We never knew this until LLMs came along

People often ask me, what is it that's going to be left for them after LLMs take over virtually any output that they desire? My answer is that your output was always there in part to serve as an indicator of what's inside you. There's no such thing as vibe understanding. You can vibe code, yes, you can even vibe art, but you can't vibe your way to understanding. You have to work at your understanding. Vibes can get you through Coachella, but not Kierkegaard.

Now, of course, LLMs can aid your understanding in the same way that literacy aids, or that the internet aids understanding, or tutors aid understanding, but LLMs can't convert output tokens to understanding-for-you tokens. Many of the problems that you wrestle with, say the ones about who are you, what happens after you die, what happens after your child dies, what is time, what is rightness, morality, why is it all meaningless and torturous simultaneously, etc. Themes that are on this channel.

Now, none of these are something that you outsource your decision about. Neither is it something that you want to adopt because of consensus. Otherwise you just join some local flavor of religious institution and pick up your beliefs like you pick up an accent. Getting an LLM to write for you would be the equivalent of getting a machine to lift weights for you. Machines can aid you, yes, but ultimately, the pump needs to be primed with pain.

You may challenge and say, "hey, Curt, well, if the LLMs come to consensus, then I trust the LLMs. They have some rationality behind them." Well, that's a large assumption, but let's go with it. Even granting this, you would still want to convince yourself that the rationality that the LLMs possess is of the sort that you deem relevant and worthy of a sound conclusion for these metaphysical issues. That act itself would require understanding tokens. And those understanding tokens are something that no current LLM can do for you. Potentially, no future text producer can ever do that for you. You can't vibe understand. You have to achingly understand. Grok can't grok for you.

Now an issue that doesn't get discussed as much is that when you produce text via an LLM you may mistake the LLM's understanding for your own. This is because no time in history has ever been like this. The format that we use to judge other people's understanding is now something that can be outputted with minimal input from us. Fluent natural language has comprehension just baked into its structure. And this is unprecedented. Shallowness now wears the robe of eloquence. A calculator never tempted anyone to think that the calculator understands arithmetic. At least, nowhere near to the same degree that LLMs tempt people into thinking that they themselves understand what the LLM is outputting. The output was always a proxy. We never knew this until LLMs came along.

Many tests are like this, though. So, for instance, we thought that the Turing test was a test of consciousness or minimally machine thinking. Once it was achieved, though, we then moved that alleged goalpost. However, it's not a fallacy in this case, like most people think. It's not a fallacy because we realized how wretched the measurement was. And unfortunately, much of the time the only way to realize the wretchedness of a measurement is to reach the goalpost. It measures something up until a point. Only then do we realize that what we thought was God was an idol.

Similarly, we used to think that the beautiful prose someone handed to us was entirely indicative of what was in their mind. It wasn't until LLMs came about such that anyone could produce said beautiful prose that we realized that it was only a useful measure up until a point. We used to think painting was about replicating reality until photography came along. We used to think that memorization carried wisdom until the printing press came along. There are so many times in history that we realized we were doing some task, not for reason X, but for reason Y, and this has happened, and this will happen over and over. The more something fulfills our need, the more that that thing reveals our needs as wants. It's only actually then that we realize what we're doing this all for.

So write. Not for the beautiful prose. It likely won't be beautiful. Write for the tedious and tormenting production of understanding. That's something only you can do. By the way, someone asked me about my writing process in the comments. Here are two relevant links from my Substack. This one's on my three rules for writing, and this one's on how I build habits. If you like this video, then you may like this other video on Kierkegaard or Gödel's Incompleteness Theorem. Both are linked on screen and in the description.