To complete or not complete, that is the...
That’s how this post started out. Here’s how it gets completed.
The problem is completing what I start.
I’ve got 138 unpublished (and therefore incomplete) drafts right now. This is maybe number 138. Or possibly 139. I could find out by starting another one. But I won’t. Because then I’d make the problem worse. I’d have 139 or possibly 140 incomplete drafts.
Among the most popular kinds of Large Language Models (LLMs) are GPTs, which is the kind of thing that ChatGPT is GPT means “Generative Pretrained Transformers.” What’s a Generative Pretrained Transformer? Let’s ask one.
Blah blah blah. It goes on.
How GPTs work
GPTs are trained to generate text. (See point 1 above)
They generate text one word at a time based on the previous words and the patterns of the “vast corpus of text data.” (See point 2)
It seems like they are “writing articles, answering questions, and creating dialog and more” (See point 1), but really, all they are doing is responding to what you type by guessing what the next word would likely have been in the “vast corpus of text data” if what you typed had been there.
If you type something that ends like “give me liberty or give me…” as the prompt, it guesses “death” as the next word. I tried this. It does.
If you ask it to guess what the next words would be if William Shakespeare were writing, it will guess the next words that kinda sound like Shakespeare. If you asked it to guess what the next words would be if Donald Trump were writing—or tweeting, I guess—you’d get different next words that kind of sound like Trump.
Really.
Me
What would be the likely next words if William Shakespeare was prompted with "To be..." (Just give the likely next words, up to 20. Don't explain why)
ChatGPT
"...or not to be, that is the question."
Me
What would be the likely next words if Donald Trump was prompted with "To be..." (Just give the likely next words, up to 20. Don't explain why)
ChatGPT
"...the best, believe me, nobody is better."
Me
What would be the likely next words if Mike Wolf was prompted with “To be…”
(Give the likely next words, up to 20, and explain why?)MikeGPT (because ChatGPT has not been trained on my life)
“… or not to be that is the gggzornonplatt?”
”That is because the phrase ‘To be’ prompts Mike Wolf to complete with the words of an old Bob Newhart routine that Mike Wolf can not quite remember, so he Googles for “Bob Newhart routine about an infinite number of monkeys” which leads to this Wikipedia article which gives him the complete quote.
I am an LLM
Me
What would be the likely next words if Mike Wolf was prompted with “To complete or not complete, that is the...”
(Give the likely next words)MikeGPT
That’s how this post started out. Here’s how it gets completed.
The problem is completing what I start.
I’ve got 138 unpublished (and therefore incomplete) drafts right now. This is maybe number 138. Or possibly 139. I could find out by starting another one. But I won’t. Because then I’d make the problem worse. I’d have 139 or possibly 140 incomplete drafts.
Mike Wolf then wrote this blog post as a completion, one word at a time, and continued the completion to here, and will continue to continue the completion.
And then breaking out of the self-referential loop he continued:
But will he ever complete the completion?
What does it mean to complete a completion
Me
What does it mean to complete a completion?
MeGPT
It means you stop writing and post the fucking thing.
Me
Now?
MeGPT
Now! Ask ChatGPT for a fucking image and post the fucking thing.
Me
Done
Wonderful. And you got it done. Amazing! Thanks, Mike GPT and Mike for real.