LessWrong (Curated & Popular) “Winning the power to lose” by KatjaGrace
May 23, 2025
The power struggle in AI development takes center stage, highlighting a clash between those pushing for rapid advancements and advocates of caution. Anecdotes illustrate how winning power doesn’t guarantee safety or success. The conversation delves into whether the reigning accelerationists truly understand the consequences of their speed-driven approach. Ultimately, the focus shifts from individual victories to the importance of collective well-being in navigating the complex landscape of artificial intelligence.
AI Snips
Chapters
Transcript
Episode notes
Power Contest vs Real Game
- Winning a power contest doesn't mean winning the real game against reality.
- Accelerationists' win might feel like control but doesn't guarantee success against AI risks.
Seatbelt Anecdote on Real Outcomes
- Katia Grace shares a childhood seatbelt struggle with her brother illustrating a lesson in real outcomes.
- Despite 'losing' control, her brother survived a crash proving the importance of factual correctness.
Unified Stakes in AI Risk
- AI risks put most of humanity on the same side despite infighting over strategy.
- Power grabs by the less factually correct side hurt everyone, regardless of immediate advantage.