NLML Research Engineer
Compassion Aligned Machine LearningRemote
Global catastrophic risksAI safetySoftware engineeringResearch
Why this is on the board
Whether a language model really behaves according to the values it is given during training is still hard to measure. In this role you run the experiments that test this on open models of around 12 billion parameters, and because you record every experiment reproducibly, other researchers in AI safety can check the results and build on them — even when an approach turns out not to work.
This board is in beta and this is our judgement. If you think we have it wrong, tell us. Give feedback · How we reason about it
About the role
Execute end-to-end LLM training experiments including midtraining, RL post-training, and evaluations to measure value alignment in models. - Run training pipelines for open-weight models (~12B scale): midtraining on synthetic documents, RL post-training, and evaluation under researcher guidance. - Build synthetic corpus generation pipelines with control conditions and debug training instabilities and infrastructure failures. - Keep experiments reproducible through careful configuration, check…
More about Compassion Aligned Machine LearningWhy this problemGlossary
Want to understand this first?
Start with our free course: about 100 minutes in six short units, at your own pace, with no sign-up. To go further, apply for the EA Fellowship: five weeks of reading and discussion in a small group. Also free, but selective.