Chatbot, draw!
I'm interested in cases where it's obvious that chatbots are bluffing. For example, when Bard claims its ASCII unicorn art has clearly visible horn and legs but it looks like this:
or when ChatGPT claims its ASCII art says "Lies" when it clearly says "Sip"
or when ChatGPT claims its ASCII art says "Lies" when it clearly says "Sip"
They also vastly overrate their own code.
Large language models tend to have lots of code in their training data, so one kind of code they can write is SVG, which can be used to draw simple images. Their SVG code mostly works, but the pictures it draws are... not all the chatbots advertise them to be.
I asked ChatGPT to recreate some emoji. I was prepared to be lenient, keeping in mind of course that the pictures themselves will vary between Android, Apple, etc, and that some of them are too complex to be closely recreated.
Here's what I got. In case you can't tell the difference, the ChatGPT version is on the right:
Here's what ChatGPT had to say about its recreation of "tears of joy":
"The first two circles create the yellow face of the emoji, while the path elements create the tears. The tears are created using cubic Bezier curves with control points positioned to give the appearance of a teardrop shape. The last two path elements create the white highlights on the eyes."
Here are ChatGPT's attempts at "a pretty magical pony".
For the one on the lower left, ChatGPT says "This code creates a simple but cute magical pony with a blue body, white mane and tail, and a golden horn."
ChatGPT also generated some giraffes for me.
I had Bard try to generate some SVG art but although its code generally worked, it usually neglected to use colors, resulting in images that were mostly dark blobs. Its giraffes were kind of cool though:
Mostly Bard's drawings were inscrutable. It would add labels when requested but they raised as many questions as they answered.
This large language model thing is going well.