Text redaction - if any part of job is about writing, you can now just brain dump everything, structure it more or less, throw into chatgpt, and make it produce a clear and readable article. In my case, the output is better than anything I would ever write.
Ditto all the redacting work in newspapers, intranet etc. The whole field of proofreaders was virtually extinguished overnight.
Marketing agencies - and I spoke to a few - increased their workers' producitivity 2-4 times (sic!), virtually overnight. Anything from writing briefs to writing copy.
Programming - most of my programming work is deep algorithms, so not much help here, but for writing boiler plate code with new APIs, or writing in a language that I'm a bit rusty in, chatgpt is better than anything else.
Customer service helplines / chatbots (and the same for intranet) - we don't see it just yet, because it takes a bit more time to build a good system, but there are probably thousands projects right now worldwide building those for any niche concievable.
Business intelligence - we used, with success, ChatGPT in our deep tech seedfund, for helping out with initial project ddil.
And, essentially, rubber ducking, but for every single field out there. I just discussed with a psychiatrist how he can use even boilerplate GPT-4 as an additional consultant. You need to be aware of limitations of course, but it is already immensely useful in it's current form - and dedicated solutions for medicine are coming very soon.
That's the short-term perspective and low hanging fruits. On top of that, you have thousands projects now, that are figuring out how to apply LLMs to specific niches. It was difficult before, because you had to train your own models, and now you can just fine tune the existing ones, do embeddings, or just plain prompt engineering.
Oh, and also synergy with different AI modalities - we've had a massive growth in voice recognition and generation, visual recognition, and so on. And LLMs are a glue that adds a layer of understanding underneath.