Research Scientist Intern, World Models San Francisco, CA · Internship · 12-24 weeks →

Reactor
San Francisco, US
On-site

Job Description

About Reactor Reactor is building the missing layer for world models. We are hiring a small number of Research Scientist Interns across five open research problems. We hold our research team to a frontier standard, and we actively support interns in publishing their work at top venues (NeurIPS, ICLR, CVPR, SIGGRAPH, and equivalents).

Research Areas World Model Evaluation & Benchmarking — Better ways to measure what matters in world models: action-conditioned prediction, physical plausibility, causal consistency, and long-range coherence.

Long-Horizon Consistency & Spatial Memory — Understanding and resolving the error accumulation and identity loss that limit current world models over extended rollouts.

Ultra-Low Latency Interactive Generation — Making real-time interaction with world models possible by rethinking generation architectures with latency as a first-class constraint.

Adaptive Controllability — Developing sample-efficient methods for imposing structured control over pretrained generative video models, without large-scale retraining or model-specific fine-tuning.

Representation Learning for World Modeling — Learning structured latent representations that reflect world dynamics rather than predicting directly in pixel space.

Who We're Looking For PhD student (1+ years in) in ML, computer vision, robotics, or graphics

Hands-on experience with conditioned generative video models (diffusion, autoregressive, flow-based, or similar)

Publication record (or strong in-progress work) at top venues; quality over quantity

Genuine depth in at least one research area above

Able to own a problem from formulation to written results

Strong PyTorch and/or JAX skills and comfort with large-scale training infrastructure

Intellectually honest: clear-eyed about results, serious about failure modes

What We Offer Direct collaboration with the founding research team

Sufficient compute to train large models

Dedicated support for publishing at top conferences

Clear path to full-time offers with equity for strong interns

Reactor Inc. is an equal opportunity employer.

#J-18808-Ljbffr Reactor

Skills & Requirements

Technical Skills

Machine learningComputer visionRoboticsGraphicsPytorchJaxLarge-scale training infrastructureIntellectual honestyProblem ownershipCollaborationPublication recordWorld modelsMachine learningComputer visionRoboticsGraphics

Employment Type

INTERN

Level

intern

Posted

4/12/2026

Apply Now

You will be redirected to Reactor's application portal.