At 11:58 P.M. EDT on Monday, Buckmaster posted his and Alpöge’s results, along with the statement laying out his account of events. According to the statement, he does not know how the OpenAI team so quickly reconstructed the details of its approach. Buckmaster wrote that he asked Bubeck if the OpenAI LLM was given access to his personal prompts to the company’s model Codex, which he and Alpöge were using throughout the process and thought were private. Bubeck told him no but wouldn’t answer whether the model’s training involved user data, which might have included Buckmaster and Alpöge’s private prompts, Buckmaster alleged.

Second to the achievement, I think this is the big takeaway and should be approached is if they did, in fact, use private prompts for training.