A Step Back...

### Childhood Interest

I have a deep love and appreciation of just about everything, I think. I feel a lot of things most of the time. I mostly zero in on electronics and computers though.

When I was a kid, maybe 8 to 10, I watched Minecraft videos and dreamed about owning the game so that I could build my own redstone contraptions. I would fill out a notebook of grid-paper with my redstone designs (not a single one functional btw).

A little bit older, maybe 10 to 12, I was allowed to use my moms laptop for a short time each day, during which I would write adventure games with batch script in the notepad, and spend all my time on the little details. I spent one entire day trying to create an ASCII stick that you would find while walking through the woods. I would graduate to learning Java, still in notepad, and making more command line games with it. I never missed an opportunity to create a batch game in the computer lab and show those next to me. It was magic to them.

I eventually was able to get some money and get myself a cheap laptop around 16, which allowed me to explore deeper. I installed visual code, unity, blender, gimp, lunar magic, etc.

Back to being 12, I would take apart electronics: cameras, remotes, flip-phones and just be enamored by the components. A few things I was even able to put back together and they would still function. I remember one time I took apart a camera and was shocked by a capacitor that controlled the flash. Those hurt!

Everything I discovered would only deepen my curiosity. I wanted to work on electronics I guess. I don’t really know if I had a clear vision for my future, but in 9th grade I got my DC electronics technician certification from the vocational school, where I learned to diagnose and repair electronics. Soldering and all that. DC technicians are a product of a long gone time, however, because most things are just micro-controllers now. I put my certs into Indeed and it serves me local janitor positions.

### College

Anyway! Micro-controllers everywhere? I love coding so it works out. Let’s go to college for software engineering. I breeze through my classes. I literally haven’t gone over a single thing in college that I didn’t learn when I was 14. I’m about to graduate, and it’s just been so boring. Everyday I think to myself, ”Computer science is so cool. I can’t wait to finish college so that I have time to study it.”

Something interesting is happening in the background. Of course I had been following the developments of generative AI for a while. Well, first I had been following neural network stuff. People making neural networks play Mario, older chatbots, etc. Then I was aware when generative AI started coming out. Image generation, advanced chatbots, voice synthesis. It was all very cool. I enjoyed watching OneyPlays videos where they would interact with generated images and stories.

There was a new interesting development called Chat-GPT. It was an incredible language model capable of writing college level essays. I took a look at it when it came out, read all of the scary stories about how it was going to take over the world, tested its abilities, and was not worried at all. I know many of my classmates used it for everything, but I knew that the quality of work produced by LLMs was far below what I was capable of. It always needed fact checking, reformatting, etc. Still, it was an incredibly valuable tool and saved me a lot of time with the more mundane tasks.

Around GPT 3 to 3.5, I noticed that it was pretty good with technical stuff, so I would feed it my notes and powerpoints and ask it to quiz me on the material until I had mastered it. I would answer 150, 200, 300 generated questions until I couldn’t get them wrong and then go in and score perfect on exams. I did this for chemistry 1 and 2, and some math classes. It worked so well that Chat GPT is standard if I’m studying or trying to learn something. It is incredibly valuable as an educational resource. It can be used like a private tutor.

Going back, there was discourse around how dangerous this technology could become. I knew better however. I knew that it was pretty incompetent and needed me to hold its hand to produce anything valuable. I also knew that there simply was not enough power, water, and hardware in the world for generative AI to scale to what they were talking about. I cited pipeline and refinery destruction, energy crises throughout the world, lack of energy infrastructure, reluctance to adopt nuclear and renewable, lack of financial viability, etc. And I was 100% correct if every body acts logically.

I remember joking with people at the time about a schizo accelerationist post I saw at the time, saying that humans were the catalyst for a higher form of artificial life that would take our place. I would repeat it with a straight face and make all the arguments for it. I can’t help but think about that now.

### Present

I noticed that generative AI is getting pretty good these days, and much quicker than it used to. It has far surpassed my abilities, and it feels like not using it would just be a waste of time. It feels like we are being pushed to take a step back, and let generative models do all the work, or else be left behind. Modern LLMs are so extremely capable, there’s nothing funny about it anymore. There’s many great things about it: accelerated research capabilities, cancer/medical research and treatments, a knowledgeable personal tutor for any subject. 

### A Step Back

I am sad and excited for the future. I am excited about the things I will be able to learn, and the new technology I will be able to see, but I am sad that my biological limitations will soon be surpassed by machines. I am sad at the idea of becoming obsolete before I even start.

I try to be hopeful. I think gen AI is a valuable tool, and it still takes a human to use it. I think that anyone could have invented the paperclip, but only one person did. I think gen AI just makes everything as simple as a paperclip, all it takes is the idea. But that is perhaps too optimistic.

I try to be optimistic. Things that used to take me minutes now take seconds. Things that took hours take minutes. Things that used to take days take an hour. Things that used to take me months take only a day. Things that would take a year take only a week. Things that would take years take maybe a month. Things that would have taken me a lifetime, are achievable in months. I heard it said, not sure where, that better tools will only increase ambition. But that is perhaps too optimistic.

### Where We Are

This doesn’t really fit, but I can’t stop thinking about it. I was listening to an interview of Eliezer Yudkowski, per Atrioc. Yudkowski, to demonstrate the different levels of intelligence that exist, told the story of Einstein discovering general relativity. He described how Einstein was able to form this theory with very little data to work off of. In Yudkowski’s model, Einstein is the highest level of intelligence. He can make an incredible inference using very little data. The next level of intelligence, is all of us. While some individuals may not be as smart as modern frontier models, we have to remember that we are working with a lot less data. Frontier models use literally all of the data ever created, while we are limited to a relatively small set that we are able to digest. With that in mind, our ability to make inferences from a limited amount of data is far superior to even the frontier models. Then the lowest form of intelligence in this model, gen AI. It is not the lowest form because it is stupid, it is the lowest because it takes all of the data in the world to make slightly better inferences than humans, who do it with very little data.

This model does not serve to disparage gen AI. In it’s current form its capabilities are incredible. This model serves to show what could come in the future. Imagine an Einstein level intelligence, one that is able to discover new laws of physics on a hunch with very little data, and then giving it all of the data ever created plus a higher clock speed. The result is incomprehensible.