Forecasters pull their timelines forward eighteen months in three

Eli Lifland and Daniel Kokotajlo of the AI Futures Project wrote that they had moved their timelines roughly a year and a half earlier over three months, citing faster expected time-horizon growth and coding agents impressing in the real world, after having lengthened them through 2025 (tweet). In the same paragraph Sam Altman, explaining the Sora shutdown, said OpenAI had a few times in its history realized something was working or about to work so well that other projects had to stop, and that it needed to concentrate compute and product capacity on the next generation of automated researchers (Andrew Curran’s transcript); the recursive-self-improvement reading of that remark is the newsletter’s, and Brad Lightcap added that training cycle time “is starting to collapse”. The revision overtakes the AI Futures Model’s July 2034 superhuman-gap date and reads as corroboration of the automated-research-intern target reported twelve days earlier; the shutdown itself was recorded in the AGI Deployment renaming. Nine days later a separate scorer found the 2027 roadmap 88% accurate so far.