Member of Technical Staff
P-Zero Research
Member of Technical Staff
P-Zero Research
AI research benchmarks measuring model taste
P-Zero Research is a public benefit corporation working to improve the long term trajectory of artificial intelligence. We aim to forecast and mitigate the risks of automated AI R&D.
Our recent work has involved improving takeoff forecasts by measuring the research taste of frontier models. Going forward, we’re improving our understanding of forecasts and risks of recursive self-improvement through studying strategic research taste and automated cyber uplift. As frontier labs move closer to automating their own research, we expect independent measurements and forecasts to become one of the main tools for pacing the frontier safely.
Members of Technical Staff own a research project end to end. You would reason carefully about which capabilities are important for RSI and where existing takeoff models are least constrained by evidence: turning that reasoning into hypotheses and then into large-scale experiments that you execute. Depending on your expertise you may focus more on building infrastructure, running preliminary experiments, doing conceptual thinking or communicating results.
In this role you will:
Conduct ambitious experiments to measure automated AI R&D capabilities
Identify new threat vectors from automated AI R&D and how to measure them
Run scrappy prototypes and rapidly iterate
Build robust, production-grade infrastructure
Update forecasters, policy makers and the AI safety community about the risks of RSI
People will thrive in this role with:
Excellent strategic taste and research taste: you can prioritise effectively in domains with weak feedback loops
High agency and conscientiousness
Comfort with ambiguity and adapting fast
Strong, fast software engineering skills and concise written communication
High attention to detail: you actually read the datasets
Alignment with P-Zero's mission to improve the long-term trajectory of AI
Bonus points for:
Frontier-lab or METR/AISI-adjacent expertise
Prior research experience in evals
Being based in, or able to relocate to, London or the San Francisco Bay Area
We prefer people to work in person in London or the San Francisco Bay Area. We're open to remote for exceptional candidates, with regular overlap during our core hours (8-10am Pacific / 4-6pm UK).
We offer a relocation allowance for moves to London or the San Francisco Bay Area.
San Francisco Bay Area: we sponsor US work visas (typically O-1A, E-3 or TN) and cover legal and filing fees, including for dependants. Sponsorship isn't guaranteed for every candidate, but if we make you an offer we'll work with you and our immigration counsel to find the right route.
London: we can't sponsor UK visas at the moment, so you'll need an existing right to work in the UK or a visa you hold in your own right, such as a Global Talent visa.
AI research benchmarks measuring model taste
P-Zero Research builds benchmarks to measure AI models’ ability to conduct research and make scientific judgments, centered on its TasteVal tool. TasteVal evaluates how well a model can design experiments and interpret results, yielding a score based on the computational effort needed to reach expert-level performance. The company distinguishes itself by measuring experimental design quality and interpretation, not only task accuracy, and by using computational efficiency as a key metric. Its aim is to provide concrete, comparable benchmarks that track progress in AI research capabilities and safety, with implications for future self-improvement.
Explore more opportunities at P-Zero Research