OpenAI Chief Scientist Predicts AI Will Drive Its Own Development
Pachocki's statement frames recursive self-improvement not as a distant scenario but as OpenAI's stated research direction, with the company treating it as the only way to stay at the frontier of AI research.
Reporting from 1 source: GIGAZINE.
OpenAI chief scientist Jakub Pachocki says AI will increasingly drive its own development, a path he calls recursive self-improvement. He traces the view to a 2023 research project and says reasoning models now produce results that push science forward. OpenAI has concentrated its research on RSI, and Pachocki warns that no one is fully prepared for the consequences of rising machine intelligence.
Pachocki dates his conviction to mid-2023, when the RLSlow research project convinced him that reasoning model training could be scaled and that models would gain the ability to form their own thoughts. Three years on, he says reasoning models are producing results that push the limits of science, which is why he now considers machine-driven self-improvement likely.
OpenAI has concentrated its research on RSI because the company sees it as the only way to remain at the frontier of AI research. Pachocki describes the current moment as one where machine intelligence is beginning to exceed human intelligence, a trend that becomes more pronounced as compute grows.
He is explicit about the risk. Pachocki says no one is prepared for the consequences of rising machine intelligence, and while OpenAI works on alignment, he argues broader intervention may be needed. The weak point is alignment generalization, the area most likely to lag behind the speed of intelligence gains. His proposed path combines stronger alignment and monitoring with a coordinated slowdown when reliability demands it.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.