Research Engineer/Scientist

SaferAI is seeking a Research Engineer/Scientist with a strong ability to perform technical research & software engineering. Ideal candidates will have experience both with conducting research on large language models and with traditional software engineering work. 

As a Research Engineer/Scientist, you will be responsible for the core technical work we conduct to advance quantitative risk modelling, i.e. the ability to quantify the probability of AI risk scenarios and their steps. By breaking down scenarios leading to harm in steps whose probability or quantity can be estimated, we intend to advance our understanding of AI risk and be able to systematically translate empirical indicators such as benchmarks, evaluations and incident reports into real-world risk estimates. The initial area of focus for development of our risk modeling methodology is assessing the impact of LLMs on cyber.

About SaferAI

‍SaferAI is a fast-moving, mission-driven organization advancing and promoting AI risk management to reduce AI risks, in particular extreme risks from advanced AI systems. We’re uniquely positioned at the intersection of technical research work and policy, ensuring that the knowledge we produce has an impact on AI risk management and the governance of LLMs. 

As the only NGO member of a G7 OECD taskforce in charge of writing a reporting framework for frontier AI, a founding member of the US AI Safety Institute Consortium and a primary contributor to EU standards and GPAI Code of Practice, we’re responsible for significant contributions to AI risk management of large language models in the policy realm.

Our technical work is key to maintaining and furthering the unique expertise we bring to governments and companies. We released the first AI risk management framework combining traditional risk management and frontier AI safety policies, cited by NVIDIA’s risk assessment work. We co-drafted the G42 risk management policy. We developed the first AI risk management rating system for AI companies’ risk management practices, featured in TIME and Euractiv, informing the decisions of major investors. 

Our current core research focus is to develop risk models that enable us to aggregate existing empirical measurements (benchmarks, evaluations, incident reports) and turn them into granular risk quantification. To render this scalable, we’re accelerating the process by utilising LLMs to complement expert Delphi studies and perform predictions. 

Responsibilities

Your core objective will be to perform research & engineering focused on rendering quantitative AI risk modelling more scalable for AI developers. Your responsibilities will include:

  • Developing LLM-powered risk assessment workflows to render risk assessment increasingly scalable and easily usable.
  • Building  the methodology and infrastructure to obtain valuable insights from risk models, e.g. by propagating uncertainties or enabling sensitivity analysis.
  • Constructing a risk-model informed benchmark as a proof of concept of how benchmarks can maximize their value of information when grounded in specific risk modeling.
  • Developing software to enable the development, update and communication of the risk models, and the integration of numerous existing publicly available empirical indicators that feed into the risk models.

Once our LLM for cyber risk modelling work matures, you will play a key role in our expansion into modeling risks related to misalignment and accidental harms. This will likely involve implementing or reimplementing control setups and evaluations to feed into our risk models in these areas.

We are excited about our team members shaping our medium-term research directions and we are keen to support and enable new research ideas that align with our mission. Here are a few examples of areas we’d like to explore in the medium-run: 

Skills and Experience Required
  • Detail-oriented and conscientious
  • Strong problem solving skills
  • A significant ML research record (e.g. have published several papers)
  • Programming and software development skills (e.g. as can be demonstrated by an extensive GitHub), including on libraries like Transformers or PyTorch
  • Technical and research paper writing abilities 

Nice to haves
  • Experience developing applications or enhanced workflows with LLMs
  • Advanced statistical proficiency 
  • High degree of creativity
  • Strong LLM tinkering abilities 

Working Conditions
Location: We have a central office in Paris. We also offer remote working options.

Wage Range: For U.S.-based candidates, the wage range is $65,000-90,000. For candidates based outside the U.S., $65,000-80,000.

Benefits: 
  • Health insurance coverage and retirement plans adapted to the location
  • Transportation home to work covered at 50% 
  • Productivity expenditures up to $2k annually
  • Office space if relevant
How to Apply

To apply for this position, please complete this application form. We will evaluate the candidates on a rolling basis starting now until we fill the role. We hope to fill the role by mid April. 

We encourage you to lean towards applying, even if you don’t have all the skills and experience required. 

If you have any questions or concerns throughout the application process, please don't hesitate to reach out to us at careers@safer-ai.org. We look forward to reviewing your application.