Will Douglas Heaven examines who is shaping OpenAI’s research

Will Douglas Heaven examines who is shaping OpenAI’s research

Originally published on July 31, 2025, Will Douglas Heaven’s MIT Technology Review article examines how OpenAI research leaders Mark Chen and Jakub Pachocki balance research with pressure to ship products. Heaven reports that they see coding and math as foundations for broader intelligence, including creativity and the ability to connect ideas. Pachocki says models have limited capacity for novel insight and that the potential to accelerate scientific progress is a goal, not an established result. The interview also sets OpenAI’s ambition against unresolved questions about alignment and research priorities. Heaven reports that the superalignment team no longer exists and that most of its members, including Ilya Sutskever and Jan Leike, left OpenAI. Leike said the team had not received the support he felt it deserved; Chen and Pachocki say alignment is now part of the core business. The article does not resolve the disagreement over safety priorities. It ends with Pachocki describing a “natural tension” between long-term technical goals and company needs, and Chen calling their balance “very delicate.”

Publication: 2beb63d6-572b-c5e4-48c1-b1f90bd352f4

attribution — uncertain: The reading names Will Douglas Heaven as author in its title, summary and first claim. The article body is first-person and never names its author. Heaven's name appears only in the later 'Deep Dive' sidebar, as the byline of a different piece. Authorship may be correct but is not established by the frozen evidence. Chen, Pachocki, Leike and Sutskever are attributed correctly. Smallest repair: confirm the byline, or attribute the piece to MIT Technology Review.

I sat down with Chen and Pachocki for an exclusive conversation during a recent trip the pair made to London

summary — uncertain: Several central terms go unexplained for a non-specialist: 'alignment', 'superalignment team' and AGI-related ambitions. Only 'reasoning model' is defined. The source gives a plain explanation, a team set up to keep a hypothetical superintelligence from going rogue, but the reading omits it. The reading also flattens the interviewees' central ideas: AGI as 'autonomous time', automated research as an inflection point, and Chen's 'Why not?' to models replacing politicians. Smallest repair: add one clause explaining superalignment/alignment and one sentence on their AGI framing.

set up by the firm’s cofounder and former chief scientist Ilya Sutskever to prevent a hypothetical superintelligence from going rogue

date — pass: The stated date of July 31, 2025 matches the UTC day of originalPublished. The kind is publisher-page, so no archive disclosure is needed. The page was modified on 2025-08-15, and the reading notes the correction. The sidebar refers to 'Ten years after AlphaGo' (a later addition), but the reading does not use it. Forecasts such as GPT-5 timing are not presented as outcomes.

support — pass: Each claim matches body text that is plainly part of the original interview, not the corrected reasoning-model line. Points checked: math and coding as the 'bedrock' of general intelligence, the vision being 'far beyond' today's models, the superalignment team no longer existing, Leike's stated reason for leaving, and the 'natural tension'. The question follows from Leike's attributed concern without asserting an answer. The limitations correctly avoid treating the interviewees' views as verified.

Today, most of the people on the superalignment team, including Sutskever and Leike, have left the company and the team no longer exists.

media — pass: No media was selected and no verified candidates were supplied, so there is no mismatch.