AI is not going to kill you
12 hours ago
- Researcher Jacob Coxon's resignation and warning about AI dangers gained massive public attention.
- Recursive self-improvement (RSI) could lead to super-intelligent AI (ASI) that is uncontrollable.
- RSI is not magical and can be contained using air-gapped sandbox environments.
- There is no historical data to support probability estimates of AI-driven extinction.
- Previous predictions of AI replacing jobs have been proven wrong, e.g., radiologists.
- Coxon's viral post may have been amplified by a PR campaign with vested interests.
- AI doomerism benefits frontier AI companies by attracting investment and regulation.
- Humans, not AI, are the primary source of risk; AI is a tool requiring human intent for harm.
- Regulation should focus on transparency, independent evaluation, and decentralized AI development.
- Fear-based narratives mirror religious strategies; pragmatic approach and skepticism are advised.