building science on why models start to cheat: study RL dynamics, create model organisms, work towards scaling laws for scheming. Keywords: experience in empirical research; experience in RL training of LLMs; you have thought a lot and read about scheming and AI cognition in general; strong math background (e.g., in statistical physics); SWE/experience. Here you don't have to have everything from the list, but it's important to have cool experience in some of the points.

tl;dr: research scientist / research engineer, $135-270k + equity, relocation to London

Want to test frontier models before release? 👁

Here, Apollo is simply looking not only for engineers, but also for researchers. OpenAI/Anthropic/Deepmind send them new models a few weeks before release, and researchers check them for scheming: deception, hidden goals, sabotage attempts. And then their reports go into the system card and influence how and when the model will be released 😉

Apollo is doing very well right now: last year they officially partnered with OpenAI, CEO Marius Hobban was included in Time 100 AI, their research is in higher demand than ever, because of that.

They relocate to London, salary range is $135k-270k + equity, provide lunches and dinners at the office, unlimited vacation, and a budget for self-development. Write to Diana recruiter contact ❤️