Research Scientist, Foundation Models

Good AI Labs is developing foundational models for robots that can learn new tasks and act in unfamiliar environments. We are hiring a Research Scientist to advance the capabilities of these models through new architectures, training methods and learning signals.

You will own a research direction from the idea to the trained model. The scope includes multimodal pretraining, representation learning, adaptation and the use of simulation and reinforcement learning. This is a role for someone who has trained models, understands why a run succeeds or fails, and has a considered view on what the next generation of embodied intelligence should learn.

Responsibilities

• Develop and train foundational models that connect visual observations, language and physical interaction.
• Investigate pretraining objectives, data mixtures and model architectures that improve transfer across tasks and environments.
• Design training experiments that separate the effects of model capacity, data quality, optimisation and compute allocation.
• Explore how simulation, demonstrations and reinforcement learning can extend capabilities beyond pretraining.
• Own the experimental work: implement the method, diagnose training behaviour, run ablations and establish where the improvement holds.
• Work with research and systems engineers on training efficiency, inference constraints and the path from a research result to robot behaviour.

Qualifications

• BS, MS or PhD in computer science, machine learning, mathematics or a related field, or demonstrated research and engineering work of equivalent depth.
• Original contributions to foundation models or multimodal learning, with direct responsibility for model implementation, training and evaluation.
• A strong research record demonstrated by first-author or leading contributions at venues such as NeurIPS, ICML, ICLR, CVPR or CoRL, or substantial industrial or independent research with comparable originality and technical depth.
• Hands-on experience training deep-learning models: preparing data, choosing objectives and optimisers, managing checkpoints and resolving instability or poor generalisation.
• Strong Python and PyTorch skills, with the ability to implement research ideas beyond an existing training recipe.
• The judgement to identify an important open problem, test competing explanations and decide which result warrants further investment.

Preferred experience

• Multimodal foundation models, generative modelling, video learning or representation learning from sequential data.
• Distributed training, mixed precision, training-data curation or analysis of scaling behaviour.
• Reinforcement learning, simulated environments or transfer from pretrained representations to physical tasks.

Application materials

Send your CV and two or three contributions that best represent your research. For one model you trained, describe the problem, your own implementation work, the training setup and the result. Papers, code, technical reports and accounts of substantial proprietary work are all relevant. Tell us which research question you would want to own here.