2 ms·
I prefer fusion power as the go-to vapourware technology. It’s been “10-20 years out” for 70 years and counting. I don’t see any reason to believe that “AGI of
by chongli 11mo ago
I prefer fusion power as the go-to vapourware technology. It’s been “10-20 years out” for 70 years and counting.
I don’t see any reason to believe that “AGI office workers” will be ready to go by 2030. All signs right now are pointing to a looming plateau in their capabilities.
- marcosdumay 11mo agoFusion has halved the "interval until we get it" on those 50 years. AGI has doubled the "interval until we get it" twice already since 2022. Those things are not the same kind of vaporware.
- gjm11 11mo agoCould you name a specific person whose estimate of when we might get AGI has doubled twice since 2022? Or do you mean you found one person with a really short estimate in 2022, another person with a longer one in 2024, and another with a longer one now? Also, if you compare with 50 years ago, AGI has also (better than) halved the interval experts are commonly predicting since then. (Of course the experts could turn out to be hilariously wrong, for fusion or AI or both. I just don't think your comparison is anything like apples-to-apples.)
- thfuran 11mo agoFusion is always years away because funding has been at or below the minimal "fusion never" levels since the '70s: https://fire.pppl.gov/us_fusion_plan_1976.pdf https://fire.pppl.gov/us_fusion_plan_1976.pdf
- otabdeveloper4 11mo ago> pointing to a looming plateau Latest LLM's have already been declining in performance. It's not a plateau and it's not looming. We don't know what the hell we are doing. Adding more training data and parameters only degrades LLMs after a certain threshold.