Alexandra Woolner
Alex Woolner asked ChatGPT to write crochet patterns and posted the results online.
The chatbot generated some illogical patterns but many viewers thought the results were cute.
Some people saw the AI patterns as an example of the new tech’s limits.
The latest viral experiment with OpenAI’s ChatGPT may be good news for artists worried about AI taking their jobs.
Alex Woolner, one of the cofounders of the arts-advocacy group Attack Bear Press, decided to put ChatGPT’s crafting skills to the test. She had been making crochet animals for her nieces and nephews when she heard about the viral chatbot.
She thought it might be helpful for generating crochet patterns, which are guides that lay out what knots a crocheter needs to make in what order to create something.
Woolner asked the bot: “Write me a crochet pattern for a narwhal stuffed animal using worsted weight yarn,” referring to a thick yarn commonly used for crochet. From a first look at the bot’s pattern, she knew the results when she made it were going to be interesting.
“I thought, ‘This is going to look really messed up or a bit weird,'” Woolner told Insider. “I knew that it wasn’t going to be anatomically correct for what a real-life narwhal looks like because of the placement of the tusk and the fact that it told me to make only one fin.”
She followed the pattern, filmed the process, and uploaded it to TikTok.
Gerald — the viral crochet narwhal
The first animal Woolner made from the AI-generated pattern became an internet celebrity. Fans affectionately named it “Gerald.”
Gerald, a crochet narwhal made from an AI-generated pattern.
Alexandra Woolner
A real narwhal.
iStock / Getty Images Plus
“I know everyone always says I wasn’t expecting my video to go viral, but I genuinely thought this is such a niche thing,” Woolner said. “I think overnight it got something like 14,000 views.”
Woolner’s first video now has 1.1 million views and thousands of comments. “I would buy this in a heartbeat, good lord. It’s adorable,” one TikToker commented.
She’s made other critters from ChatGPT patterns too, including a cat.
A crochet ‘cat’.
Alexandra Woolner
Woolner said she’s been selling Geralds, but in limited numbers as they take time to make. “The first time I listed 10, and within two days they were sold out,” she said.
Woolner also sells the ChatGPT patterns, along with her own edits and instructions, for $2.
Viewers seized on it as an example of AI failings
While some viewers thought the narwhal was strange but cute, others saw it as a sign of AI’s limits.
“I think a lot of people were concerned about AI taking their jobs or AI work becoming impossible to differentiate with human-made creations,” Woolner said. “So, a lot of people commented with ‘this is clearly not going to steal my job’ in a very sarcastic manner.”
A crochet newt made from an AI-generated pattern.
Alexandra Woolner
“There are some things AI cannot steal,” one person commented on her video of Gerald.
Some creatives have taken issue with AI art generators which can copy the styles of specific artists in seconds, saying they could make them lose income. Art made by AI has even snagged first place in an art competition.
Not getting any better
Woolner said she often had to go back to the chatbot when it gave impossible instructions. “For example, it will ask me to make an essentially a flat piece of crochet and then it will tell me to stuff it — a human would understand you can’t stuff a flat plane.”
Woolner keeps making crochet patterns, but said the chatbot doesn’t seem to be getting any closer to making “results I would expect a human to produce.”
“There is no sense of visual space. There’s no sense of what this will look like in three dimensions, it’s not good at math.”
Norma the normal fish and Gerald the narwhal.
Alexandra Woolner
Other AI programs have faced similar issues. Some AI-image generators have struggled to replicate the complexity of human hands, often adding extra fingers or blurring them all together.
Woolner said for now she’s just enjoying the bot’s output and has no plans to help it learn and improve.
“My goal is to keep giving it relatively specific questions so that it will actually produce a pattern, but keep it loose enough so that the AI generates what it generates,” she said.