Severin Field, an IAPS fellow, surveyed 25 researchers from OpenAI, Anthropic, Google DeepMind, Meta, and US universities regarding recursive AI self-improvement. In a new blog post, he compared their previous forecasts with current developments.
According to The Decoder, several milestones named by the researchers have already been passed. The source does not specify in the provided material which specific milestones are involved or when they were achieved.
This is significant because estimates regarding the timeline for automating AI research are used to understand the industry's pace of development. However, the available data does not allow for a conclusion about the onset of full-scale recursive self-improvement or its consequences.
editorial commentary
Why it matters
The probable consequence is increased attention to verifying timelines for AI research automation rather than abstract forecasts. The next observable signal will be the publication of a precise list of milestones and the criteria for their achievement. Significant uncertainty remains: the provided source does not disclose details and is not confirmed by a second independent publication.