I would argue we have seen return. Documentation is easier. Tools for PDF, Markdown have increased in efficacy. Coding alone has lowered the barrier to bringing building blocks and some understanding to the masses. If we could hitch this with trusted and solid LLM data, it makes a lot of things easier for many people. Translation is another.
I find it very hard to believe 95% got ZERO benefit. We’re still benefiting and it’s forcing a lot of change (in the real world). Example, more power use? More renewable energy, and even (yes safe) nuclear is expanding. Energy storage is next.
These ‘AI’ (broadly used) tools will also get better and improve the interface between physical and digital. This will become ubiquitous, and we’ll forget we couldn’t just ‘talk’ to computers so easily.
I’ll end with, I don’t say ‘AI’ is an overblown and overused and overutilized buzzword everywhere these days. I can’t say about bubbles and shit either. But what I see is a lot of smart people making LLMs and related technologies more efficient, more powerful, and is trickling into many areas of software alone. It’s easier to review code, participate, etc. Literal papers are published constantly about how they find new and better and more efficient ways to do things.
OCR was more my thinking, not Pandoc. LLMs enable OCR to achieve greater accuracy through context enhancement for example.