Escape From Plan A

Ep. 578: Tell Me About DeepSeek In the Style of 3 Asian Dudes Bullshitting (ft. Mike and Filip)

·59 min·5 clips
A Chinese quant fund trained a top-tier AI model for under $6 million, challenging billion-dollar tech giants.
The episode begins with casual banter about personal updates, including Mike's experience with wildfires in Santa Monica. The discussion quickly pivots to the main topic: DeepSeek, a new AI model developed by a Chinese quant trading fund. The hosts note that DeepSeek was trained for under $6 million, which is three orders of magnitude cheaper than frontier models from OpenAI or Anthropic. They explain that DeepSeek is open-source under the MIT license, making it freely accessible. The conversation delves into the technical aspects, such as parameter sizes, inference efficiency, and the ability to run models on consumer hardware like Macs or iPhones. The hosts debate whether this breakthrough will 'rug-pull' the AI investment bubble, referencing the $500 billion Stargate project and hype from figures like Sam Altman and Larry Ellison. They compare the current AI cycle to historical tech bubbles, like the dot-com era, where infrastructure was overbuilt before use cases emerged. Philosophical questions arise about whether AI is approaching general intelligence, with mentions of AGI and sci-fi comparisons. The hosts critique the business side of AI, highlighting companies like Scale AI and the role of confidence in valuations. They also touch on geopolitical angles, such as reactions to Chinese tech advancements and data privacy concerns. The episode concludes with a teaser for a Patreon-exclusive part two.

As heard by us

A loose, pointed argument about AI, open source, and control.

Escape From Plan A turns a loose exchange with Philip and D-Gen Mike into a broader argument about AI, power, business, and technological control.

Read the full review in PlayNext →

Why you'd press play

Press play if you want open-source AI argued through tech history, corporate resentment, and a toaster that might talk back.

Read the full recommendation in PlayNext →
Listen to the show on