Reading through that old post, it's truly staggering how much LLMs and agents have improved in less than 3 years. And it was just as amazing how far they had come in the few years prior to that - they basically didn't have any useful capability before the 2020s. I remember that talk about achieving any sort of 'AGI' was still a fringe concept in 2023, which very few people would admit they considered was actually a possibility in the near future. I remember telling a colleague, when he asked about what to expect with LLMs, that in a few years we'd simply tell 'GPT 6' to 'build an application that does ___'. That seemed like ridiculous science fiction at that moment...