Ryan Doolittle Email & Phone Number

Portland, OR, US

Train AI models by reviewing and correcting API usage from a wide tool library including YouTube, Google (maps, flights, hotels, search), and Tesla. Rate up to four side-by-side responses to each prompt and write a detailed justification outlining the faults of each response (tool selection, correctness of parameters, hallucinations) and arranging them in.

Trained AI models by solving advanced and nuanced math prompts that AI models regularly fail to solve. Such prompts were multi-step, high complexity problems involving topics such as advanced geometry, multivariable calculus, and linear algebra. Precision was a must, as any inaccurate submission would incorrectly train the model, so attempters would be.

After receiving consistently high ratings as an attempter (4.5/5 - 5/5 from 16 reviews), moved to Reviewer position within the Flamingo Code project. Reviewed and rated submissions made by attempters based on prompt complexity, adequacy of their analysis of the deviations, and precision of their "ideal" response. If any portion of the attempters submission.

Remotely trained AI models by creating programming related prompts specifically designed to cause deviations in model responses. Rated side-by-side responses given by the model, highlighting precise points of success and failure, before creating the ideal response to the given prompt.