We recently received access to OpenAI’s ChatGPT, and we decided to test a few simple concepts in which a traditional chat agent would perform poorly. In a sense, it was an attempt to perform a double Turing test where both AI agents are being put to the test to see if they can plausibly pass as humans. What we found was rather exciting. We tried a few different variations of conversation between the names of portfolio companies and branded/unbranded counterparties to see what the system generated, with no additional insights in the prompt.