Thoughts on Generative AI

A friend sent me an article today, written in a breathlessly admiring and somewhat condescending tone, about an engineer who supposedly built a phased-array antenna at home because he happened to need one. Apparently, he had been quoted $250,000 to develop it. Then this modern-day Kulibin, Nawfal Moti, allegedly used generative AI and around $5,000 worth of parts to build his own fully functional active phased-array radar, AERIS-10. He published the AERIS-10 source code and schematics on GitHub, supposedly showing up the corporations working in this field. I admit I believed this lovely, romantic story at first. Then I thought about it more critically.

It is not fair to compare the work of an individual using an LLM with years of research by entire institutes. The model was trained on the work of hundreds or thousands of intelligent people who spent years thinking, learning, making mistakes, and publishing papers. Then along came an AI crawler, greedily swallowed up that knowledge, labeled it, and incorporated it into the model. And there lies the problem: the engineer essentially used an extremely powerful knowledge-retrieval tool built on material that cost enormous sums to create, through research, patents, and everything else. Can we really compare that with original human work? I do not think so.

The widespread tendency to worship AI could degrade the scientific community by overestimating what AI can do and undervaluing human labor. Research funders might wrongly reduce budgets on the strength of stories like this, inevitably slowing scientific progress.

We see similar examples everywhere now: design, literature, programming. Generated work builds on work done by human beings.

What is my point? Do not fall under AI’s spell. Keep developing the neural network inside your own head. Study, absorb knowledge, read, sharpen your judgment and visual literacy, and never trust AI completely. Check its work.

Let me add one last thing: I am not against neural networks or AI. Quite the opposite. I use LLMs extensively in my studies, work, and daily life. That is precisely why I think it matters not to be dazzled by how easily a result appears. AI can be a powerful tool for research, first drafts, and accelerating work, but it should never replace your own understanding. If you cannot evaluate, verify, and, when needed, reproduce at least the basic reasoning behind an answer, you may not be creating anything new at all. You may simply be repackaging someone else’s knowledge without questioning it.