OpenAI says AI agents finish nearly half of research tasks

11 sources
  • OpenAI estimated that its coding agents' success rate without human help on two-to-four-hour research tasks rose from 23% in March to 46% in October.
  • Results were uneven: agents succeeded 79% of the time monitoring training runs but only 25% writing inference code, and researchers flagged "AI slop".
  • OpenAI renewed calls for labs to publicly report signs of AI speeding up its own development, saying it doesn't intend to pursue full self-improvement.
Sources (11)
  1. 1 Research acceleration: The jagged frontier openai.com
  2. 2 OpenAI Reports Agent Gains on Multihour Research Tasks | Frontier thedefiant.io
  3. 3 David Holtz (@daveholtz) on X x.com
  4. 4 OpenAI Reports Agent Gains on Multihour Research Tasks | Frontier thedefiant.io
  5. 5 OpenAI reports agents complete an estimated 46% of two-to-four-hour research tasks letsdatascience.com
  6. 6 OpenAI's top researchers now burn $23000 in AI tokens a day cryptobriefing.com
  7. 7 OpenAI says its coding agents finish longer research tasks—and still write ‘AI slop’ digg.com
  8. 8 OpenAI estimates coding agents finish 46% of its two-to-four-hour research tasks superpowerdaily.com
  9. 9 Patching AI-found flaws becomes the new bottleneck sharedsapience.substack.com
  10. 10 Apple researchers say FastFrames generates video in 36 milliseconds ... superpowerdaily.com
  11. 11 OpenAI Research openai.com