I have worked on a few projects that use language models over the years. Most of them don't use the models people would call large these days, but have had cases where we used those.

In all of the cases that I have used them it has been for categorization. Basically in areas where we can't write enough if statements to do it well enough so the only option is human review. For the most part language models do great here.

Replacing the first tier of customer support for sure. Every time I see support bots mentioned, some ex-support employees bring up that 90% of those customer support questions are really dumb. The more people you can funnel to your knowledge base, the better.

On a more depressing note, translations. Translators are really hurting from LLM translations. They must focus on certified translations that must be done by a sworn translator.

Many websites got wrecked by Google's reaction to the rise of AI-generated content. The internet got funnelled to a very small number of sites controlled by 16 companies. A lot of independent websites got practically killed.

My use case (not very sexy) is process documentation. I take a basic outline, run it through GPT, manually edit some things here and there, boom, done. 10x productivity in that regard.

According to my testing, Claude 3 Opus makes drafting patent applications about 50% faster.

I read an article about an anime place that fired 80% of their illustrators, because the remaining 20% were 10x more efficient.

As a consultant, it's been extremely useful for rapid prototyping for POC projects.

We were able to fire 200 of our workers and replaced them with an AI chatbot.

Klarna allegedly has saved millions from using AI to replace most of their support folks.

Might just be a PR stunt tho

Agreed, It's just a bubble. I have a feeling now a days these bubbles are intentionally created to scam investors out of their money.