AI Trainer

Contributed to multiple AI data annotation and evaluation projects across Italian and English, spanning prompt creation, response evaluation, and conversational AI assessment. Core responsibilities included: - Designing complex, multi-constraint prompts across diverse categories (creative, adversarial, audio, category-specific, and web/computer agent tasks) in Italian and English - Evaluating AI-generated responses against rigorous criteria including language proficiency, factual accuracy, instruction-following, and harmlessness - Assessing adversarial prompts and rating the harmfulness of model outputs to support AI safety objectives - Translating prompts and responses between English and Italian - Conducting multi-turn conversations with leading AI models (Gemini, ChatGPT) and evaluating their human-likeness, enjoyment, and personality consistency across multiple axes - Manually guiding AI agents step-by-step through websites and desktop environments to complete structured tasks, then judging the quality and efficiency of automated decisions - Identifying prompts that expose model failures in instruction-following and truthfulness across specific subcategories (numerical reasoning, correction implementation, etc.) - Reviewing peer submissions to ensure evaluation quality and consistency Achievements: Consistently worked respecting performance thresholds, unlocking extra bonus projects and premium bonus rate tiers.