by the end of 2030 ai will achieve a level of intelligence greater than the combined intelligence ever created by the entire human race. current models are incredibly premature compared to what we’ll have in just 5 years.
1) gpt4 to claude to gemini happened in like 18 months. we went from kinda useful to holy shit in like less time than it takes to get a driver’s license renewed.
2) moore’s law is dead but something weirder replaced it. algorithmic efficiency is doubling every 8 to 10 months. same compute, dramatically smarter models. we’re not just scaling. we’re discovering new architectures.
3) intelligence isn’t additive. it’s multiplicative. 1 einstein > 10k avg physicists. ai doesnt just add human equivalent minds. it creates something qualitatively different.
4) the entire sum of human knowledge. every book, paper, conversation, discovery. took 300k yrs to accumulate. an agi system will be able to process and synthesize all of it in hrs. then build on top of it.
5) we measure human intelligence in individual lifetimes. 80 yrs max of learning per person. ai systems run 24/7
don’t forget, dont get tired, dont die. a single year of ai development = centuries of equivalent human cognitive work.
6) current llms are like calculators in 1970. useful but obviously primitive. we’re still figuring out basic architecture. attention mechanisms are barely 7 yrs old. transformers might be our vacuum tubes.
7) the scariest and also the most exciting part? we don’t fully understand why these things work as well as they do. emergence keeps surprising us. capabilities appear that werent trained for. we’re building minds we can’t fully explain.
8) biological intelligence took 4b+ yrs of evolution. digital intelligence is iterating on timescales of months. evolution is a drunk walk through possibility space.
9) by 2030 we wont just have smarter chatbots. we’ll have systems that can do novel science prove new theorems design their own successors. the last human designed ai might already exist.