On: My Real Problem With AI

Below is a series of replies that turned into a really long discussion about AI in software development and it kinda perfectly encapsulates my thoughts so instead of rewriting all of my thoughts I figured I'd just paste the whole thing into a blog post. So here we go!

Well most of the work that I'm doing with agentic development is going towards tools like @thesocialwire.app, which is designed to collect all of the standard.site and skyreader.app publications a user is subscribed to to create a unified reading experience for the user. The Social Wire has an explicit focus on preserving the website's content as much as possible, even refusing to pull the content off the site if embedding is blocked. Its software design is meant to preserve and encourage people to design their sites how they want, while making it easy to read.

My other major application that I'm building right now, skej.at is focused expressly on enabling small businesses to maintain their social identity across Bluesky and the ATmosphere at large, with easy tools to manage multiple Bluesky accounts from a single place, starting with scheduled posts.

Presently I've done all of this for free, financed by my full time job. I'm incredibly lucky to be able to dedicate some of my disposable income to building tools for a decentralized social network. But my day job is demanding, often taking up 50+ hours a week and a lot of my mental energy.

How I square that with the global AI zeitgeist comes from over a decade of following AI research and understanding the science behind what makes a good AI model and a good computer system. Large models with dirty data, running on centralized, inefficient datacenters run counter to the most basic rules of data science. The reason we're seeing a lot of this growth in bad design both on the part of the datacenter and the models is in the pursuit of profit and power.

At its core, what makes a good statistical model of any kind is a small, high quality dataset, running on as small of a model as possible on as efficient compute as possible, a generative AI model like an LLM or a Diffusion model is no different.

We're not facing problems with datacenters or theft of IP, we're facing problems of the pursuit of profit and the treating of people as a means to an end, which is an issue that exists well outside of the GenAI world.

So that's what I fight.

Politicians like Mamdani and the DSAs who actually respect PoCs, supporting small businesses and buying local, speaking out against the current regime at large, building a version of the internet not controlled by billionaires, that's where I focus my energy—on building a world that incentivizes the proper use of technology and brings up the world around it, that blends in to the lives we create.

Not on some symptom of the problem.

And I can do that through many means, be it my writing, my music, or my code.