It’s day 5 of the rewrites. I remember writing this one. It was pure stream of consciousness with very little editing involved. It’s funny how when one writes I think and the editor says, “Be bold, don’t hedge.” It’s like, come on, why you gotta make this personal, Hemingway app?
I’m in what I call reverse withdrawal stage. It’s when one is looking for excuses to not continue doing what one is doing. Meaning, while writing this entry, I was thinking about how writing is obsolete now. Algorithms that can generate text like GPT-2 exist and are only getting better. I fed some prompt to the algorithm here. A lot of times it spits out better gibberish than I ever could. For the moment it seems I found the perfect reason to not write.
But doing this led to several other thoughts. Those thoughts then provided an opposite excuse to continue writing. Why did the algorithm need a prompt? Would it be able to write something unprompted? What would it write? Of course it could and it would pick something random from its training data.
My initial reaction was to compare what the algorithm was doing to how people think. Aren’t humans also responding to prompts most of the time? In the physiological sense, at least? Thoughts pop up in random from the synthesis of our training data, i.e. our experience, all the time. And when that happens, it appears as if we come up with something original. Down this path, isn’t GPT-2 also sort of thinking? Or it could be the opposite. We assume we are thinking but like GPT-2 we are not.
But that’s a repulsive thought. GPT-2 writes for the sake of having written something. Humans write as self-reflection. Or do they? Also, how sure am I GPT-2 doesn’t have a self? Or that I do? Pretty soon this whole train of thought devolves into asking what a self is.
At this point, I have to confess I wrote the following yesterday: “One’s writing reveals one’s depth. I know this is self-serving to write, but thoughts are not thoughts unless written.”
I have no intention to even try to answer what a self is or what a thought is. I will be sure to be out of my depth pretty quick. Instead, my question is meta. Did I write all that so I could quote myself? Not in the beginning. The proof is that the first two paragraphs in this entry contradict what I wrote.
But after realizing I would like to use those two sentences, I started to think about how I’m thinking. I noticed that in the context of admitting that, the quote would become relevant. Now I’m writing about how I thought of that, so this entry could land on something.
That’s what separates humans from GPT-2. We don’t limit ourselves to thinking about stuff, we also think about our thoughts. We even think about how we come up with them. This includes instances when we propose writing as a way to think better. When we write our thoughts down, we reach some semblance of clarity. But then is it not possible to transfer all these to an algorithm?
This post has gotten too long. So much for beginning with coming up for an excuse not to write. Enough for today.
In the spirit of amusement, I fed the first two of the last three paragraphs above to GPT-2. It wrote, “At this point, I realize I need to learn about the program I am about to propose.” What’s proposing what now? GTFOOH.