I'm Hiring: Postdoc in AI + Biosecurity
I have a grant starting in September to set up a new AIxBio / biosecurity program, and I'm hiring a postdoc in AI biosecurity evaluation and study design
I have a grant starting in September to set up a program in AIxBio / biosecurity here at the University of Virginia School of Data Science. We’ll be evaluating how biology benchmarks relate to real world performance on biosecurity related tasks, along with some interesting model editing and interpretability research, working with our National Security Data and Policy Institute to turn our scientific findings into policy action. I’ll be working closely with three superstar faculty members here: Tom Hartvigsen (SDS), Chirag Agarwal (SDS), and Phil Potter (NSDPI). And I’m hiring!
Please see the job ad, description copied below. More may be on the way. The position is in-person, here in lovely Charlottesville, Virginia.1
Postdoctoral Research Associate, AI Biosecurity Evaluation & Study Design
About the Position
This position leads the design, execution, and analysis of a large paired human-subjects study that links AI biosecurity benchmark scores to performance on realistic biological protocol-design and troubleshooting tasks. The researcher reports to Associate Professor Stephen D. Turner and coordinates closely with co-investigators Tom Hartvigsen and Chirag Agarwal, and a laboratory partner. The role draws on study design, psychometric and statistical analysis, and management of participant data collection. The researcher will also contribute to a parallel machine unlearning effort, supporting measurement of how capability-mitigation methods affect both benchmark scores and real-world task performance.
Key Responsibilities
Lead day-to-day execution of the paired evaluation study, including benchmark administration, participant recruitment and scheduling, and coordination with the laboratory partner on protocol intake and outcome reporting
Develop scoring rubrics and instruments for protocol-design, feasibility, and troubleshooting tasks
Build and run mixed-effects and item-level analyses that estimate how benchmark scores predict real-world task performance, with quantified uncertainty
Produce the benchmark validity report and the benchmark-to-task crosswalk and calibration framework
Contribute to the machine unlearning and model editing components through safety-utility analyses, including within-subject comparisons of model variants on shared task batteries
Prepare manuscripts, preprints, and documentation for open release, and support IRB compliance and data quality
Minimum Qualifications
Doctoral degree (PhD or equivalent) in data science, statistics, biostatistics, psychometrics, quantitative psychology, a biological or biomedical science, or a related field, completed at the time of hire
Experience designing and analyzing human subjects studies, including power analysis and study design
Strong statistical programming in R or Python, including mixed-effects or hierarchical modeling
Working knowledge of molecular biology or microbiology sufficient to evaluate biological protocols and collaborate with laboratory scientists
Preferred Qualifications
Research experience in biosecurity, biodefense, or dual-use risk assessment
Familiarity with AI or large language model evaluation, benchmarking, or psychometric validation of model capabilities
Experience with machine learning tooling such as PyTorch or Hugging Face Transformers, and with model finetuning workflows
Prior work under IRB oversight and with controlled or sensitive data
Record of open-source software or reproducible-research practices
Experience coordinating academic-industry or multi-site research collaborations
I also have another job ad open, but will be closing this one soon: Postdoctoral Research Associate, Machine Unlearning and Model Editing for AI Biosecurity.
It really is lovely here. This little crunchy college town is a real gem in central Virginia. I moved here over 15 years ago from Hawaii (Kailua/Lanikai), paradise on earth if such a place ever existed. And I absolutely love it here. The cycling is world class. The trail running is also. 20 minutes from a national park, an airport in town with the bigger Washington Dulles airport 2 hours away, 2 hours from the coast, bike-friendly, progressive, and welcoming.