All models hallucinate more on obscure topics, recent events, and specific numerical data - All models produce better output with explicit format specification than without - All models improve significantly with context — the "blank slate" problem affects all platforms equally