Do the Biorisk Evaluations of AI Labs Actually Measure the Risk of Developing Bioweapons?

Audio versions of blogs and papers from BlueDot courses. 

Original text: https://epoch.ai/gradient-updates/do-the-biorisk-evaluations-of-ai-labs-actually-measure-the-risk-of-developing-bioweapons 

Author(s): Anson Ho and Arden Berg. 

A podcast by BlueDot Impact.

Chapters

• 3:17 1. Publicly described benchmarks are common but saturate rapidly, with uncertain implications for biorisk
• 12:43 2. We know little about most other AI biorisk evaluations
• 12:49 Evaluations are generally light on detail, and often by design
• 17:42 In practice, many biorisk evaluations only tell us about one model
• 22:24 3. Existing evaluations do not fully address the positions most skeptical of LLM-driven biorisk
• 23:14 The need for somatic tacit knowledge
• 25:11 The importance of infrastructure access
• 26:19 These objections are valid but do not rule out AI biorisk concerns
• 27:36 Discussion