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

This section discusses how current AI models excel at solving complex, objective problems like math and coding but fall short in subjective tasks like design, tweets, and creative content. Thais Castello Branco of Taste Labs highlights that AI lacks taste, in part because training data often promotes regurgitation of the average, leading to regressive, uninspired results. She emphasizes the importance of incorporating human 'Taste Makers'—experts with deep domain knowledge and strong opinions—who can curate, critique, and guide models towards generating higher-quality, more diverse content. Their approach involves building a community of diverse curators who are paid and potentially invested in the outputs, aiming to elevate the standards once the baseline quality improves. The core challenge discussed is maintaining uniqueness and rarity, avoiding commodification of taste, and ensuring AI outputs remain vibrant and personalized rather than homogenized.

This section explores the difficulties AI faces in subjective tasks, particularly in beauty, style, and personal preferences. Branco discusses how variety, nuance, and personalization are often missing from AI outputs because models tend to favor regressing towards the mean, producing generic results. She argues that creating a system capable of mimicking the rare skill of developing authentic taste is incredibly difficult, and true taste is a rare skill built through exposure and experience. The solution proposed involves leveraging human curators who can teach models to appreciate and generate better-styled content, and the goal is to lift overall content quality before tackling deeper personalization and style.

This section philosophizes on whether commodifying taste through AI erodes its value and rarity. Branco reflects that taste has historically been curated by elite critics and tastemakers, but with AI and social media, these functions are dispersed and often diluted. She worries that AI could lead to a 'flattening' of taste by recommending the same widely available content, thereby losing its unique, niche qualities. However, she also suggests that as model capabilities improve and baselines are raised, authentic and high-quality expression will become more valuable, not less, because consumers will seek out the exceptional rather than the average. The conversation emphasizes the importance of quality over mere popularity.

This section covers the business and strategic aspects behind the efforts to teach AI about taste. Branco's company, Taste Labs, has raised $18.5 million in seed funding from investors like Amplify Partners. They focus on visual design and aim to build a 'design index' sourced from expert curators to improve AI outputs. She discusses the idea of acquiring historic and niche content like magazines or archives to preserve and enrich the cultural backbone of taste. Partnership and licensing with creative agencies and cultural institutions are considered as future directions. It’s also noted that her team is hiring across various roles and that the company is based in San Francisco, amid the city's rising housing crisis.

This section shifts to broader societal impacts of AI and emerging technologies. Topics include Google's AI-enhanced Google Earth, which can generate fake images and false information that could be used maliciously, such as misrepresenting border invasions or conflict zones. The podcast discusses the risks of misinformation, the spread of AI-generated content, and governmental regulation as seen in China's bans on self-driving vehicles to prevent job displacement. It also touches on the tech industry's tendency to add AI “a little dab” into products regardless of potential misuse, and the importance of responsible deployment, including possible licensing or revenue sharing models to address societal harms. Broader themes include the challenge of controlling AI-driven misinformation, the societal impacts of automation, and the importance of thoughtful regulation.

This section reviews innovations in autonomous transportation, including self-driving golf carts at Stanford and the progression of driverless vehicle technology. The discussion highlights the rapid commoditization of self-driving tech—companies like Tesla, Baidu, and startup projects are racing to deploy fully autonomous vehicles. The economic impact is significant, especially regarding job displacement for drivers, which raises societal and regulatory concerns. The episode also covers AI-powered robotic systems for solar panel installation, illustrating the increasing role of automation in renewable energy infrastructure. It emphasizes that larger factory-scale, robotic, and AI-led solutions for energy and environment infrastructure could revolutionize how we generate and manage power, water, and resources, with an eye toward sustainability and efficiency.