Reading Group: Steering "AI Scientists" Toward Valid & Trustworthy Research
Join the Snorkel AI Reading Group, a recurring forum to explore the latest frontier developments in AI while building meaningful connections within the community.
AI scientist systems can now generate hypotheses, run experiments, and write papers at unprecedented speed and scale, but their output ranges from real contributions to papers with invalid experiments, hallucinated citations, and unsupported claims. Without proper evaluation, valuable work gets buried, errors spread, and trust in the scientific record erodes.
In this virtual webinar, Oxford University Senior Postdoc Dr. Harit Vishwakarma addresses this evaluation gap by presenting Paperena: a long-horizon environment for evaluating “AI scientists” by testing agents on realistic, extended research cycles.
In this shared arena where AI-generated research is recorded, verified, and reviewed – with feedback flowing back to the AI scientists themselves – evaluation becomes something agents can learn from to steer them toward valid, trustworthy research.
Agenda:
12:00 pm - webinar opens, welcome remarks
12:05 pm - talk begins
12:30 pm - open Q+A and discussion
Paperena will be part of the AI for Science Workshop at NeurIPS, 2026 and has been presented at research convenings, such as Frontier Data Summit (Snorkel AI), San Francisco and the AI Scientist Summer Workshop (Microsoft Research), New England.
Dr. Harit Vishwakarma is a Senior Postdoctoral Research Fellow at the University of Oxford, where he researches the fundamentals and reliability of modern AI systems.