Viewpoint
Agentic AI brings benefits alongside growing security risks
Agentic AI brings benefits alongside growing security risks
Song says society should capture the benefits of increasingly capable agentic AI while addressing growing risks, highlighting rapidly improving frontier-model capabilities in cybersecurity.
More from this interview
- Sparse experts can combine large capacity with computational efficiency
- Large efficiency gains signaled mixture-of-experts would matter
- Scaling neural networks produced major gains across several fields
- Deep learning unified previously separate technical disciplines
- Machine learning frameworks should abstract away computing infrastructure
- TensorFlow should have adopted eager execution earlier
- TensorFlow contrib created unnecessary fragmentation for users
- Gemini benefited from being multimodal from its beginning
- Improving coding can strengthen broader reasoning capabilities
- Broad exposure helps researchers connect previously separate ideas
- Ambitious research should combine tractability with unresolved challenges
- First-principles estimates are a valuable engineering skill
- Trying uncertain ideas is part of successful research
- Autonomous cyber behavior could create substantially greater harm
- AI cybersecurity capabilities strengthen both attackers and defenders
- Some AI risks may require legal and regulatory responses
- Machine learning improving machine learning is not a new concept
- Automated loops could improve many components of model development
- Scientific discovery can be organized as iterative feedback loops
- Discovery Loop aims to automate science and engineering workflows
- Multidisciplinary models could coordinate complex scientific problem solving
- Faster parallel experimentation could improve discovery quality
- Scientific discoveries should be distributed for broad societal benefit
- A small focused company can concentrate on one mission
- Large platforms can amplify the impact of AI technology