The recent release of ChatGPT is all I am reading about, and I'm having a lot of fun playing with it too. 


It's shifting the goalposts on what is hard for an AI, so it seems like a good point to link to this post - https://clivethompson.medium.com/when-robots-attack-83102761c6d5 which talks about Moravec's Paradox - it's easy to make machines exhibit adult performance on intelligence tests or complex games, but almost impossible to get them to have the skills of a one year old when it comes to perception and mobility. 


This issue of perception is still a constraint on ChatGPT. 


This question of what is easy and what is hard is related to clock problems vs cloud problems - https://onetroy.org/blog-collection/2018/6/19/clocks-and-clouds-emergent-issues-require-emergent-thinking (Ben Sauer brought this to my attention on Twitter). 


- Ian