Рет қаралды 771
in this video chris looks looks at the specific book datasets that openai chatgpt is trained on, in comparison to other models such as llama, alpaca, vicuna 13b, stabilitylm, dolly and mosaicml.
chris looks at how you can test whether it's ingested the book directly or using a secondary source and how it affects hallucinations, in addition we look at how you can prompt openai to reveal it's training data.
this is the original data that i used to make this video, it's a little more detailed
www.dropbox.com/s/s6i8ogmeebv...