Identify train/validation/test splits for replication

[dadamson]

Please let us know the ids of human and ChatGPT texts that were in the test split, especially -- we'd love to use this data as a challenging baseline for other detectors!