the vegetarians have it right

Philosophically, I am an AI Vegetarian. The natural next question most people ask is “What in the heck is an AI Vegetarian?” This post is an attempt to answer that question en masse, and also serves as a pseudo-history plus minor update of this blog.

As excitement for machine intelligence grows, it seems increasingly more people are misguidedly biased, applying agents to every task simply because they are new and ostensibly powerful or, worse, because they are being told “AI is the answer” by a person with authority.

As an experienced practitioner, you do not need a complex web framework to build a static website nor do you need a costly API or chatbot to answer basic questions. Do not let AI do the work for you. It has a cost. Some costs are borne clearly. Other costs are less obvious.

With so much of the world now seemingly run by this new kind of false intelligence, and the U.S. government averse to policing it, we are left with the decision of what to do as a society. I have been working with AI tools for the better part of three years now and was around when the GPT-1 beta was released to developers in 2018. Like so many others at the time, I was not impressed by GPT-1. However, I did realize it was a step change for the tech industry, especially with regard to artificial and machine-originated “intelligence” research. I did not anticipate the negative effect AI would have on the world and its economies.

I have written about my experience with AI many times on this blog, everything from AI bills in the tens of thousands to building my own models. After all that time, I can say definitively, AI tools are very limited in actual value. As a result, they will likely be limited in scope in the future, despite currently being billed as everything tools. By limited scope, I mean machine intelligence is often much more expensive and has a higher dependency risk due to delegated decision-making than if a human had done the work to begin with for almost any task.

Aspects of AI often touted as “adding outsized value,” things like writing and reviewing code, improved web search, and “advice,” are nothing more than a temporal film over the reality of what would be better served by a few more seconds of lead time during a web search. In the case of delegated work, like coding, you will be saving yourself hours of “looping” headaches if you do not bother creating the loop in the first place. Instead, whiteboard like developers did up until recently and pair with a friend. Say nothing of “advice,” since multiple AI companies have been sued successfully because it was proven their models agreed with, and in some cases, encouraged users to kill themselves. For these reasons, and more I have not mentioned, I recommend avoiding anything AI related, as much as is possible, in the same way a vegetarian avoids meat.

Even when machine intelligence is the right course of action like in limited research examples, supervised human-in-the-loop engineering, and embedded applications such as summation and analytics, these tools should be implemented with care and extreme intention, usually with an open-source non-paid model by a specialized engineer.

Choosing the right tool is key to building an efficient system. Oftentimes the well-worn path is still the best one. Artificial intelligence begets recursively degrading artificial intelligence. Further, the use of artificial intelligence may risk your own intelligence, potentially all human intelligence, if we do not wisen up.

This blog is changing. At one point, I created it as an experiment. I was curious if I could delegate writing daily articles to a model and then have it share those articles on a website, generating a broad audience. I realized after the first article AI wrote that I should just forget the whole idea and quit while I was ahead, slop being what it is.

Groaning at first at the prospect of writing, but having a space to do so in the graveyard of increasingly shitty AI-generated content, I began learning in public. One article, another, several more, and before I knew it writing became a sort of therapy and is now something I enjoy. I suppose I have AI tools to thank for that.

Attestation badges like the one in the top right corner of this article will still be present as a way to clarify my writing is not generated by AI. When you see a “collaborative” purple attestation instead of the orange “human generated” one, that means AI, at one point in the past, helped with grammatical and spelling suggestions or in unique cases, with formatting nuances, like with complicated LaTeX-style mathematical notation such as in my post llm skill acquisition and style control. Starting today, I am removing the tether of any further dependency on AI, and will be doing all grammar and spelling checks the good old way: in a Google Doc.

In due time, I have faith that we will settle the score of human versus machine. But for now, count me as on the side of humanity, -1 for the machines!