AI made up these words. The bobweiser can vouch for their quality
Instagram's former engineering director trains an algorithm to invent vocabulary. Sometimes, nonsense can make perfect sense.
Leslie KatzFormer Culture Editor
Leslie Katz led a team that explored the intersection of tech and culture, plus all manner of awe-inspiring science, from space to AI and archaeology. When she's not smithing words, she's probably playing online word games, tending to her garden or referring to herself in the third person.
Third place film critic, 2021 LA Press Club National Arts & Entertainment Journalism Awards
Sometimes, I feel so damn happy I just want to break out into a skunkie. Oh, don't know what a skunkie is? It's a dance combining thrash and rock music, usually with a marching band.
I've been doing the skunkie forever, but I just learned what it's officially called, thanks to ThisWordDoesNotExist.com. The amusing site, created by Instagram's former engineering director, collects nonsense English words generated by AI, along with strangely convincing dictionary-style definitions.
The site has greatly expanded my vocabulary. I now know, for example, that a heapslepper is a "small piece of iron, glass or similar material to be blasted out of something." A triuplex is a "climbing dog that resembles a quadruped with long black hair and two-tipped toes" and a creature I'm pretty sure I saw in my nightmare last night. A bobweiser is a person "whose job it is to check the quality of something," and thankfully for job seekers, doesn't need to be named Bob. All definitions are followed by a second one: "a word that does not exist; it was invented defined and used by a machine learning algorithm."
Software engineerThomas Dimson created the site with the help of GPT-2, a neural net that serves up predictive text culled from the internet. "GPT-2 is trained with a simple objective: predict the next word, given all of the previous words within some text," explains the site for the tool by OpenAI, a research lab based in San Francisco. GPT-2 was trained on a dataset of 8 million web pages.
Dimson came up with the idea for ThisWordDoesNotExist.com after leaving his job at Instagram earlier this year and starting to ponder potential names for a new AI company. He decided to train an algorithm to help with with the task, and "while building the algorithm, I stumbled upon using the dictionary as training data and discovered I could make arbitrary words," he tells me.
I typed in "qwertyuiop," the first line of letters on the keyboard, and learned that it's "a high-pitched sound originating in Tonga, used chiefly for talking to children." And that lockdownfever is a tropical North American plant with brightly colored flowers. If only.