I recently looked at the prices of pre built desktop in my country and the PS5 offers better price to performance, if it is jailbroken that is. The prices are ridiculous. I wanted a desktop to play games, the price isn’t worth it. Every month I see an increase in the price.
Maybe Jesus will kiss me on the cheek and I’ll be able to perfectly render GTA 6 rtx in my mind.
But maybe we live in reality.
(science side of my opnion, LLMs are almost the same algorithm as a search engine in the background, making true novelty impossible. No one has yet created another novel OS like you describe for the AI companied to scrape, repackage, and resell yet, so its impossible for the LLMs to output such an OS.)
(The other other side of it is what do you even mean by a novel, stripped down, adapted OS? Like what has FOSS been developing for the past 40 years lmao? Anyone in the OS dev world knows usually the system itself is the most optimised part of the whole computer. Let’s shit on uncompressed 140gig games if we want to shit on someone)
Probably should have said novel, stripped down, OR adapted OSs. That is to say they could be novel, OR they could be stripped down OR adapted to work on more limitted h/w. So, really, more likely stripped down or adapted; novel is the least likely. In the case of Linux on PS5, adapted.
I see that, and with the new ZRAM/CRAM developments its reasonable (to me) to assume more optimisation is on the horizon.
Unfortunately the problem of novelty is very much a difficult area with AI, as inheritly it is defined by its training data - so finding a new optimisation (which tends to have the greatest performance benefits) is generally out of the question.
This leaves it to reshuffle is training data in ways that appear novel, like running Linux on a PS5, but that just means these options were already possible just unexploited.
I feel like that brings me to the same “benefits through adaptation” conclusion as you reached, but with less optimism. I’m confident enough that existing devs and humans are capable enough to find 99% of these adaptations independently, which leaves any AI devs stuck trend chasing well past an ideas expiration date.