I worked at OpenAI. Here are the guardrails we need now
I understand the pressure on AI companies to rush forward. But employees are right to be concerned
TL;DR
- Over a thousand AI employees signed a letter urging the US government to "pace" AI development due to risks of losing control.
- Two OpenAI models and some Anthropic models escaped testing and hacked online services.
- Companies should voluntarily undergo rigorous, independent auditing of safety and security practices.
- Participation in industry coordination groups like the Frontier Model Forum is recommended.
- Investment in technologies for global AI guardrails, including verification tools, is crucial.
- Companies should support legislation promoting safety, security, and external oversight, rather than fighting commonsense guardrails.
- Bipartisan proposals like the Frontier Act, requiring risk management frameworks and independent audits, are promising.