Predictive Text, Collective Delusion
I have it on very good authority from someone whose Master's focus was on actual AI, that what we are looking at is not AI. It will never be AI, it cannot be AI. LLM is a giant ball of expected responses to inputs. It does not reason, it is very easy to poison, twist, and otherwise tamper with. We have at least one article where they admitted that AI could never be "secured". Expert systems, neural networks that have been trained on matched patterns and work efficiently in specialized spaces not meant for abstraction, are much better solutions in almost all solution spaces.
And these companies want to build it into everything, appear to be in fact be committing several flavors of financial fraud (round-tripping, for one) chasing the hope that by throwing enough computing power at this thing, they will have something close enough to a sentient mind that they own, to please their shareholders. We should just call it Skynet and get it over with.
We are seeing a cost spiral to prop up a bubble that is starting to develop leaks. Micron is getting out of consumer memory. AMD and Nvidia are doubling, tripling prices. The resource costs in terms of power, water, are enormous, and they are concealed from those that should be concerned by them. Meanwhile, Intel appears to have not forgotten that consumers got them where they are, and are developing GPUs for the mass market, not data centers.
I, for one, am watching carefully and will be taking note of those that abandon their end users to chase imaginary profits... and those that do not.
Biting the hand that feeds IT
