AI help me with my development work and my view on it.

It's been a while since I last wrote about my development work. Lately, I've spent a lot of time on the both product side and the marketing side. AI tools have helped me build several features and utilities to support these marketing efforts. While it's clear that AI is a huge helper, does it mean it will eventually take over my job?

First of all, I dislike the term 'AI'. I belive it is a victim of 'meaning inflation'—a buzzword marketing teams slap onto everything to sell products. I've grown frustrated with similar hype cycles throughout my life. Yes, how you define 'intelligence' is a big question, and what about consciousness? Does intelligence require consciousness? Drawing from the book Gödel, Escher, Bach (GEB), I know there is a fundamental difference between static symbols on a page and the human mind that gives them meaning. But that goes beyond the scope of this post. The traditional Turing machine model has limitations, meaning there are problems it cannot solve. We - humans - have been able to stand outside the box and seeing the limitaions and solving the problems.

So what I'm discussing here are tools that help me and free me from boring, repetitive work — things like SEO best practices and content creation for social media. I know best practices exist out there to implement, but I don't have access to them, and that's exactly where these tools help and solve my problem. Take something like implementing in-app purchases: this must have been done millions of times, and there are established best practices to follow. But I still have to spend the time and energy to find them and implement them myself. Or take the navigation model in a SwiftUI app — I don't know what other developers have tried or what practices they use, so I end up investing time to build one that fits my case from scratch. The LLM works like a "knowledge base," and I can query it using natural language. It understands context, so I can narrow the scope or tune the direction slightly to zero in on the right solution for my specific case.

As for the limitations, I believe there's a gap between remembering a solution and understanding it — between memorizing and thinking. An LLM is guessing the next token at every step, and it's surprisingly useful. But it's still guessing, just with high probability. It reminds me of the random walk problem: you start at the origin and can move in any direction with a step size of 1, with equal probability for each direction. You'll find that after several iterations, you end up far from the origin. I think LLMs are like this — after several iterations, they start to drift away into nonsense.

Yes, that's the main point I want to get across. A person who pretends to understand something is often highly valued in our society — or at least in my experience. And others who do understand, but lack the ability to express it, are often undervalued.