Miju Labs

← All buyers · OpenAI

A

Aditya Ramesh

World Simulation Lead, OpenAI · OpenAI

medium confidenceTier B · Map and watchFrontier labsimage · video

Creator of DALL·E and DALL·E 2, later a Sora lead, and now OpenAI's World Simulation Lead — credited in that title on both Sora 2 (September 2025) and ChatGPT Images (December 2025). He owns the video and world-model line at OpenAI, which was reshaped when the standalone Sora app was discontinued in March 2026.

Background

Ramesh joined OpenAI as a researcher and built the original DALL·E, the model that made text-to-image a mainstream idea, then DALL·E 2. He moved into video with Sora and, by the Sora 2 release, was credited for the organisation of the effort rather than a specific technical component — the signature of someone who has become a research manager.

The context around him changed sharply in 2026. OpenAI announced on 24 March 2026 that it would discontinue the Sora video app, and the product was retired the following month. 'World simulation' as a framing survives the app: the bet is that video models are world models, which is a research agenda rather than a consumer product. What that means for data demand is a shift away from feed-friendly clips toward physically and temporally consistent footage.

What they run now

World simulation: video generation as a model of physical dynamics, with synchronised audio, controllability and character insertion. Sora 2 (30 Sep 2025) was framed around physically accurate motion that prior video models could not produce. The consumer Sora app is gone; the research line is not.

What they buy

Inference. A world-simulation programme buys video data and video judgement: professionally shot footage with known camera and lighting, physical-plausibility judgements ('does this violate physics'), temporal-consistency grading, and human preference over motion quality. It buys less pure aesthetics than the image line does. Note the contrast with Google DeepMind, where Dumitru Erhan has said publicly that professionally shot content beats random web video — the same constraint applies here.

Career

  1. 2019Research scientist (DALL·E, DALL·E 2), OpenAI — Dates approximate
  2. presentSora lead → World Simulation Lead, OpenAI — Credited 'World Simulation Lead' on Sora 2 and GPT-Image-1.5
  3. —New York University, BS, computer science

Interviews & talks

Why it matters here

Medium. Video judgement is a real and expensive category, and he is the person who owns it at OpenAI. But the Sora shutdown makes his near-term data appetite uncertain, and physical plausibility is a different sale from creative taste — it needs cinematographers and VFX people, not graphic designers. Worth a dossier; not worth the first call.

How to reach

Only after the Google DeepMind video conversation is live, so Julian can speak from a position of having done the work. Speculation: the credible opener is a small graded set of video generations rated by working cinematographers on motion, continuity and lighting, with agreement statistics — the same artefact that works for Erhan.

What we could not establish
  • No verified X handle, LinkedIn or personal site found.
  • Education listed from general public record, not confirmed in this pass.
  • What the World Simulation org's scope and headcount are after the Sora app shutdown.
  • Whether any video data procurement is external.

Why they are on the map

'World Simulation Lead' in ChatGPT Images credits (Dec 2025); Sora 2 'Organization' lead (Sep 2025). Sora app shut down Mar 2026 - current remit unverified source — evidence dated 2025-12

Decides whether the problem is funded.

How OpenAI buys data

Has a dedicated Human Data team that 'turns human feedback into reliable signals for training and evaluation' via bespoke campaigns with 'external vendors and AI trainers' and has open PM/TPM/EM/TLM roles on it (2025-26 job posts). Ended its Scale AI relationship after Meta's stake (Jun 2025); reported to be a major revenue source for Mercor and to use Surge and Handshake, while building in-house data capacity. GDPval (Oct 2025) and HealthBench (May 2025, 262 physicians) show it buys large expert-graded eval sets. Image models ship with a named 'Data & Evaluation' team (Dec 2025 credits).

A central Human Data team (PMs, TPMs, engineers) works with all research teams and vendors. Glaese, a former Human Data lead, is now VP Research & Safety (Jul 2026). Image generation sits under Multimodal (Dhariwal), with its own data leads and an 18-person Data & Evaluation group, plus Model Behavior. Sora was shut down in Mar 2026 and Peebles left in Apr 2026. Image work was folded into GPT Image 2 / 2.5.

Known vendors: mercor, surge-ai, handshake-ai, scale-ai (ended Jun 2025)

Sources

  1. Sora 2 is here — credits
  2. The new ChatGPT Images is here — credits
  3. OpenAI to discontinue Sora video app
  4. What to know about the Sora discontinuation