Program and guests

a16z’s Working Theory aired with OpenAI Chief Research Officer Mark Chen and co-founder John Schulman as guests; the episode runs 16:19. Original English title: Working Theory: Recursive Self-Improvement with Mark Chen & John Schulman.

Key takeaways

  • AGI is a spectrum, not a moment: Chen says the AGI spectrum has advanced about 80%, while the recursive self-improvement spectrum is about 25–50%. 4:55
  • A Meter report dated 5 November estimates that the pace of AI research has accelerated by about 2 times over the past year. 5:22
  • The bottleneck to faster research is research “taste,” not computing power. 5:49
  • New progress in alignment includes model “confessions”: models proactively reporting their own errors. 12:18
  • Automated researchers can fix 10 different types of alignment failure. 13:02

Background and notes

a16z’s Working Theory features a discussion with OpenAI Chief Research Officer Mark Chen and co-founder John Schulman on the definition and progress of recursive self-improvement, as well as alignment. Subtitle coverage note: the video’s subtitles contain 495 timestamped segments, covering 0:00–16:18 (total runtime 16:19), and cover the full video. The views and timestamps in this article are paraphrased from the original subtitles; each timestamp marks the start of the corresponding segment of the original statement. Publication time note: the publication time is 2026-11-18 02:15 Hong Kong time, converted from the relative time shown by YouTube when retrieved (7 hours ago), with hour-level precision.

Source links

Further reading

Related topics