I expect rapid progress but not towards general superintelligence
Rapid AI engineering progress is coming but won't produce general superintelligence or broad superhuman economic value
“I expect the effective cost of model intelligence to decline near-exponentially over the coming years (potentially faster than recent trends).”— Nathan Lambert
Nathan Lambert argues that AI will become superhuman at infrastructure and engineering tasks within a few years, driven by inference efficiency gains and AI-assisted optimization, but this won't produce general superintelligence or economically valuable superhuman capabilities outside math and coding. The bottleneck in AI research is shifting away from engineering execution back toward ideas, as coding agents lower implementation friction. Near-term progress is scaling inference-time compute with existing architectures rather than fundamental model breakthroughs.