As a programmer, it ended up being useful for a few things in the beginning, but I don't even use it anymore. Maybe every other week to reformat some text or make some boilerplate code.

GPT is really good for finding insertion points into large-breadth topics where you don't really know where to begin. I've found great joy in asking it particle physics questions. I don't know the math to google the right terms, but it's been really good about taking my layperson questions and translating it into answers that are digestible.

I’ve used it as a software engineer most days ever since week 2 when I first learned about chatgpt. It is incredible at making me more efficient by answering dumb questions such as what are edge functions, and getting unstuck on new topics such as how oath2 works in youtube apis because i’m forced to use it for a server to server app.

This with Bing so it searches for articles and reads and summarizes them has been great! I mean, admittedly it'd probably be better if it used a better search engine but at least with this workflow you know it's less likely to regurgitate garbage.

I've used it to find a couple of different python libraries. Asked is questions about top level ontologies and OWL. it was pretty insightful on the ontology stuff.

I've found LLMs to be very useful for, well, text-based things. I've not found the "bot" implementations useful, but they're better tech for summarization, highlighting important sections (i.e. what sentence of this product review should I show in bold, given the search term), and entity recognition (what are all the products mentioned here).

They are expensive to run, in terms of GPU cycles, but they are noticeably better than the previous models.

It's also hard to constrain them well. If you want 95% accuracy, it takes some tuning work. If you also want to avoid 1% total batshit nonsense (repeat "chicken" 50 times), then you have to check for that. Earlier models were sometimes wrong, but they were not quite so aggressively wrong as the 1% case of LLMs.

That's just my anecdotal experience, but it leaves me both optimistic about applications in the right spaces and worried that people are just shipping something that's OK 75% of the time and calling it a product.

To your last point, LLMs can translate your writing into different styles. Try feeding it one of your writings, then asking it "now make it sassy." It's also good at writing low-effort things like Instagram product blurbs.

I like the other comment somewhere on this post, the real uses will just be small improvements. I used image generation to make a quick favicon for a website, it came out nice enough, better then what I could do. I also thing the semantic search stuff with language as a query is pretty useful. I was able to download my emails, generate embeddings, and write queries for my emails better then what the Gmail search would do.