| The "Why AGI will not happen" argument here seems to hinge a fair bit on hardware limits and "computation is physical" but the arguments don't seem very good. Much better in my opinion is Hans Moravec's 1997 paper "When will computer hardware match the human brain?" which seems quite solid in it's reasoning - he was a roboticist and spending time trying to make robots do things like vision and given at the time they understood the retina quite well he could compare the amount of compute needed to do something equivalent to a given volume of neurons and then multiply up by the size of the brain. Conclusions were > human brain is about 100,000 times as large as the retina, suggesting that
matching overall human behavior will take about 100 million MIPS of computer power 100 TFLOPS roughly. Or similar to a top graphics card now. Also: > Based
on extrapolation of past trends and on examination of technologies under development, it is
predicted that the required hardware will be available in cheap machines in the 2020s. which seems to have come to pass. I think we don't have AGI yet because the LLM algorithms are very inefficient and not right for the job - it was more a language translation algorithm that surprised people by getting quite smart if you threw huge compute and the whole internet at it. (Moravec's paper which is not a bad read https://jetpress.org/volume1/moravec.pdf) It also has >This paper describes how the performance of AI machines tends to improve at the same pace
that AI researchers get access to faster hardware. My guess is that will come to pass. AI researchers have the hardware and the software will improve shortly. |