What graph best tracks AI progress? A post questions benchmark limits
The post calls METR long horizons “effectively saturated” and says Frontier Math and ARC-AGI are saturated too, prompting a question about how to chart progress.
TLDR
A user asks which quantitative graph best captures what they call AI’s “current exponential progress curve.” They describe METR long horizons as “effectively saturated,” citing Epoch as finding that pre-Fable agents could do 18 weeks of human work. They also say Frontier Math and ARC-AGI are saturated.
Combined views
54.5K
1 Source, first seen ago