NLProject Tailwind: AI Governance & Policy Research
Coefficient GivingRemote
Global catastrophic risksAI safetyResearchSoftware engineering
Why this is on the board
A good way to test AI models only has to be worked out once: after that, other labs and regulators can adopt it without redoing the work themselves. You research how to tell whether a model is developing unwanted behaviour, and summarise it for developers and policymakers who use such findings because they cannot work it out themselves.
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
Multiple research institutions tackling AI safety through independent oversight, evaluation, and foundational research on model behaviour and alignment. - Develop monitorability benchmarks and audit frontier model chains-of-thought to detect degradation over time. - Build rigorous evaluation methodologies for AI character, behavioural tendencies, and adherence to developer specifications. - Establish large-scale training pipelines to systematically identify which procedures induce misalignment…
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.