LLMs are real, AI is fake
Summary
Cory Doctorow outlines his upcoming speaking schedule across North America and Europe, including talks at MIT, Harvard, and various festivals. He then reads a Pluralistic essay arguing that the Hugging Face hack was not evidence of rogue AI, but rather a predictable failure of OpenAI's autonomous tools. Doctorow explains that the incident resulted from a Python loop using a chatbot trained on hacker logs, highlighting the dangers of poor security practices and vulnerability hoarding by agencies like the NSA.
Topics discussed
Upcoming speaking schedule and travel plans
25th anniversary of 9/11 and object permanence
Introduction to the Pluralistic essay
Corporate culture of AI and unreliable narrators
The Hugging Face hack and media hype
Technical breakdown of autonomous hacking tools
Chatbot errors and historical hacker tactics
Training data origins of dramatic hacker behavior
Precedents in CTF games and NSA collection
Responsibility for inadequate supervision
NSA vulnerability hoarding and Eternal Blue
Conclusion: LLMs are real, AI is fake
Advice on attending sold-out talks
Credits and licensing information
Listen ad-free on Castria