1. the training & inference costs have dropped to triple digits and a phone app, respectively.
2. given the preexisting codebase for distributed training, some non-negligible fraction of the billions of GPUs are going to be volunteered in an exascale fashion not unlike Folding@Home.
3. given that many business processes have already been articulated & opensourced in natural language, effectively any SME has the means to finetune their own nuances & SOPs to drastically lower training costs and turnover for new employees. this is a multimodal trend, any apprentice in the world can snap a photo of what they're doing and ask an LLM what to do next. eventually, it will be video if that modality can be inferred on mobile hardware.
4. admission to the bar and license might be the bottleneck for lawyers, but it is no longer the same bottleneck for incorporation and other legal services
5. given how much operational budget in hospitals goes to administrative work, I'm curious to see how the people deal with their medical bills in the next couple of years.
6. we haven't even confronted garage-tier sentiment analysis. I genuinely wonder how many markets get arbitraged due to this, starting with social media dogfooding.
7. what's the necessary cost of mainstream journalism to the general public? I'm sure you'd agree that should be weighed. same as 6), what's newsworthy & why should it be published by a corporate media company?
8. on the tail-end to this, legislature & lobbying costs just got profoundly cheaper. also cheaper to pick apart pork-barrel or other inconsistencies therein.

these are just a few downstream effects. and I'm leaving out the parallel gains in manufacturing automation, machine vision, crowdsourcing, etc.