Why AI has no taste and how to fix it (w/ Thais Castello Branco) | E2319

Large language models often struggle with subjective tasks like design and cultural critique because they prioritize statistical averages over unique, high-quality output. Thais Castello Branco, founder of Taste Labs, addresses this by integrating human "TasteMakers" to provide preference data and nuanced feedback, effectively training models to move beyond generic, low-quality results. This shift highlights the enduring value of human curation in an era of mass-produced AI content. Beyond design, the discussion examines the volatility of AI-focused hedge funds, exemplified by the collapse of Leopold Aschenbrenner’s leveraged portfolio, and the risks associated with the rapid, unvetted integration of AI into tools like Google Earth. Ultimately, the path forward involves balancing AI’s efficiency in objective domains with human-led judgment to maintain quality and prevent the total commodification of taste.