Your position
The successful candidates will investigate how large language models (LLMs) and AI agents can drive their own improvement. Research directions include:
- Post-training LLMs with reinforcement learning for long-horizon, multi-stepreasoning and agentic tasks
- Frontier learning: an open-ended post-training approach that keeps generating newtraining problems at the edge of a model's capabilities and tracks that edge as themodel improves
- Building RL environments for agentic tasks such as search, terminal use, and games,including synthetic environments, tasks, and rewards that LLMs generate and verifythemselves
- Weight-space programming: agents that improve open-weight models by actingdirectly on their weights, opening a path to models that improve themselves
- World models that let agents plan, imagine, and learn from simulated experience
- Open-ended and evolutionary search over programs, agents, and training recipes
These directions are closely connected, and there is plenty of room to shape your ownpath through them. Ideal candidates have a strong background in machine learning andare excited about large-scale experimentation with frontier AI systems, supported byprincipled analysis where it helps.
Starting dates are flexible. The duration of a PhD in Switzerland is 4 years.
Your tasks include:- Conduct research on recursive self-improvement and open-ended learning, from newideas to working systems
- Design and run large-scale experiments with state-of-the-art open LLMs and agents
- Write research papers and present results at leading international conferences. Wepublish in venues such as ICML, NeurIPS, and ICLR
- Interact with national and international teams of AI and machine learning experts, aswell as industrial collaborators
Your profile
Applicants must possess a Master's degree in computer science, mathematics, physics,or a related field. Candidates should have an exceptional academic record and a solidmathematical foundation. They are also expected to have strong coding andimplementation skills, with the ability to run large-scale experiments.
Experience with LLMs, reinforcement learning, generative modeling, or evolutionary andopen-ended methods is a plus. Research experience is advantageous but not mandatory.Above all, we are looking for curious, ambitious people who want to work on hard, openproblems.
Excellent communication skills and fluency in English (spoken and written) are required.
We also seek to increase the number of women and other groups in areas where they areunderrepresented and therefore explicitly encourage such groups to apply.