Kagi Translate, the AI-powered translation tool from search company Kagi, is going viral for translating plain English into registers like 'LinkedIn Speak,' 'Gen Z slang,' and 'horny Margaret Thatcher.' The tool launched in 2024 as a direct competitor to Google Translate and DeepL, built on a combination of LLMs selected and optimized per task.
The collective discovery exposes a real tension: generalized LLMs do not draw hard lines between translating Uzbek and performing open-ended stylistic transformation. Kagi acknowledged at launch that this architecture 'can occasionally lead to quirks.' What users are finding now is not a bug report. It is a demonstration of how wide the capability floor actually is.
The full article is worth reading for what it does not answer cleanly: where the line is between a translation tool and an unconstrained text generation interface, and whether Kagi intends to draw one. That question has product, safety, and regulatory implications that go well beyond Margaret Thatcher.
[READ ORIGINAL →]