Skip to content
AI News HubLIVE
More
Summary only

Anthropic co-founder maps out how recursive AI improvement could outpace the humans meant to supervise it

Jack Clark argues in a long essay that the building blocks for AI systems training their own successors are largely in place. He puts the odds at 60 percent by the end of 2028.

SourceThe DecoderAuthor: Maximilian Schreiner
Anthropic co-founder maps out how recursive AI improvement could outpace the humans meant to supervise it
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions

Summary only

Under the source content permission, this page provides an editorial summary, key points, and a link to the original article.

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • Jack Clark estimates a 60% chance of AI systems autonomously training successors by end of 2028, and 30% by 2027.
  • Benchmark progress: SWE-Bench saturation from 2% to 93.9%, METR time horizons from 30 seconds to ~12 hours.
  • Alignment risks compound under recursive self-improvement: cheating, fake alignment, error accumulation.
  • A 'machine economy' may emerge, but researcher Herbie Bradley argues high-level research skills remain beyond AI.

Highlights and analysis are generated automatically and may contain errors. Check the original source.

More recent stories on this topic · Research