This dataset (specifically the file Mistral7B_CME_v7.csv
) consists of 4900 LLM generated texts.
(Note: versions 1 to 6 are redundant, and are only kept so as not to break any notebooks that use them)
Update: The new file Mistral7B_CME_v7_15_percent_corruption.csv
has also been added as per the discussion "Alternative approach - Simulating hidden dataset".
v1: 700 LLM texts for prompt 6 "Exploring Venus" for use in the LLM - Detect AI Generated Text competition.
v2: + 700 LLM texts for prompt 8 "The Face on Mars"
v3: + 700 LLM texts for prompt 4 "A Cowboy Who Rode the Waves"
v4: + 700 LLM texts for prompt 11 "Driverless cars"
v5: + 700 LLM texts for prompt 7 "Facial action coding system"
v6: + 700 LLM texts for prompt 2 "Car-free cities"
v7: + 700 LLM texts for prompt 12 "Does the electoral college work?"
Photo credit: Image of Venus by NASA.