Applied Research Scientist

Generality Labs; Greater London; Híbrido; GBP 100 000 - 200 000; Tempo integral

Vantagens oferecidas por esta oferta de emprego

Visa sponsorship

Relocation assistance

London-based team

Resumo da oferta

Generality Labs in London seeks an Applied Research Scientist to scope and deliver R&D projects addressing AI evaluation challenges. You will move between deadline-driven projects with partners and self-directed research to advance evaluation science.

The role requires rigorous scientific reasoning, strategic prioritisation, and pragmatic problem-solving to deliver incremental value. Visa sponsorship and relocation support are available; the team is London-based with in-person expectations.

Qualificações

• Relentless commitment to scientific rigour and clear communication.
• Strategic foresight and prioritisation to solve the field's most pressing problems.
• Pragmatic and practical delivery to provide immediate value to the field.

Responsabilidades

• Scope and deliver R&D projects that solve key challenges in AI evaluation.
• Move between deadline-driven, project-based work and self-directed research.
• Be a rigorous methodologist and a pragmatic problem-solver on different days.
• Collaborate with partner organisations to ensure adoption of research outputs.

Conhecimentos

Scientific rigour

Strategic foresight

Pragmatic & practical

Descrição da oferta de emprego

• We’re building a team of the world’s most forward-thinking, ambitious scientists to ensure that humanity can measure the safety of advanced AI
• Successful candidates will join the founding team of a well-resourced, fast moving nonprofit start-up, at the centre of the AI safety ecosystem.
• Motivated by the urgency of the problems we’re solving, we focus on integrating research outputs into standards and tools which can be rapidly adopted by the organisations that need them - governments, safety orgs, and frontier labs

About us:

• Generality Labs is an R&D lab based in London
• We were founded in 2026, spinning off from Arcadia Impact with $4.4M in seed funding from Coefficient Giving
• Our goal is to support the field of AI in accurately evaluating risks and mitigations.
• We’re a team of 2 cofounders, and 5+ contractors who have previously worked at places like METR, CERN and Anthropic.
• Our mission is to build tools and standards which address key scientific challenges in AI evaluation, and then drive adoption of these solutions where they are most needed
• Our ultimate goal is to ensure the field can accurately evaluate the safety of advanced AI (which is currently not the case).
• Our flagship product, Inspect Evals, has been used by every major AI safety organisation.
• While we do sometimes publish, we prefer to drive adoption of our research proactively:
  • We consult directly with organisation, providing auditing services and advising internal teams on best practices
  • We build tools that incorporate best practices, and work closely with key organisations to ensure they are adopted
• After launching our research workstream 3 months ago, we’re unable to meet the significant demand from our partners & collaborators

Summary: as an applied research scientist, you will scope and deliver R&D projects that solve key challenges in AI evaluation.

We’re open to accommodating different preferences about the scope of the role. In general, you can expect to move between deadline-driven, project-based work with partner organisations, and self-directed research projects in which you genuinely advance the science of evaluation. This requires being a rigorous methodologist on some days, and a pragmatic problem-solver on other days.

Here’s an example of what the role might look like:

• Applied Research (50%): you identify problems that need to be solved, and then develop methodologies to solve them. Key findings are published, e.g. via a blog post and/or workshop paper, and/or delivered directly to partner organisations with a memo
• Research Implementation (25%): based on your research, you incorporate methodologies into tools to promote broader adoption. Our main tool for this purpose is Inspect Auditor, which is being developed to automate the analysis of evaluation quality
• Auditing & consulting (25%): you will work with our partner organisations to assist them in their efforts to evaluate AI safety. In some cases, this may involve audits of internal evaluations, where you will innovate new QA methodologies as required.

What we’re looking for:

• Relentless commitment to scientific rigour: This role demands significant skills in scientific reasoning and communication. You must be comfortable decomposing and questioning assumptions, and you must never be confidently incorrect. If you often find yourself appalled by the scientific rigour of research in the field, that is a great sign!
• Strategic foresight and prioritisation: you must be motivated by the need to solve the most pressing problems in the field, rather than by intellectual curiosity or popular trends. You will think from first principles and pre-empt the needs of the AI safety community.
• Pragmatic & practical: while building theories and elegant models is important, the urgency of the problem demands that we deliver as much value as soon as possible. This means being able to scope research projects to deliver incremental advancements that can be immediately adopted by the field
• Base compensation is 100-200k GBP, with a 5k productivity budget
• We’re able to sponsor visas, and we’re open to discussing relocation packages and support
• Our team is based in London; for this role we expect at least 3 days in-person per week

Why should I work for Generality Labs as opposed to the organisations that Generality partners with?

As a member of staff at Generality Labs, your role is to make all research across key organisations more impactful. While it isn’t possible to have the best researchers working in every single role where their skills are needed, it is possible to build an elite team of scientists which rolls out scalable tools & standards, which then uplift researchers across the field.

We’ve seen early evidence that this is the right approach. Organisations may specialise in a particular area, but the fundamental research methods are surprisingly similar across organisations, and yet best practices aren\'t adopted uniformly.

How do you make sure best practices are adopted?

Our mission is to define best practices and roll them out across the field. We do this with a carrot-stick method: adoption is streamlined with nice tools and forward deployed scientists (carrot), and incentivised with audits (stick)

What is an example of a research project that I might work on?

Eval awareness is clearly a key priority, and yet there is no definitive set of SOTA methods for measuring and mitigating eval awareness.

As an Applied Research Scientist, you could tackle this problem by:

• Stress-testing existing methods for measuring and mitigating eval awareness, identifying weaknesses and promising directions
• Advancing the SOTA through systematic experimentation and benchmarking
• Building a tool suite and skill file which is incorporated into Inspect Auditor, which then allows researchers to effortlessly measure eval awareness in their own experiments, with confidence that they are using the best methods available
• Working with forward-deployed scientists to ensure the tool is adopted by researchers, and gaining feedback from them to help you improve the methods you developed