Postdoc: AI Biosecurity - Unlearning & Model Editing

University of Virginia; Charlottesville (VA); Sur place; USD 60 000 - 75 000; Plein temps

Avantages offerts par ce poste

UVA Health Plan

Vision Coverage

Dental Plan

Benefit Savings Plans

Life Insurance

Disability Benefits

Paid Time Off

Résumé du poste

The University of Virginia School of Data Science seeks an accomplished researcher to develop and evaluate machine unlearning and model editing methods. You will implement, compare, and benchmark techniques that reduce hazardous biological capabilities in AI systems while preserving useful scientific functions.

Reporting to Assistant Professor Tom Hartvigsen, you will design experiments, lead the open UBS-Bio evaluation suite, contribute to interpretability studies, and publish reproducible

Qualifications

• PhD in data science, CS, ML or related field completed by hire.
• Strong Python programming and experience with ML frameworks like PyTorch and Hugging Face.
• Proven publications in ML/NLP and/or biosecurity.
• Experience training, finetuning, or post-training for large language models.
• Software engineering practices for reproducible and reusable research tools.

Responsabilités

• Implement and compare machine unlearning and model editing methods.
• Design experiments measuring safety–utility tradeoffs.
• Develop adversarial testing protocols with the lab partner.
• Lead development of the UBS-Bio evaluation suite.
• Support interpretability analyses of model representations.
• Publish manuscripts and release reproducible code.

Connaissances

Python

PyTorch

Hugging Face Transformers

LLM fine-tuning / post-training

Reproducible research tooling

Publications in ML/NLP/biosecurity

Formation

Doctoral degree (PhD or equivalent) in data science, CS, ML or related field

Outils

Description du poste

The University of Virginia School of Data Science seeks an accomplished researcher to develop and evaluate machine unlearning and model editing methods. You will implement, compare, and benchmark techniques that reduce hazardous biological capabilities in AI systems while preserving useful scientific functions.

Reporting to Assistant Professor Tom Hartvigsen, you will design experiments, lead the open UBS-Bio evaluation suite, contribute to interpretability studies, and publish reproducible