WildChat-1M
A public dataset released by the Allen Institute for AI containing conversation logs between real users and a free ChatGPT front end. It is used to study real-world traffic patterns and user interaction behavior.
Timeline (last 90 days)
Under the default 5-minute window on the WildChat-1M dataset (59,857 conversations, 172,418 user turns, 25.3 days of traffic), only 2,568 calls (1.49%) were exact duplicates; with the window stretched to 24 hours it was 4.39%, and across the full 25 days it was 4.77%.
Within the 5-minute window, 91% of duplicates came from the same user (same hashed IP) sending the same content again, and the most-duplicated request was the system prompt used by AutoGPT (beginning with 'CONSTRAINTS: 1. ~4000 word limit...'), which was repeated 76 times.
The study sampled 1,242 English-only prompts from six data sources: Dolly-15k, LMSYS-Chat-1M, WildChat-1M, MMLU, GSM8K, and HellaSwag
All Facts (3)
Under the default 5-minute window on the WildChat-1M dataset (59,857 conversations, 172,418 user turns, 25.3 days of traffic), only 2,568 calls (1.49%) were exact duplicates; with the window stretched to 24 hours it was 4.39%, and across the full 25 days it was 4.77%.
50%UnverifiedWithin the 5-minute window, 91% of duplicates came from the same user (same hashed IP) sending the same content again, and the most-duplicated request was the system prompt used by AutoGPT (beginning with 'CONSTRAINTS: 1. ~4000 word limit...'), which was repeated 76 times.
50%UnverifiedThe study sampled 1,242 English-only prompts from six data sources: Dolly-15k, LMSYS-Chat-1M, WildChat-1M, MMLU, GSM8K, and HellaSwag
50%