The discussion shifts to the value of the "harness" or application layer versus the base AI models. Jason Calacanis highlights the emergence of companies like Micro One that create proprietary data sets for training. He also references legal troubles at OpenAI involving Greg Brockman's internal comments about bypassing paywalls. Calacanis predicts that open-source models will eventually catch up to frontier models, allowing businesses to maintain data privacy.