[russellbeattie]

Those of us who have spent the last several decades learning how to break down information technology problems into their constituent pieces, many times down to the byte, are being suddenly presented with a system that requires none of that expertise and produces results which are nearly as good. The more you know about technology, the more you are amazed. How long would it take you to code by hand some of the things that LLMs can do with a simple prompt? The less technical people are just like, "Oh, the computer is doing magic again, shrug," because they have no idea the difference between what Siri does and what ChatGPT is doing. The rest of us however are mind boggled by what we're seeing. When I saw how Microsoft "programmed" the Bing chat (Sydney) simply by prepending a series of rules in plain English to the prompt, I nearly fell over.

Prompts are like programming without the details. Like a product manager's user stories - simply describing what is expected is enough to get the LLMs to spit out useful results. And it's getting better faster than any of us thought possible. Even if progress stopped today, what's already out there is going to have a profound impact on office work, and thus society, as we know it. But it's not stopping.

I've explained to my 20yo that he needs to learn to embrace the new tools as soon as possible, understand what they can do and can't do, and make sure he focuses his skills and education on the bits that AI aren't great at. As for me, I'm focusing on figuring out the best way to package the power of AI in various niches so that I can use it to provide that power to users in an easily understandable way, which is what I've always done in tech. I'm just using going to be using different tools to do it.

There's plenty to be done to train models, feed AIs data, and package the output for use in a million contexts. That Quick Oil Change company down the block still entering information on Windows GUIs (or worse, DOS computers) aren't going to see AI affect their interfaces for another decade. Between now and then there's plenty to do.