Intrepid Growth Partners panel questions METR AI progress “time horizon” metric
Intrepid Growth Partners managers debated METR’s AI “time-horizon” benchmark, which estimates how long models can complete software tasks at a 50% success rate. METR data cited a roughly 10x rise in frontier-model task horizons over the past year, from about 1.5 hours to about 15 hours. Richard Sutton called the metric