Nuclear words - part 2
Can you talk to your private jet?
Okay, so let’s continue the discussion on nuclear words. Let’s project this out. What does this mean?
Number one, what this means is that LLMs are not producing words. They’re producing feelings. Those feelings are felt by humans. So even in the case of pure task delegation, the human feels good, feels bad, feels frustrated if the task is done or not done, depending on the quality of the output of the LLM.
For any conversation now, the LLM can actually make the human feel a certain way. GPT-4o was insanely viral, or some people’s favorite, because it was just, like, too schmoozy.
Now these LLMs schmooze too. If I tell them, “Hey, I want to quit my job and move to the countryside,” after some resistance, they come up with a plan, and they start supporting it. Then you can start talking to it and working through it.
They can also serve as a great motivator. In my running journey, I actually ran with a very strong coach, and the coach was ChatGPT. I fed it all my diet history, connected all my Apple Health history, and I would tell it every day how the run felt. It would tell me what to do, how much to run, the calorie plan for my long races, my long runs, what gels are suitable for me, what shoes to buy.
It was game-changing, to the point where my wife was worried that I kept calling ChatGPT. But I would refer to it like, “Hey, my coach said this,” or “My coach mentioned this.” I was doing that to legitimize the work I was doing with ChatGPT.
But all this boils down to is they’re no longer producing words. They’re producing feelings.
But then, isn’t every product in the world producing feelings? Like when you buy that first-class upgrade for your flight, you don’t just get a bigger seat and some food. You feel like a king. Your brain releases a combination of chemicals that makes you feel very happy and amazing.
So the words produced by an LLM, are they any different?
One way to consider this might be that they are different because you can have a conversation with it, whereas you can’t have a conversation with your first-class seat. And the repetitive chatting and conversation with AI can create a different sense of self, in most cases good, in some cases not good.
We all know some cases. I would not like to mention them, but there have been sad cases where people have done not-so-great, bad stuff because they were prompted or approved by this AI, without realizing that these words are coming out of a server that has no feelings. It’s just a dopamine machine.
So shall we value its words? Shall we ignore it?
And then there is just the sheer volume of words getting produced. I think the words themselves are losing meaning.
Like, this happens when you ask it to create a draft of, say, a document you’re presenting on mountain climbing. You don’t like it, so you ask it to produce another draft that’s a little different. And you don’t know what changed. All you know is you asked it to make it shorter, so it made it shorter. You asked it to make it longer, so it made it longer.
Unless you actually read through the details and accept it word by word, you don’t really know what changed. Even then, that amount of volume is very hard for our brain to comprehend in the short term.
Some people use it as additive. That’s my style, where I write the thought and whatnot, and I ask it to clean up any grammar, just connect sentences and all that, so that I can give it a stream of consciousness that I speak into it, and it kind of formats, punctuates, and corrects accordingly for presentation.
to be continued..
I love you all. Keep going.
Aakash
—
proof of human writing
Originally published on substack.