GPT-4 is about 95% less hallucination for my use cases (coding, brainstorming, planning). GPT-3.5 is actually stable enough with some training. You can use it

GPT-4 is about 95% less hallucination for my use cases (coding, brainstorming, planning). GPT-3.5 is actually stable enough with some training. You can use it on the typical grunt stuff: formatting text, sorting a pile of data into tables, writing boilerplate code. It converts an API to code that accesses the API perfectly fine, with a lower rate of mistakes than I make. But you need to give it examples. I use it as a form of metaprogramming, where I give it instructions and it spits out the code with caching and all, better than what Copilot can do. As a rule of thumb, treat GPT 3.5 as a worker with a thousand years of experience and an IQ of 80. GPT-4 is more like one with an IQ of 130; you don't have to prompt engineer and all but it's not omniscient.