it is the original windows-intel scenario of the 90s. in 40 years computer performance increased 1 million times since 1980s part from hardware, part from software advancements hardware will advance with more dedicated AI ASIC, currently many startups are doing these specialized AI inference chips. software architecture will also advance at same time including optimizations deepseek has done. on application side, analogous to windows, AI scientists will come out with new thinking models to do more advanced thinking. New models likely will require more complex thinking and more compute time/resources but might not be applicable to consumer, but more applicable to scientific/medical/others. china might actually come ahead in lower end consumer space. US likely will dominate in scientific/other space. IMO
hardware will advance with more dedicated AI ASIC, currently many startups are doing these specialized AI inference chips.
software architecture will also advance at same time including optimizations deepseek has done.
on application side, analogous to windows, AI scientists will come out with new thinking models to do more advanced thinking. New models likely will require more complex thinking and more compute time/resources but might not be applicable to consumer, but more applicable to scientific/medical/others.
china might actually come ahead in lower end consumer space. US likely will dominate in scientific/other space. IMO
量子计算应该是2040年的事了
硬件就是architecture innovation
嗯, 好像是这样的, 人类学开车, 肯定不需要预先认 1万 条路, 特斯拉认 1万条路, 也白搭。 缺少基本的逻辑思维, 尽管语言方面已经很惊艳了。