AI Tool Translates into Slang, Gibberish

Kagi Translate, a tool from the company known for its search product, has recently demonstrated surprising capabilities beyond standard language translation. Users have discovered that the AI-powered service can "translate" text into unconventional "languages," including internet slang like "Reddit Speak" and professional jargon such as "McKinsey consultant speak."

This unexpected functionality came to light when users experimented with URL parameters and direct input fields, revealing the AI's ability to generate outputs mimicking distinct speaking styles. Examples range from imitating famous personalities to generating humorous or critical commentary, such as "LinkedIn Speak" or "rude man with a Boston accent."

While these discoveries highlight the playful and creative potential of large language models, they also underscore the risks associated with generalized AI tools. The open-ended nature of Kagi Translate's input allows for imitation of various linguistic patterns, sparking widespread online amusement and experimentation. However, the ease with which users can prompt for potentially problematic outputs, like "someone who keeps saying slurs," indicates a need for input sanitization to prevent misuse. This feature, launched in 2024 and evolving since, brings back a sense of whimsy to AI, reminiscent of early LLM explorations.

The company has acknowledged these "quirks" and is working to resolve them, but the incident serves as a reminder of the fine line between creative AI application and the need for responsible deployment. Unlike more critical AI applications, the potential harm from Kagi Translate's "language" play is considered minimal, positioning it as a fun, albeit potentially risky, digital toy.