/b/ - Random

>>121670 Computer does not scale linearly or exponentially with realism for a certain model unless mathematical or logical breakthroughs are made.

>>121678 AI is the best it could be... these models are just glorified autocorrect, AGI is at least 5 decades away, the most viable path forward is bio-hybrid computing.

>>121678 *compute

>>121680 Although I like ur pov of bio-compute. But the economy of that will not make sense just from the ethical standpoint. And do not forget everything the ai bro's said came true. The points were: >we need 7 trillion dollars for agi. <Massive build outs ranging from. 500BTO 700B JUST this year expected to increase more in next year. >Every year cost of compute will fall tox in price <The api cost for the SOTA models have actually fallen in accordance to what altman said. >The models will be the level of elementary school level, high school level, college level,phd level and now Frontier mathematicians-scientist levels. >Every single word that came out of tech bros was nothing but truth yet u r still skeptical why? The amount of NPUs that will come online will be so much that every iteration which was taking 6 months in 2024 will take 3 HOURS.

>>121685 While these glorified autocorrects are genuinely nice tools of convience but they are not what they claim them to be, and they never will be, they have already sold the dream and are hoping that if they throw enough compute, it will eventually actually become what they claim it to be, but as I said it doesn't scale linearly. And AGI is not a compute but architecture problem. The few proposals that exist are just multiple AIs layered over each other. The only viable path towards AGI is bio-hybrid computing, muh ethics are a meme, they only exist to discourage research on humans since that may lead to results the Jews don't want to be scientifically validated. If it becomes financial beneficial they will quietly push them aside.

>>121687 Also I'm not on the AI hate bandwagon, I'm just on the pessimistic side of things. I'm actually very much hoping we will get AI assisted drone swarms linked in a kill web with other weapon systems, they take out the equipment that takes out tanks while tanks manuevre to destroy support assets so more drones can be utilised in mass. This forms a solid bee and iron loop. The Chinese are already experimenting with these while Pajeets procure obsolete Americuck equipment. I have some other ideas of my own and those I've read, like 1 sector double penetration, kek. And also adding 2 whole brigades of artillery either dedicated to precision and the other to massed fires.

>>121687 In the end people will be disappointed since the dreams are too unrealistic to be achieved, and then it will settle down to its actual worth.