Fun With ChatGPT

AI did not write this subtitle

This one has a ton of pictures, so your email client may not handle it well. You can also read it at the Substack website.

The woke-ification, for lack of a better word, of chatbots using AI is inevitable, but there is still a great deal of fun to be had here in the early days of such applications.

During last night’s slow reveal of Twitter Files, part 2, I bided my time by playing with it and learning a bit about how to get good results from it.

I asked it for descriptions of encounters I’ve actually had, to compare its attempt to reality; creative writing about people I know; questions about how to practice religion; a self-help plan; code snippets; mathematical proofs, and several other things, including spooky commentary on the present by figures from the past.

It only consistently failed in the realm of poetry, which is interesting.

Some of the results are hilariously funny, creepily realistic, or surprisingly deep and thought-provoking. Enjoy!

Failure Mode: Poetry to Order

It theoretically understands rhyming, but in practice, it does not. This suggests that it knows enough to check a rhyming dictionary for words that rhyme, but not to do a reverse check for words that do not rhyme, if I had to guess, which is a very interesting failure mode.

It wrote a haiku on the second try, but other than that, totally failed.

I have three more free subscriptions to Joseph Massey’s excellent poetry Substack to give away. Leave a comment or email me if you want one! Clearly the ChatGPT sucks at poetry, so get the good stuff.

A Story About Dr. Roller Gator

It has a memory of each session, so it had already answered some questions about Gator for me (which is how it knew about his slick leather jacket). Still, I was impressed. Only the implication that the Arby’s food was good is even mildly off-base.

AI Humor Is Pretty Funny

I laughed for ten minutes. Also, I thought it was interesting that the AI went with “Heather and Bret,” when most people say “Bret and Heather.”

A Political Contrast

It’s Very Good at Theory of Mind, Including for Arguments

After some friends failed to get it to say positive things about various personality disorders, I tried having it take on an explicit perspective. That led to having it try other explicit perspectives.

It Can Write Niche Jokes

It Has Some Advice for Practicing Buddhism

We Weren’t Sitting on A Couch, But Not A Bad Approximation

Especially for someone who wasn’t there…..

AI Confirmed As the Ghost of Thomas Jefferson?

Christian Songwriters May Be Out of Work Soon

A Self-Help Plan

It Can Write Decent Code

This worked, and produced the correct answer.

This code also worked, producing the following simple but accurate graph:

But Its Math Proofs Need Some Work

This one is pretty good:

For the uninitiated — the question of whether or not there are infinite prime numbers has been settled for centuries (there are) but the question of whether or not there are infinite twin primes (primes separated by 2, like 3 and 5, 11 and 13, or 17 and 19) is an open one, so this was something of a trick question.

This is mathematical gobbledygook, not surprising given it’s a trick question.

---

ChatGPT reached a million users in less than a week. It’s not hard to see why; the technology is quite fun to play with, and it’s going to be really interesting to watch it develop.

---

Housekeeping: comments are open for paid subscribers. Email hollymathnerd at gmail dot com if you want a paid subscription but cannot afford one.