Two studies reveal Google's Gemini 1.5 Pro and 1.5 Flash don't effectively handle data like 'War and Peace.' Correct answers given by models were only 40% to 50% in document-based tests.
Despite processing large contexts, Gemini models may lack true understanding of content, according to Marzena Karpinska, a postdoc at UMass Amherst. The context window of Gemini models can include up to 2 million tokens.
Collection
[
|
...
]