kenzie524/

gdpval-gpt4o-exp

	🧮 GDPval LLM Scaffolding Experiment (GPT-4o + Claude Sonnet)




	
	Overview

This dataset contains model completions for a controlled behavioral experiment conducted by the Data Innovation Lab, UC Berkeley Haas.It explores how assistant scaffolding — structured planning and self-review guidance generated by Claude 3.5 Sonnet — affects the performance of GPT-4o on professional tasks drawn from the GDPval “gold” subset (OpenAI 2024).

GDPval benchmarks… See the full description on the dataset page: https://huggingface.co/datasets/kenzie524/gdpval-gpt4o-exp.