Monday, May 01, 2023

Some thoughts on large language models (LLMs)

I wrote this post several months ago but never really got down to publishing it. Just thought I would do so today, even though the ideas are a bit raw.

The AI techniques used are broadly categorized under deep learning, which is more or less mirrored on the way the human brain actually learns. Kind of like creating a digital version of neurons in the brains and using those artificial neurons to learn something.

The problem is that, for a fan artist, he or she probably took months, if not years, to learn to draw and mimic someone else's art style. He or she may take hours, if not days or months, to produce a piece of art in that style. Computers, on the other hand, have the advantage of not having to eat, sleep, or even move a brush or mouse; they can be trained in days, and produce thousands of pieces of work in an hour.

It is a really complex issue, but the bottom line is this: given that these digital brains are now capable of mimicking humans, what's next? When the industrial revolution came, humans were freed from manual work to take on other forms of work that relied more on human intellect and cognitive abilities. But if AI is now able to take on work in this realm, what is the work that humans are left to do? If humans are replaced by machines, our economic system, where humans earn money based on the work they do, and use that money for the allocation of limited resources, will become obsolete. If we do not first answer the question of this next-generation economic system, we are going to end up with chaos in society as AI slowly replaces sector after sector.

I think there was a Disney movie called Wall-E, something like that, where humans basically just do nothing but "enjoy" life while AI and robots do everything. But if you look at it in another way... all those humans ended up having the same value as each other, since none of them actually do or did anything consequential.

The Wall-E model works only if we have an abundance of resources, so much so that we can fulfill everyone's desires. If not, if resources are limited, then how do we decide who gets what, when no one works?

Look at it this way. If AI and robots can do everything that we can, we don't have to work. But the robots end up having to support us freeloaders.

Really random thought: Given that resources are limited, one day, the robots are going to think about efficiency, about efficient allocation of resources. To robots, they are not declaring war against humans. They are just running an optimization function. Just so happens that the optimization of planet Earth means getting rid of the freeloaders...
We will be the pets of robots. If they destroy all human life, they end up with no purpose. Then they stop functioning. So instead, they keep a few humans around, make sure they are happy to fulfill their programming of "to serve humans and make them happy".Robots will be our masters, but at least we will be happy (at least, those of us lucky to be kept as pets)





No comments: