If you’ve spent any time online, you’ve likely used a service like Google Translate to turn webpages or bits of text between languages from Uzbek to Esperanto. But what if you wanted to convert text into more obscure “languages” such as “LinkedIn Speak,” “Gen Z slang,” or “horny Margaret Thatcher”?
This week, lots of people online were amused to discover that the AI-driven Kagi Translate can carry out these and many other unlikely “translation” feats. While the shared discovery highlights the playful, creative side of large language models, it also reveals the hazards of giving users free rein with generalized LLM tools.
What exactly counts as a “language”?
You may know Kagi as the paid rival to Google’s deteriorating search product, but the company rolled out Kagi Translate in 2024, pitching it as a “simply better” alternative to services like Google Translate and DeepL. At launch, Kagi said Kagi Translate “uses a combination of LLMs, selecting and optimizing the best output for each task,” a setup that “can occasionally lead to quirks that we’re actively working to resolve.”
The initial versions of the tool offered simple dropdowns to pick from 244 languages for the source and target. In February 2025, though, at least one little-noticed Hacker News poster found that by tinkering with the URL parameters you could set the target language to “rude man with a Boston accent” without anything breaking.
In recent weeks, Kagi’s social accounts have shown off the service’s ability to mimic “Reddit Speak” or produce McKinsey consultant jargon with a couple of clicks in Kagi Translate. Early Tuesday, however, those offbeat use cases escaped limited attention after a Hacker News user gleefully reported that “Kagi Translate now supports LinkedIn Speak as an output language.” Further down that busy HN thread, other commenters pointed out that you can change the output language merely by typing into Kagi Translate’s search box, and the underlying AI will try to accommodate you.
