Page 132 of 506 · Back to latest writing
Miles Brundage is an AI policy researcher focused on compute-based governance, expert red teaming, and tests for dangerous frontier-model capabilities. His work asks how verifiable systems and readiness planning can turn broad safety concerns into practical oversight.
Jonathan Frankle is Databricks’ Chief AI Scientist, a MosaicML co-founder, and co-author of the Lottery Ticket Hypothesis. Finding efficient subnetworks inside dense neural networks informs how he builds, evaluates, and scales machine learning systems without hype.
Stella Biderman is a mathematician, AI researcher, and EleutherAI’s Executive Director. Her work on public models, interpretability, and training dynamics argues that open access, rigorous evaluation, and alignment’s technical difficulties must be studied together.
Alex Ratner is Snorkel AI’s CEO and a University of Washington computer science professor. His work on programmatic weak supervision argues that reliable enterprise AI depends less on continually changing model architecture than on disciplined, scalable data curation.
Douwe Kiela is an AI researcher who co-invented Retrieval-Augmented Generation and co-founded Contextual AI. His work examines secure external data use, static benchmarks’ weaknesses, and the retrieval capabilities agentic systems need beyond controlled evaluations.
Dan Hendrycks is a machine learning researcher, Center for AI Safety leader, and MMLU co-author. He connects capability evaluation with catastrophic-risk policy, asking how alignment, evolutionary dynamics, and AI governance should address artificial superintelligence.
Aleksander Madry is an MIT computing professor studying reliable machine learning. He shows that adversarial vulnerabilities are not random glitches but non-human patterns models actively use, reframing what it takes to make AI secure in practice.
Jakub Pachocki is OpenAI’s Chief Scientist, who led development of GPT-4 and reinforcement learning systems for OpenAI Five. His journey from competitive programming to multi-step reasoning asks how deep learning can become a rigorous science for automating discovery.