Lambert expects AI to automate infrastructure work but not yield general superintelligence
Overview
Nathan Lambert of Interconnects says AI models will become superhuman at distributed GPU engineering within a few years, though he does not expect this to make models fundamentally different in nature.
He predicts inference costs will fall near-exponentially as agents optimize training and serving stacks, and that pretraining architecture and data selection will be automated within 2 to 3 years. He also says the quality of reinforcement learning environment data is currently low but fixable. The report is a single post, and no timelines or cost figures beyond these predictions are given.
Written by AI from the articles below · updated Oct 9, 5:56 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- Interconnects (Nathan Lambert)BlogResearcher expects rapid AI infrastructure gains, not general superintelligence
AIInterconnects' Nathan Lambert says AI models will become superhuman at distributed GPU engineering within a few years, but that will not make models dramatically different in nature. He expects inference cost to fall near-exponentially as agents optimize training and serving stacks, with pretraining architecture and data selection automated in 2-3 years. He also says RL environment data quality is low and fixable.
Heat trend
Not enough continuous observations to show a trend yet.