More AI

Another week passed, and I am taking a moment to synthetize everything that has happened to a short essay to update my understanding of the world. I am grateful to be in a such a stimulating environment, both at the university, and at the company, which both bring new things to think about.

This week I was at a short presentation by Forgent AI who are trying to apply AI agents to help companies to apply for public tenders. However, the talk was not about that, the talk was about how AI coding assistants, Claude and the like, are changing how their company works. This made me reflect about the fact that this is now a reality: the world is changing extremely fast, and is time to adapt. The cognitive work is being automated, and that means we have to start adapting. We have to adapt the way we collaborate – organisations will change to work faster, and with less human supervision. And as individuals, technical skills in some domains will matter less. I am thinking a lot for myself what I should be focusing on. It is certainly still valuable to learn and understand concepts, design good experiments and architectures, be able to explain value of things to others. But current skillset I focused on (mostly engineering) will not be that valuable anymore.

When I was in the middle of thinking about this, I came across this talk by venture capital firm Sequia Capital on YouTube. The pointed three interesting takes:

Artificial Intelligence is to cognitive work the same as photography was to painting.

Before we had cameras we had to paint to capture things. There were many people learning to do this, and be as accurate as possible. However, then cameras came and the need to capture the reality by human labour has disappeared instantly. Art has remained though, but it started focusing less on the technical, and more on the description of the human feelings. This is beautiful. It also means that in the future we will still do what we do now, but with different motivation.

Artificial Intelligence is the steam of the new industrial revolution.

Before industrial revolution, most of physical labour was done by humans. This has changed quickly with the usage of steam. Small tasks are still done by humans, but bigger operations are automated. Same will be done to cognition. You might read one document, but nobody not hundreds of them a day, that will be automated. The demand for cognitive work will sky-rocket as it gets cheaper, and humans will still be doing small specialized parts of it, but not the bulk.

Large language models are formed by data and rewards. In contrast to human creation which are formed by curiosity and motivation.

This is noted by Andrej Karpathy. I think we should keep that in mind, since it shows us that LLMs are fundamentally different than we are. That informs how they work, and what they will be good at. Good to keep in mind when using them.

As a final thought, I think we should not be scared by this. I at first thought we are all doomed: AI is gonna take our jobs. But no, world is not efficient. There are millions of bullshit jobs. People are not paid for what they do, they are paid for positions in organisations, and their responsibilities. Those will still be needed. Actual labor not, but position and prestige is still there to be fought for.