Research
Developers and evaluators should note that benchmarks like FrontierML are beginning to measure whether AI systems can autonomously delegate and spend resources to improve themselves, which bears on how agentic systems are scoped, monitored and governed.
This edition of Import AI covers research on AI systems that sub-contract tasks to improve themselves, reporting that AI can now productively spend money to do so. It introduces FrontierML, a benchmark for measuring this kind of capability. The issue also touches on a wet lab for AI and subterfuge during model development.
Read the full article at Import AI
CoFabrix summarises and comments on this story. The original reporting belongs to Import AI.
Find out where your organization stands -- and what to do about it.