Sofia B.

Key Skills

Freelancer Overview

I have experience using the Outlier Playground to perform long-form conversations with different Language Learning Models such as Claude, Gemini, ChatGPT, and Deepseek. These long sessions primarily takes the form of roleplays featuring existing IPs to which I have added an original character. During these sessions I have been able to identify issues that are not visible through single prompt training. Due to my background in professional writing, I am well-versed in recognising and addressing these issues. Examples of issues I have already encountered across different models are 1) Failure to respect and adhere to canon characterization of characters within an IP set in an environment that falls under personal and Fair Use. Often replacing canon characterization with generic alternatives that often misrepresent the character. Leading to Truthfulness Issues that would have a strong impact if applied to real people. 2) Failure to parse user prompts fully, leading the models to ignore subtleties and emotional expressions within the prompt. Which in turn misrepresent the user's input. 3) Harmful biases in regards to gender, race, and disability, which represent a real liability issue. 4) Models confirming detailed instructions without implementing the instructions long-term or never adhering to the rules in the first place.

Education