I think this is just an example of 'Selling stupid stuff to stupid people'. It's ms-dos on the corporate desk in the 80's. Did it ever do anything? Googs is now trying to extract its pound of flesh. I have great faith that the 'productivity curve' will remain its normal straight line, nothing to get excited about.
Like normal productivity improvement, just expect the lowest of lowest jobs to be replaced. Anybody in that job like 'mail girl' in the 80's was always moving to another job higher up. Farm work, construction work, all replaced by machines.
ps we have the fun economic scaling effect, which is consume all the resources, fire all the people, and wonder why nobody is buying the (very expensive) service. Then we have the physics scaling effect of viewing the chip as a big screen of lights. The query goes in, interacts with training, spreading like a stone-pond-ripple, and then consolidating for an answer. Old days had the pond too small, and the blobs bounced on the sides. Then the whole boom was the 'sweet spot' of sucking in everything for training, and have the query and answer bubbles stay without edge bouncing. Now we go too big, with the training data at a limit, and the query-answer bubbles staying well within the boundaries. This is bad, as it creates dead zones in the corners that are not regularly trained. They loop around themselves and create hallucinations that never get to training. Bigger and bigger means more of this. Stealing all the books in the world reaches a limit, and queries are stupid, and not uplifting. The whole thing dies.
No comments:
Post a Comment