Kagi Translate now supports LinkedIn Speak as an output language

Original Article ↗ Hacker News Discussion ↗

What Kagi Translate Is Doing

  • Kagi Translate now offers “LinkedIn Speak” as an output language and appears to use an LLM behind the scenes.
  • Users discover that the from/to language fields are effectively free text: you can type any “language” label and the model will try to mimic that style.
  • Some “fun languages” like Reddit, Hacker News, Trump, Kamala Harris, Metallica lyrics, etc., are surfaced in the UI; others work only if manually typed.

Playing with Arbitrary “Languages”

  • People successfully generate text in styles such as “Hacker News speak”, “angry guy”, “Karen speak”, “Metallica lyrics”, “Jim Cramer speak”, “Gen-Z slang”, “unhinged Trump rant”, “unhinged Elon Musk rant”, “poopoo peepee”, “a dumb guy”, and many more.
  • The model not only mimics tone but often thematically adapts structure (e.g., repetitive input like “ass” thousands of times becomes a hustle-culture post about scaling output).

LinkedIn Speak Reactions & Uses

  • Many find the LinkedIn-style output uncannily accurate, triggering strong recognition of real LinkedIn “broetry” and corporate jargon.
  • Users test it with trivial, vulgar, or bodily-function inputs; it consistently converts them into polished, hashtag-heavy corporate motivational posts.
  • Others feed in famous texts (Gettysburg Address, Lord’s Prayer, Shakespeare, Declaration of Independence, Kennedy/Berlin, Churchill, national anthems, Bible/Genesis, Nietzsche, rap lyrics, memes) and enjoy the bathos of seeing them flattened into LinkedIn jargon.
  • Some see a serious use: softening blunt or harsh feedback into more palatable corporate language, or auto-generating LinkedIn posts, recommendations, or Slack replies.

Reverse Translation & Custom Styles

  • Reverse LinkedIn→English is inconsistent: sometimes works well, sometimes leaves jargon intact; several users call this “bad” or unreliable.
  • Defining custom “target languages” like “Telling you how it is cut the crap” or “answer” can yield fairly direct, de-jargonized paraphrases.
  • Repeated round-tripping (English ↔ LinkedIn) tends to drift into absurdity, illustrating how the style exaggerates and obscures meaning.

Model, Prompt, and Safety Concerns

  • Commenters infer it’s “just” an LLM wrapper with a system prompt tailored for translation; one user even coaxes out a detailed translator-style system prompt.
  • There appear to be minimal or no safety filters: users get output with slurs, graphic violence, and politically charged content in whatever style they request.
  • Some express concern about this lack of guardrails; others treat it as part of the joke.

Kagi Product & UX Notes

  • Brief side discussion on Kagi Search: some say results are no better than free engines; others argue it’s noticeably better than Google/DDG, and customizable (blocking/upranking sites, AI-slop detection).
  • A few report technical issues (Cloudflare captcha, Firefox quirks, 429 rate limits).
  • Multiple people want browser extensions to auto-translate LinkedIn posts into plain (or salty) English.