According to TechCrunch AI's report, Inherent's announcement highlights a growing, and arguably more interesting, trend than raw performance: the push for AI with 'taste.' The focus on a tiny model beating giants is a classic underdog narrative, but the more substantive bet is on reinforcement learning to instill judgment, not just knowledge. In our view, the risk is that 'taste' remains an ill-defined, unmeasurable goal, making it a convenient shield for any shortcoming. The real test won't be replicating known results but generating novel, testable hypotheses that advance a field—a leap the company itself admits is still a 'north star' and not a present reality.