Is this one of those times you discard "scientific" tests and trust your instinct instead? OpenAI is doing it well using curated context files [1]. Maybe these

Is this one of those times you discard "scientific" tests and trust your instinct instead? OpenAI is doing it well using curated context files [1]. Maybe these "scientific" explanations are not useful or misleading. [1] https://openai.com/index/harness-engineering/ https://openai.com/index/harness-engineering/