Skip to content
LessWrong· Tom Davidson·Indexed by this site · Originally published· Opinion →· Its place in the scan record →

Author updates judgment based on the METR report: AI R&D automation timeline shortened, intelligence explosion far from inevitable

Updates from METR report: short timelines vs IE

AI summary

Based on the METR report, the author believes full automation of AI R&D by AI is closer and faster than previously expected, only modestly raises the probability of an intelligence explosion, and considers it far from inevitable. The author assesses the feedback loop of AI improving AI in three stages: labor augmentation has already brought about a roughly 2x speedup without hitting compute bottlenecks, but compute bottlenecks may still emerge later. The author believes the overall evidence supports acceleration of progress, but it is far from overwhelming; even without acceleration, automated AI R&D would make progress 2-3 times faster.

Read originallesswrong.com