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

The episode discusses AI's lack of taste in creative outputs and how Taste Labs uses human taste makers to train models. They cover Leopold Aschenbrenner's leveraged AI-focused hedge fund blow-up, Google Earth's AI image manipulation enabling misinformation, LinkedIn and Substack's anti-AI slop measures, and a migration crisis in Ceuta. Jason closes with optimism about solar energy, robotics, and autonomous driving solving major global challenges. - Guest Thais Castello Branco explains AI models regress to the mean on subjective tasks, making their creative outputs generic and tasteless. - Taste Labs built a community of paid taste makers who critique and curate AI outputs to improve aesthetics and quality. - Leopold Aschenbrenner's AI hedge fund faced a margin call, sold its public stock to Citadel, but retains private holdings including $5B in Anthropic. - Google Earth integrated Gemini's Nano Banana image generator, allowing users to create realistic fake satellite views, raising misinformation fears. - LinkedIn and Substack introduced tools to flag or limit AI-generated slop content on their platforms. - A Stanford grad built a self-driving golf cart using only cameras and vision models, illustrating commoditization of autonomous driving tech. - Jason argues solar energy and batteries are on the verge of solving energy, water, and agriculture, supported by robotic installation and solar canopies on canals.