Careers
We are sixty people across two offices, and we hire a few at a time. Most of us split our weeks between running experiments and arguing about what they mean. We pay at the top of the Edinburgh and Montréal markets, publish under our own names, and do not ask anyone to work weekends to hit a release date.
5 open roles
01 · Alignment
Edinburgh · Full-time
Design training objectives and evaluations for models that hold correct answers under pressure and abstain when they should. You will own part of the Steadfast and Lantern suites and the experiments that feed them.
Apply by email →02 · Interpretability
Montréal · Full-time
Build the infrastructure that lets researchers train sparse dictionaries on billions of activations and browse what they find. Strong systems engineering matters more here than prior interpretability work.
Apply by email →03 · Reasoning
Edinburgh or Montréal · Full-time
Write evaluations that catch quiet failures in long documents: answers that sound right and are not. You will extend LongTrace to messier, real-world archives and work out how to grade synthesis, not just retrieval.
Apply by email →04 · Policy
Edinburgh, hybrid · Full-time
Help turn our evaluation practice into commitments others can audit, starting with held-out test sets and disclosure norms. You will write for regulators and researchers, and sometimes for both at once.
Apply by email →05 · Infrastructure
Montréal · Full-time
Keep our training runs reproducible: data lineage, shard hashing, checkpoint provenance, and the tooling behind our commitment not to train on Commons-Held. Experience with large distributed jobs is expected.
Apply by email →A person reads every application and replies within three weeks, whatever the answer. If you need adjustments to any part of the process, say so in your email.