Why
I'm a data engineer by trade, and if I may be candid, this AI boom has roughed up my mental health. On a daily basis, I've been bombarded with how XYZ model is infinitely better than another. Or how a new "AI agent will replace you before the end of the day." It has sincerely weighed on my mental state.
But my intuition says there's a lot of manufactured fear. My theory is, these AI companies adhere to the old adage, "No press is bad press." So, they continue to fan the flames of fear. That causes ClosedAI and their brethren to generate angsty copy for the Media. The Media laps it up, as the fear causes engagement. Those engaged are the CEOs of large companies looking to reduce their single greatest expenditure, labor costs. My elected officials do nothing to prevent the hamster wheel of fear, as they are being bought by the Tech Oligarchs. And even if they did want to intervene, they are too ignorant of the details of AI technology to do much.
Pass the foil? I need make a hat.
Living in fear is a horrible state. As a hacker (the build stuff kind), I've always found the best way to reduce fear is by understanding better what is causing the fear. And that's the point of this series--understanding enough of the nuts-and-bolts of "AI" to reduce the fear, and maybe, even find some joy in the madness.
Exploring the Singularity
The first thing I'm going to attempt is to create a series of Python packages wrapping different AI models. These packages will have a sole purpuse, e.g., converting text-to-speech. Regardless of the utility of the package, it will be wrapped with a FastAPI server to allow RESTful interactions with the utility of the model.
The purpose of these componentizing and serverizing these AI packages is to begin to string them together in a semi-distributed system, with the goal of eliciting uncanny behavior, similar to human intelligence.
As a concrete example, I plan to create the following:
- text-to-speech (TTS) server for converting my speech to text
- speech-to-text (STT) server for converting text to speech
- A chat model a server, for reasoning through unstructured text
