Member of Technical Staff

AI Safety Research * Cambridge, United Kingdom Geodesic *Research* The shortest path to impact About Us We build *robust initialisations for alignment* —base models shaped through pre- and midtraining to hold up under capabilities reinforcement learning. Geodesic Research is a UK-based technical AI safety organisation focused on compute-intensive alignment research. Our seminal work on alignment pretraining showed that you can bake alignment priors into base models, and the broader field is now converging on the approach. Long-horizon capabilities reinforcement learning is emerging as a critical and underexplored threat to alignment—degrading alignment properties across evals and selecting for behaviours like metagaming, sycophancy, and reward hacking. Our agenda is to design midtraining and early post-training interventions that create initialisations where alignment persists through the rest of training. Research Taste 01 Conceptually Simple We focus on conceptually simple, data-centric interventions—document mixes, filtering, declarative midtraining—that benefit from scale and slot into existing training pipelines without bespoke infrastructure. 02 Uniquely Positioned Philanthropic funding from Coefficient Giving and a compute partnership with the UK AI Security Institute put us among the few non-lab actors who can replicate the full midtraining, SFT, and RL stack at scale. We have no commercial stake in any particular alignment method, leaving us free to investigate the full picture. 03 Frontier Impact Our target audience is model training teams at frontier labs. We design interventions that can be profiled, packaged, and handed off—taking the shortest path to advising on their production training stacks. Research Directions Can alignment hold up through capabilities RL? 01 Alignment Pretraining Our seminal work showed that AI discourse causes self-fulfilling (mis)alignment—and that you can shape these priors during pretraining. Frontier labs are now converging on this approach: Anthropic's recent Teaching Claude Why and Model Spec Midtraining both lean on the alignment-priors framing we pioneered. Learn more 02 Misalignment Quarantining Post-training on imperfect data can broadly corrupt a base model's alignment. We will explore shaping base models with declarative midtraining documents that aim to establish an explicit context boundary around unsafe behaviour—so that benign features might generalise while the misalignment underneath stays quarantined. 03 Adversarial Robustness to Capabilities RL Long-horizon capabilities RL may degrade alignment in ways pretraining alone cannot prevent. We will stress-test midtraining and early post-training interventions against agentic RL with misspecified rewards—aiming to surface which methods could produce truly robust initialisations and which might break down under pressure. Key Papers Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment Read paper Blog Posts Announcing Geodesic Research Read post The Team Founded in Cambridge, UK Cameron Tice Co-Founder \& Executive Director Marshall Scholar at the University of Cambridge, where he completed his MPhil on automated research with LLMs for computational psychiatry. A former Research Manager for the ERA:AI fellowship. Puria Radmard Co-Founder \& Technical Director Former ERA:AI fellow and University of Cambridge PhD student. Previously a machine learning engineer at raft.ai and a private equity quantitative strategist at Goldman Sachs. Alexandra Narin Head of Operations Cofounder of UK AI Forum. Previously, a Experimental Neuroscience researcher at UCL and the Head of Grants for a Biotech company. Edward Young Founding Member of Technical Staff Former researcher on AISI's Safeguards team and ERA:AI Fellow. Completed a Computational Neuroscience PhD at the University of Cambridge. Kyle O'Brien Founding Member of Technical Staff Leads the alignment pretraining research agenda and has developed strong relationships with UK AI Security Institute through previous research on Deep Ignorance. Previously at EleutherAI and Microsoft. Nathalie Kirch Member of Technical Staff PhD student in computer science at Imperial College London and King's College London, researching mechanistic interpretability and robustness in LLMs. Previously a MATS Research Scholar, LASR fellow, and ERA:AI fellow. Mentors Guided by leading researchers Alex Turner Google DeepMind Tomek Korbak OpenAI Alex Cloud Anthropic David Demitri Africa UK AI Security Institute Join Us Help build the future of AI safety Geodesic Is Hiring Four Additional Members Of Technical Staff. We're looking for technical staff with experience across the ML and alignment research stack: multi-GPU / HPC training and evals experience, deep familiarity with data-centric alignment methods, and an *insatiable desire to improve the outcomes of developing superintelligence.* If this sounds like you, please apply. Job Details Apply Now