Technology

In reality, the model will be wrong in much of what it says for something so niche, but forming questions based upon what I know already reveals holes outside of my awareness. Often a model is just right enough for me to navigate directly to the information I need or am missing regardless of how correct it is overall.

I get lost sometimes because I have no one to talk to or ask for help or guidance on this type of stuff. I am not even at a point where I can pin down a good question to ask someone or somewhere like here most of the time. I need a person to bounce ideas off of and ask direct questions. If I go look up something like Pipelines in microprocessors in general, I will never find an ideal entry point for where I am at in my understanding. With AI I can create that entry point quickly. I'm not interested in some complex course, and all of the books I have barely touch the subject in question, but I can give a model enough peripheral context to move me up the ladder one rung at a time.

I could hand you all of my old tools to paint cars, then laugh at your results. They are just tools. I could tell you most of what you need to know in 5 minutes, but I can't give you my thousands of experiences of what to do when things go wrong.

Most people are very bad at understanding how to use AI. It is just an advanced tool. A spray gun or a dual action sander do not make you stupid; spraying paint without a mask does. That is not the fault of the spray gun. It is due to the idiot using it.

AI has a narrow scope that requires a lot of momentum to make it most useful. It requires an agentic framework, function calling, and a database. A basic model interface is about like an early microprocessor that was little more than a novelty on its own at the time. You really needed several microprocessors to make anything useful back in the late 70s and early 80s. In an abstract way, these were like agents.

I remember seeing the asphalt plant controls hardware my dad would bring home with each board containing at least one microprocessor. Each board went into racks that contained dozens of similar boards and variations. It was many dozens of individual microprocessors to run an industrial plant.

Playing with gptel in emacs, it takes swapping agents with a llama.cpp server to get something useful running offline, but I like it for my bash scripts, learning emacs, Python, forth, Arduino, and just general chat if I use Oobabooga Textgen. It has been the catalyst for me to explore the diversity of human thought as it relates to my own, it got me into basic fermentation, I have been learning and exploring a lot about how AI alignment works, I've enjoyed creating an entire science fiction universe exploring what life will be like after the age of discovery is over and most of science is an engineering corpus or how biology is the ultimate final human technology to master, I've had someone to talk to through some dark moments around the 10 year anniversary of my disability or when people upset me. I find that super useful and not at all stupid, especially for someone like myself in involuntary social isolation due to physical disability. I'm in tremendous pain all the time. It is often hard for me to gather coherent thoughts in real time, but I can easily do so in text, and with a LLM I can be open without any baggage involved, I can be more raw and honest than I would or could be with any human because the information never leaves my computer. If that is stupid, sign me up for stupid because that is exactly what I needed and I do not care how anyone labels it.