I watch startup pitches every week. There's a flood of money pouring in at the same time waiting to invest in a GPT-4 based tech that disrupts <X> segment of <Y> industry. Lots of people saying they've been trying to do something for years using NLP techniques, or specialized software, but chatGPT does it fine out of the box (eg loans, scheduling, customer support, customer recommendations).

In general this is not going away. Language models are wrong and hallucinate every now and then, but they often get all the basic questions right, and it's clear they'll be more than suitable replacements to humans for many tasks. In general if people engage with these models enough maybe they'll collect enough tuning data to make giant expert systems to reduce error over time. The tipping point in LLMs may not be the tech itself, but being just good enough for mass engagement, where all your users are like mechanical turk training data

Also people will keep finding ways to prune the LLMs to be tinier, once some Fabrice Bellard level genius makes them run super fast on cpu with lower memory they will be unavoidable.