When access to knowledge is no longer the limitation

Let's do this thought experiment together. I have a little box. I'll place the box on the table. Now I'll open the little box and put all the arguments against large language models in it. I'll put all the arguments, including my own. Now, I'll close the box and leave it on the table.

Now that that is out of the way, we are left with all the positives. All the good things that come from having the world's information at our fingertips. I can ask any question and get an answer almost instantly. Well, not all questions. The East has its sensitivities around a certain square, and the West about a certain island, but I digress.

I can learn any subject I want to learn. I can take the work of any philosopher and ELI5 it. I can finally understand "The World as Will and Representation" by Schopenhauer. A friend gifted me a copy when I was still in my twenties, it's been steadily collecting dust ever since. But now I can turn to the book and ask questions until I thoroughly understand it. No need to read it cover to cover.

In fact, last year I decided I wanted to learn about batteries. I first went to the Battery University website and started to read lesson by lesson. But I had questions. How was I going to get them answered? The StackExchange network is not what it used to be, so I turned to ChatGPT. It had all the answers. I learned and read so much about batteries that I am tempted to start a battery company.

My twin boys are at that age where they suffer from the infinite WHYs. Why does it rain? Why does the earth spin? why does California still use the Highway Gothic font on some freeway signs? I do not have answers to these questions off the top of my head, but I have access to the infinite knowledge machine, so of course my kids know the answers now.

Just the other day, I had a shower-thought about cars. "Are cars just a slab of metal on wheels?" And now I learned that the answer is "essentially yes." But then I kept reading on the subject and learned about all those little devices and pieces of mechanical technologies that exist that I had never heard of. For example, the sway bar link. Did you know about it? Did you know that it reduces body roll and maintains stability during turns? Fascinating.

Ever since LLMs made their public debut in 2022, we've been gifted with this knowledge base that we can interact with on demand, day and night, at work or at home. The possibilities seem endless. I can learn or understand any codebase without being familiar with the programming language. And yet it feels like something is missing.

The more I access this knowledge, the more I feel the little box on my table is starting to open. Now this is just my opinion, but I'm starting to believe that the sum of all parts is still just one. Let me explain.

In 2022, the Japanese Prime Minister Shinzo Abe was shot and killed. It came as a shock to me, Japan is not a country known for gun violence. So in December of that year, I decided to learn more about him, about Japan, and about their stance on guns. With the holiday season and the rolling code freeze at work, I spent a good amount of time just reading through Wikipedia, some translated Japanese forums, and some official documents. A whole lot of material. Long story short, I still don't have a definitive answer as to why exactly he was killed, but I came away with a richer understanding of the story and the perspectives of the people around him.

Reading more material is not going to give me a definitive answer, but it helps paint a richer picture of the event. I spent enough time with the subject to appreciate the knowledge I gathered over those weeks.

When you ask ChatGPT why Shinzo Abe was shot, it will give you a satisfying answer. It will be correct, it will include some of the nuance, and will probably ask you if you want to learn more. The answer satisfies your curiosity and you move on... to your next question.

It could be the chat interface. Even though the words on the page clearly ask you "if you want to know more," somehow you are more keen on starting a new subject. And rare are the times we go back and re-read the material we have been provided with.

With the books I've "read" through an LLM by asking multiple questions, I can hardly tell you that I understand them. Yes, I know the gist of it but it doesn't replace the knowledge you build by reading a book at a steady pace. You save a whole bunch of time by using an LLM, but the knowledge is fleeting. Reading original sources is slow, but you get to better immerse yourself in the subject.

It seems like reading through an LLM removes the friction of learning, but in doing so it makes knowledge shallow and disposable. The problem is the way we process information as humans. We don't become experts by learning from summaries. The effort of learning is part of the process.

Those endless questions my children have, there is a snack-like quality to the answers I give them. Because the answers are so easy to get, we treat them like a social media feed. I scroll through and one post is about batteries, the next is about sway bars, and somehow I land on California highways.

Having the world's information at your fingertips is a gift, but knowing the gist of everything is not the same as understanding something deeply. We do not form character by reading the gist of it. Instead, character comes from the hunt for information. The limitation of a manual process forces us to focus, to dwell on a subject, until we truly internalize it.

You can hardly spot a hallucination unless it concerns material that you already have knowledge in. Wait a minute. What's happening here. Ah! I see. The box has crept back open.

---