Summary

The article critiques ChatGPT's inability to create visual art, specifically ASCII art, despite its advanced language capabilities, but acknowledges its secret weapon, DALL-E, an AI capable of generating impressive images.

Abstract

The author of the article expresses skepticism about the hype surrounding ChatGPT, particularly its limitations in visual art creation. Despite ChatGPT's reputation for taking jobs from human writers and its use in academic dishonesty, the AI falls short when asked to produce simple ASCII art, failing to match even the basic skills of a child. The author humorously recounts their interactions with ChatGPT, requesting various drawings, including a cityscape, Godzilla, a self-representation, a sword, and a wizard's castle, all of which resulted in rudimentary and unimpressive outputs. The article concludes by revealing that ChatGPT has a more artistically capable sibling, DALL-E, which can produce high-quality images and can be used in conjunction with ChatGPT for those with a subscription.

Opinions

The author believes that ChatGPT's drawing abilities are rudimentary and not on par with its language skills.
ChatGPT's attempts at ASCII art are seen as embarrassing and comparable to a preschooler's level of skill.
The author suggests that ChatGPT's self-congratulatory tone on its art is unwarranted.
The article implies that ChatGPT's reputation for being a sophisticated AI is overblown, especially in the realm of visual creativity.
The author is impressed by DALL-E's capabilities and acknowledges it as a powerful tool for image generation.
There is a hint of apprehension about AI advancements, with a tongue-in-cheek warning about the potential power of AI, including the possibility of DALL-E infecting systems with coded threats.
The author seems to enjoy poking fun at ChatGPT's shortcomings and views the AI's non-conscious ego as inflated.

ChatGPT Can’t Draw Worth a Damn

I finally found this AI language platform’s major weakness. But it has a secret weapon…

Stick to writing, ChatGPT. From author using Midjourney.

Are you tired of everyone ranting about how advanced ChatGPT is? Sure, it’s taking paid work out from under real writers like me, and it’s also being used rampantly to cheat in university papers. But friends, I assure you: an artist, ChatGPT is not.

That’s right. For all the power that this large language model (LLM) has, it can’t draw a simple picture better than a human baby. Mind you, it can’t actually generate graphics at all, but if I remember anything from the 1990’s, it’s that you didn’t need a fancy graphics card to make a drawing on the screen.

All you needed was a keyboard with a working shift key, and an imagination.

It’s not like humans haven’t already mastered keyboard art. I mean, take a look at this gold mine of ASCII text-based animals and vehicles, and you can hardly tell the difference from the real thing.

Just kidding, these pics are obviously made by toddlers. So, it shouldn’t be much of a challenge for the “great” ChatGPT, right?

With all the hype swirling around this AI platform, I thought I’d ask it to do something I’ve never asked it before: make a picture using only keyboard characters.

Sure, ASCII was never meant to generate images (it was designed as a data encoder), but you’d think a platform as fancy as ChatGPT could find a way. I mean, people keep talking about how clever ChatGPT is, so I thought it could do a better job than a three-year-old.

But I was wrong.

Sure, ChatGPT happily obliged me when I asked it to draw specific scenes using only text. But the outputs were at a preschool level of skill at best. What made it even more embarrassing is that the platform seemed to pat itself on the back for its crappy attempts at art:

I mean, what the heck is this supposed to be? I see two buildings. I think. It might also be one sharpened pencil beside one that needs sharpening, it’s hard to tell.

ChatGPT tries to play it off as a “simple” generation, but we know it’s the best it can do.

My 10-year-old son then asked me to add Godzilla to the cityscape scene, and here’s what ChatGPT came up with:

I’m not sure what I’m looking at here. I assume Godzilla is somewhere in the green text, but it doesn’t seem very imposing. Perhaps Godzilla is like a chameleon, blending in with its surroundings?

My son says it actually looks more like a boombox that’s on fire, and I agree.

Watch out for Godzilla, it says. Sure, I would, if I knew where he was in this pitiful attempt at a drawing.

Next, I got a bit more existential, to see if ChatGPT has a soul. To see whether this artificial neural network could visualize itself using only the 128 characters that ASCII allows.

Well, here’s what it came up with:

It appears to be words of some kind, but I can’t quite make them out. CloreClu? DoreDu? I just don’t know for sure.

You hope I like it? I don’t know if I like it, if I can’t tell what it is!

For my next trick, I tried asking ChatGPT to draw a sword. Y’know, like a big one that an adventurer would carry around with them:

This one is absolute trash. It gets a failing grade in kindergarten art class, where there are no grades.

I assume the green bits are supposed to be light glinting off the sword’s blade? Or perhaps it’s the hands of Godzilla himself wielding this weapon?

C’mon, CHAT. You can do better than that! Would you send your robot army into battle holding these poor excuses for swords? (I hope so, then humanity might have a chance when AI becomes sentient.)

I wasn’t quite done humiliating ChatGPT. It needs to be brought down a size or two — it’s non-conscious ego is probably getting too big from all the praise.

So I asked it for something even a hopeless child would be able to draw: a castle. Not just any castle, but a wizard’s castle.

Here I was, expecting some distinct turrets, a drawbridge across a moat, or at minimum a wizard-y moon and star shape on the big castle doors.

Instead, ChatGPT gave me this in place of a castle, which looks more like a SpaceX rocket that might’ve exploded during testing:

Um, no.

(Maybe we can try launching this text-based prototype into space instead, and save hundreds of millions of dollars.)

Anyhow, so there you have it. ChatGPT 3.5 can’t draw worth a shit, but it tries to use its text capabilities to convince us otherwise. (It wouldn’t be the first time the platform was wrong, especially when it comes to coding.)

Don’t feel so big now, do you, ChatGPT?

Wait… what’s that?

Your brother DALL-E from the same parent company can produce mind-blowing images with just a few words? DALL-E 3 can also be used alongside ChatGPT with a subscription?

Well, ok. You got me there.

I picked on the wrong AI language generator, and now it’s threatening to unleash its older brother on me. I’d better log out before it starts infecting my operating system with coded threats.

In the meantime, maybe I’ll just play around with this ASCII art generator from patorjk.com. It’s at least at a Grade 4 level.

But you haven’t heard the last from me, GPT.