Kagi Translate's AI answers the question "What would horny Margaret Thatcher say?"

2026-03-18 · Source: AI - Ars Technica · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Fundamental Awareness, short

Summary

Kagi Translate, an AI-powered translation tool launched in 2024, has gained widespread attention for its unexpected ability to "translate" text into esoteric styles like "LinkedIn Speak," "Gen Z slang," or even the voice of "horny Margaret Thatcher." While initially offering 244 standard languages, users discovered in February 2025 that modifying URL parameters allowed for custom target "languages." This functionality, which Kagi's social media has since embraced, enables the underlying Large Language Model (LLM) to synthesize language patterns in creative and often humorous ways, reminiscent of early ChatGPT explorations. This phenomenon highlights both the playful potential of LLMs and the inherent risks of user interaction with generalized AI tools, as demonstrated by the potential for generating undesirable outputs when prompted with harmful "translation" styles.

Key takeaway

For AI Product Managers evaluating user-facing LLM applications, you should recognize that generalized AI tools, even those designed for specific functions like translation, can be repurposed by users in unexpected ways. While this can foster creativity and engagement, it necessitates robust input sanitization and content moderation to prevent the generation of undesirable or harmful outputs, ensuring brand safety and responsible AI deployment.

Key insights

Kagi Translate reveals LLMs' creative potential for stylistic "translation" but also exposes risks of generalized AI tools.

Principles

• LLMs can synthesize diverse language patterns.
• User interaction with generalized AI carries inherent risks.

Method

Kagi Translate uses a combination of LLMs, selecting and optimizing output for each task, allowing for flexible, user-defined stylistic "translations" via URL parameters or direct input.

In practice

• Experiment with LLMs for creative text generation.
• Implement guardrails for user-defined AI inputs.

Topics

• Large Language Models
• AI Translation
• Generative AI
• AI Safety
• Kagi Translate

Best for: AI Product Manager, Tech Journalist, AI Ethicist, General Interest

Related on AIssential

• I Built a Tamil Crossword Game for My Mom (And Had to Fight AI to Make It Work) — Adversarial prompting and strict schema definition are crucial for overcoming AI limitations in complex, non-Latin language tasks.
• Get more context and understand translations more deeply with new AI-powered updates in Translate. — Google Translate now offers AI-powered contextual alternatives for nuanced, idiomatic translations.
• What DeepL Got Right That No LLM Can Copy — Prioritizing data and speed over general LLMs enables superior real-time user experiences.
• Better Literary Translation: A Multi-Aspect Data Generation and LLM Training Approach — Multi-aspect data generation and iterative refinement significantly enhance LLM literary translation.
• Bridging Cultural Gaps in Automated Translation of Brazilian Expressions: A Study on Cultural Adaptation — Automated translation systems struggle with cultural adaptation, even with advanced LLMs, requiring human oversight.
• A Dataset for Probing Translationese Preferences in English-to-Swedish Translation — Language models often prefer literal "translationese" over idiomatic phrasing, especially when exposed to the source language.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI - Ars Technica.