Yeah, it's the perfect use case for generative AI. I worked on it because it seemed like a hard and niche enough problem, and yet not important enough that someone else would try to solve.

It follows certain patterns, aka tropes. Story tropes have been around since Classical Greek civilization. It uses procedural generation to make sure that characterizations don't conflict. It uses some techniques to make it easy to get into the character rather than just being a third party description.

As you can see, it's far from perfect. Search traffic has gone down about 50% but direct users have gone up by about 30%. So it does appeal to a niche.

I've been using GPT-3 as a kind of "pre-render" because it's so damn tiring to write descriptions of someone being brave. But it's not quite good enough, and ChatGPT can just do the whole thing quickly and for almost free.

ChatGPT has a huge weakness here though - it doesn't quite know what makes an interesting character, and usually the people asking it don't know either. It's about setting up tension. There's intention and obstacle. Plots and characters usually need to be simplified to make the story easier to follow.

With GPT-3.5 being so cheap, we could actually plug in AI directly here on a free app. But at 5 cents per character it still adds up to $100/month.

Then again, there's interesting stuff, like more directly relevant content rather than this genericness. It can generate incredible detail on articles of clothing. And so on.

Gluing multiple systems together for a full project - whether it be web, CI/CD /devops style problems, or large systems engineering problems will surely almost never be replaced by AI. The larger the scale the more amount of places to be wrong, the more BS to debug if you get AI to generate it because it will have a million issues and nobody will know where to start looking.

I feel a sense of urgency to generate wealth before AIs become capable enough that those who didn’t make it become stuck in some sort of UBI limbo.

On the other hand, when trying to apply GPT-3.5 and 4 to do useful work, their limitations become so glaring that I think most takes on the internet in the vein of "chatGPT made an app" are incredibly naive.

Maybe all we need is to finetune a LLM for code optimization. But first you need a large enough codebase handcrafted by phds. But again not many phd are even qualified for performant code.

There will be more competition, which means both more accessibility and more bad quality.

On your side, you may have some aid to be more productive - /may/. The projects you are working on can be "redundant" only in terms of "becoming part of a number of other spawning projects doing similar things, owing to new global levels of accessibility, and competing with a lot of noise, owing to new global levels of bad quality outputs".

First I am writing code for my own pleasure. I am doing it because I like putting my thoughts in this form...

Second AI will first eliminate derivative work, script kiddies, sweatshops etc. I won't be missing them.