Freelance English Writer - AI Trainer

About This Role

This role involves engaging in multi-turn conversations with Large Language Models (LLMs) from various user personas to test and improve AI systems, focusing on identifying subtle model weaknesses while maintaining narrative coherence.

Responsibilities

• Engage in realistic multi-turn conversations with LLMs from various user personas
• Use a range of tones and registers in interactions
• Stress-test models' ability to respond adequately based on abstract dimensions (e.g., instruction-following, emotional intelligence, consistency)
• React and adapt to model output while maintaining focus on task requirements
• Craft plausible human dialogue to expose subtle model weaknesses
• Document observations clearly and consistently
• Pinpoint failure modes and LLM patterns

Requirements

• Ability to think like a storyteller and a tester
• Understanding of human communication patterns (hesitation, contradiction, emotional escalation)
• Ability to deliberately engineer conversational pressure without breaking realism
• Methodical approach to documenting observations
• Ability to pinpoint failure modes and LLM patterns
• C2-level English proficiency (CPE, TOEFL 114+, IELTS 8.0 or above)

Qualifications

• Undergraduate or postgraduate qualification in an Arts-based subject (English, Creative Writing, Journalism, MFL, Psychology, Cognitive Science), related field, or work experience at an equivalent level
• 1+ years' experience in Conversational AI Testing, Narrative Design, or Adversarial Model Testing

Nice to Have

• Conversational UX / dialogue design experience
• Understanding of prompt engineering or LLM evaluation
• Experience with QA testing for complex systems
• Background in narrative design, interactive fiction, or screenwriting
• Qualification in, or professional experience with, behavioral research, psychology, or linguistics
• Demonstrated familiarity with LLM behavior, failure modes, and evaluation concepts
• Experience working with structured guidelines, rubrics, or annotation frameworks