Good to see more on this; LLMs are inherently inefficient, and new generation improvements mostly come from making that even worse. Pair this with NVIDIA's earnings call. They see the future of AI so processing-heavy the limiting factor will be the cost and difficulty of powering the data center.
whoever solves the power generation conundrum for this will be fabulously wealthy in a very short period of time. by solve, I mean less expensive, zero emissions, 100% reliable or near it.
PUCs need to start allowing new generation to go into base rate. When new housing gets built, the homeowners are charged to build all roads, water, sewers, parks, libraries, etc. The richest corps in world don't even pay their power hookup
Spell check is out of control - STOP allowing all new generation into base rate
Let’s hope someone, anyone, listens and halts the needless “we must be A.I. first” management edicts.
The hyperscalers want everyone to believe they will be the only AI, but a lot of AI will be local or embedded. It will be better and cheaper.
Yeah, I don't see how they could ever make it cheap enough to compete with me just buying my own GPU once.