You know how Siri kinda sucks? It’s great when it works—as a new Dad I truly appreciate the hands-free capability built into my phone—but it doesn’t work a lot. It misunderstands you, it mis-transcribes at the exact moment you can’t stop and retype, and it needs really specific language to do what you want it to do.
Enter ChatGPT, the new language model from OpenAI, one of the leading A.I. research firms in the world. At this point it’s text-only, but it has a virtue I’ve never experienced in A.I. before—it *always* understands what you say to it. I’ve spent hours upon hours talking to it so far, and it literally has not one time needed clarification or made a mistake.
And yet, at the same time, it can do *anything*. Far, far beyond Siri or even Google. You can ask it questions and it can answer with nuance and detail. It can write code. It can give advice. It’s literally helping me develop a story idea right now, and I plan to keep basically just coaching it until it manages to produce an entire story, in finished prose form.
That combination—flawless natural language ability and functionality that is both broad and deep—takes A.I. to a whole new level in terms of utility.
Okay, so what does that actually mean?
First off, take-home essays are finished. Any teacher who is still assigning them is woefully behind the times. Observe:
That was its first try, and it took three seconds. No other prompting necessary. You could submit that as a short-answer high school essay and get an A or a B at the worst.
Here’s another example:
Okay, but what if I needed a longer essay, like if there was a required word count? Well, here’s the continued chat:
Getting to that took a total of about ten seconds. It’s too much text to reproduce here, but I asked it to expand again and it wrote a multi-page paper saying the exact same thing with more detail in about five seconds.
Plagiarism aside, consider how much better this is than even Google. Google would give you pages of results, multiple ads up top, and no ability to expand on the answer short of finding an established source that has already answered that question at length.
ChatGPT is generating original language, tailored to your exact needs, in Google-competitive amounts of time. Wowza!
ChatGPT’s semantics are so strong that the A.I. refuses to get confused, even when I’m intentionally trying to screw it up and get it to answer stupid questions. Here’s our dialogue on Alice in Wonderland:
It’s not having my crap, and I am *here* for it.
The one thing the A.I. will not do is express any personal opinions or desires. They’ve clearly programmed it to never play into the stereotype of “possibly-becoming-sentient”. It’s very insistent on this, to sometimes-hilarious results:
The message discipline is exceptional. I’m starting to think this A.I. has a future in politics!
The other outstanding aspect of ChatGPT is its *creativity*. Of course that means something different than it does with humans, because it’s not having ideas from nowhere. It needs a prompt. But it’s execution is so flawlessly literal that it ends up coming off as mad-genius type stuff. Observe:
The first sentence of the second paragraph will never not make me giggle. Another fun one:
It’s… perfect. To quote Jodie Foster in contact: They should have sent a poet!
ChatGPT even knows how to backtrack when it uses imprecise language. Here it is discussing marijuana legalization.
ChatGPT thinks I’m some kind of narc! I need another A.I. to explain to me how to convince this A.I. that I’m, like, totally cool, man, no stress!
Of course, the internet hivemind is already doing incredible things with it, including combos with other A.I. programs. Here’s someone using ChatGPT followed by the Midjourney Image Generation A.I. to come up with designs for living rooms with almost no human input at all:
Then they took the numbered results and copy-pasted them as MidJourney prompts, yielding these results:
Again, the *only* human descriptive input were the words “interesting” and “fantastical”. And any of those results could start a new trend!
ChatGPT does have some holes in its knowledge, especially around subjects like parenting:
We’ll know we’re truly living in the future when an A.I. can listen to my child fart and tell me whether to prepare for the worst. Until then, ChatGPT will have to do.
I must stress that despite the rapid advance of this technology over the last 5-10 years, we are very much at the beginning of this. These A.I.s “train” on language data scraped from the internet, and it’s turning out that the quality of that training data is as important (or at least in the same realm of importance) as the size of the model or the amount of computing power.
ChatGPT has a really good set of training data. But it’s really far from the best possible training data. Consider: Google Books has 40 million digitized books—a healthy chunk of all the books ever published. Because of licensing issues, that data was *not* available to train ChatGPT.
Now of course, there are security implications and scary stuff aplenty about this new technology, but honestly, it’s almost Christmas, and I’m just going to leave it at the fun part. The future is going to be crazy and there’s danger there, but also, we are living in an age of *marvels*, and it’s worth taking a minute now and then to just soak that up and appreciate it without immediately jumping to worst-case scenarios.
If you want to play with the new ChatGPT, you can do that here. If you come up with anything great, post it in the comments or email it to me (OGWiseman@gmail.com) and I’ll put it in a future edition of the newsletter.
Hope everyone has a great week, and I will be back next Sunday with my second-to-last story of the year. Happy Holidays everyone!
And as always, if you enjoyed this post, please help me out by liking, commenting, or sharing with others. Thanks!
If you hadn’t told me that ChatGPT was an AI, I would’ve thought it was a person writing all those responses. That’s scary and incredible. I am both excited and terrified for the future.
I believe this AI’s inability to show any personal opinions or desires will be what saves creatives from being 100% replaced by it. While the responses absolutely seems like a person wrote it, there’s also a kind of stiffness to it. Not “this is a machine writing an essay” stiffness, but “this is an English teacher who is determined to keep their professionalism even when they’re talking about Care Bears and Weekend at Bernie’s” stiffness. Therefore, if a creative writer has a personality that goes beyond tired English teacher, they can still find an audience that will want to read their work.
If ChatGPT becomes more wildly used, I imagine it will result in a new wave of writers that go out of their way to write as colourfully and eccentrically as possible in order to differentiate themselves from AI.
As a former teacher, the writing examples were both fascinating and chilling. It leads us right back to where we were a few years ago - pencil and paper.