A Stanford and Berkeley study showed models retrieve answers best at the start or end of context and struggle in the middle—sometimes worse than if they had no context at all.
This week’s #PythonTips breaks down the research (links in the comments) and how to write prompts.