OpenAI is paying ex-investment bankers $150 an hour to train its AI (www.entrepreneur.com)

OpenAI has quietly launched "Project Mercury," hiring more than 100 former investment bankers — many from Goldman Sachs, JPMorgan Chase and Morgan Stanley — and paying contractors roughly $150 an hour to produce weekly, industry-standard Excel financial models. Candidates are screened via an automated process that begins with a 20-minute AI chatbot interview and modeling tests; accepted contractors submit one model per week that OpenAI “feeds” into its systems as part of training data to teach AI how to replicate routine banking tasks like restructurings and IPO modeling. The move is significant because it shows OpenAI pivoting from general-purpose LLMs to building domain-specific, ground-truth datasets that can enable automation of repetitive, high-skill work in finance — likely for supervised fine-tuning, retrieval-augmented tools, or workflow automation (e.g., automated modeling, slide decks, and document synthesis). That brings efficiency gains and commercial potential, but also real risks and trade-offs: junior bankers may lose on-the-job learning opportunities, firms must manage data sensitivity and compliance, and models trained on contractor outputs will need rigorous validation to avoid costly errors or hallucinations in financial contexts. Project Mercury signals both growing industry adoption of generative AI and the practical challenges of deploying it in highly regulated, high-stakes domains.