Why AI has no taste and how to fix it (w/ Thais Castello Branco) | E2319
Summary
Hosts Lon and Jason interview Thais Castelo-Branco of Taste Labs, discussing how AI lacks subjective taste and how her company uses human experts to improve model outputs. They also cover the collapse of Leopold Aschenbrenner’s hedge fund due to leverage issues, Google’s controversial AI image integration in Earth, and Anthropic’s report on model security incidents.
Topics discussed
Intro and sponsor reads for Odoo and YSecurity
News roundup: AI image manipulation and Anthropic updates
Interview with Thais from Taste Labs on AI lacking taste
The value of human curators and tastemakers in the AI age
Commoditization of taste and the 'slop' problem
Taste Labs hiring and the importance of preference data
Leopold Aschenbrenner's fund liquidation and leverage risks
Memes about Leopold and 2014 Oscar predictions
Google Earth AI features and Substack's anti-AI labeling
Migrant crisis at the Spanish border in Ceuta
Self-driving golf carts and the future of autonomous driving
Robotic solar panel installation and floating solar farms
Listen ad-free on Castria