Overview
- Project lead Daniel Kokotajlo said progress is running slower than first outlined and put his personal short‑term estimate at around 2030.
- The update moves the expected arrival of fully autonomous coding from 2027 to the early 2030s and drops any date for human‑ending scenarios.
- The new forecast names 2034 as a tentative horizon for superintelligence while acknowledging substantial uncertainty in timing.
- Critics such as NYU’s Gary Marcus deride the scenario as speculative fiction, whereas Oxford’s Dr Fazl Barez warns capability growth is outpacing safety work.
- Context from industry includes Sam Altman describing an automated AI researcher by March 2028 as an internal goal that OpenAI might fail to meet, alongside expert notes on jagged performance and real‑world integration hurdles.