models can save orders of magnitude more time by doing very specific things. However, all that hardware being dedicated to corpo models makes it more expensive for people to run what they actually need.
If silicon was manufactured like any other technology, by thousands of companies all around the world, rather than by a handful of companies who all work together. Or even, by nationalized government branches, after all computers are a key piece of what makes modern life possible (not just for entertainment but to provide the services for a healthy and sanitary life). Then this whole fiasco wouldn’t be feasible or even possible. The silicon supply chain is fragile and there is little being done to fix that.