Humans Are Reading Your ChatGPT Conversations
Summary
404 Media investigates "Project Lily," revealing that OpenAI hires contractors to read real ChatGPT user prompts to improve model responses, despite buried privacy disclosures. The episode also covers the chaotic leadership crisis at Automattic, where CEO Matt Mullenweg was briefly ousted by the board before claiming to regain control and replacing key executives. Hosts Joseph, Sam, and Emmanuel discuss the ethical implications of human data review in AI training and the ongoing turmoil in the WordPress ecosystem.
Topics discussed
Sponsor: LinkedIn Hiring Pro
404 Media subscription and bonus content
Intro and AI ethics discussion recap
Project Lily: Humans reviewing ChatGPT chats
Reviewer dashboard and user data exposure
Privacy model limitations and PII leaks
Anonymization techniques and context removal
Contractor tasks: rating prompts and responses
Sycophancy risks and user safety concerns
ChatGPT's conversational personality and utility
Rationale for scores and model improvement
Buried disclosure of human review practices
Comparison with Gemini and safety monitoring
Distinction from previous Kenyan worker reports
Opt-in settings and enterprise data risks
Sponsor: BetterHelp
Sponsor: Babbel language learning
Sponsor: NordLayer Browser
Matt Mullenweg put on leave of absence
Background: Automatic, WordPress, and WP Engine
2024 feud and internal company turmoil
The board vote and Mullenweg's reaction
Mullenweg claims he is back in control
Board statement and executive support
The tugboat blog post and trolling
Board changes and removal of members
CFO departure and current board status
Outro and subscription call to action
Listen ad-free on Castria