A Chinese AI lab released DeepSeek, an open-source model that rivals OpenAI's but cost only millions to train.
The episode begins with an introduction to DeepSeek, a Chinese AI model from a research lab backed by a hedge fund. It was trained cheaply, runs efficiently on consumer hardware, and competes with expensive Western models like those from OpenAI and Anthropic. The hosts detail how OpenAI's Sam Altman accused DeepSeek of using ChatGPT outputs for training, sparking a debate on content theft and hypocrisy. They delve into the Chinese context, explaining that the hedge fund faced government crackdowns in 2021, leading to a pivot to AI development under state encouragement. The discussion questions whether Western AI firms inflate funding needs, citing Anthropic's Dario Amodei's analysis. The episode shifts to L4S, a new telecom standard being piloted by Comcast to reduce latency by allowing data packets to report congestion, potentially speeding up internet responsiveness. The hosts then nostalgically review 2018 tech trends, including Fortnite's impact, Taylor Swift's tour, and debates on Facebook's ethics post-Cambridge Analytica. They note how little has changed, with similar issues persisting. The episode concludes with announcements about a podcast hiatus and reflections on seven years of production, emphasizing the show's role in informing listeners during routine activities like cleaning.
