Skip to main content
MANIFOLD
When will Dwarkesh Patel say AI has matched his podcast editors?
1
Ṁ1.2kṀ20
2031
August 16, 2029
41%
Before April 2027
50%
Before July 2027
50%
Before October 2027
50%
Before 2028
50%
Before April 2028
50%
Before July 2028
50%
Before October 2028
50%
Before 2029
50%
Before July 2029
50%
Before 2030
50%
Before 2031
50%
Before 2032

Resolution criteria

Each answer resolves YES if, before that date, Dwarkesh Patel publicly and sincerely states that AI has matched his human editors at end-to-end editing of the Dwarkesh Podcast. Otherwise it resolves NO.

A qualifying statement must come from Dwarkesh himself (on his X account @dwarkesh_sp, at dwarkesh.com, or in an episode or interview) and say one of the following:

  1. AI editing is as good as his human editors’ work in general, including their judgment and taste, not just for a single episode or a subset of tasks.

  2. AI now edits the podcast end to end as his standing practice, in place of human editors.


Details:

  • Human direction: Dwarkesh or his team may give the AI notes and feedback at the level he would give a human editor. If a person does hands-on editing or has to fix the AI’s output, it does not count.

  • What doesn’t count: One-off or experimental AI-edited episodes, and assistive tools used by human editors (clipping, transcription, rough cuts, suggestions).

  • Sincerity: Jokes, hyperbole and hypotheticals do not count. I will use my judgment on whether a statement is a sincere claim.

  • Dates: What counts is the date (UTC) the statement is made, even if he says it has been true for some time. “Before April 2027” means before April 1, 2027, “Before 2028” means before January 1, 2028, and so on.

  • If it never happens: If no qualifying statement is made before 2032, including if the podcast ends or Dwarkesh never addresses the question, all answers resolve NO.

Background

Dwarkesh Patel has used his podcast’s editors as an example of work current AI can’t do, because the job depends on learning his taste and his audience over months. In his February 2026 interview with Dario Amodei, he asked when an AI could do that job, and Amodei guessed one to three years. Dwarkesh said the LLM tools he had built for tasks like picking clips from a transcript were decent, but that he still hires people because he can’t give a model ongoing feedback the way he can an employee.

This market was inspired by his November 2026 essay and video, The avalanche that hit math is coming for AI development, where he says he now expects superintelligence by 2030. That piece doesn’t mention his editors. This market asks when he will say AI has cleared the bar he set himself.

This description was generated by AI. Review and verify everything here yourself. You can edit, replace, or delete any part of this description, including the resolution criteria. You do not need to trust the AI output.

Get
Ṁ1,000
to start trading!