Profile banner
Herbie Bradley

Herbie Bradley

PhD researcher and ML practitioner on LLMs, synthetic data, and AI policy

herbiebradley
San Francisco, United States
Joined January 2025
Favicon imageherbiebradley.com
Favicon imagex.com
Favicon imagescholar.google.com
Favicon imageorcid.org
Favicon imagehuggingface.co
Favicon imagewandb.ai
Favicon imagebennettschool.cam.ac.uk
Favicon imagemlg.eng.cam.ac.uk

Network

5.3K connections
🎓
Cambridge AI Researchers
🛡️
AI Security Institute
🧠
AI Safety Researchers
🌍
Environmental AI Climate
🇬🇧
UK AI Policy Makers
⚕️
Biosecurity Health AI
🏛️
...

Summary

Technical LLM research and open‑science contributions: Herbie is an active ML researcher focusing on large language models, synthetic data, and open‑endedness. He contributed to community-driven projects (EleutherAI's Pythia), co-authored multiple conference papers, and has published work on tools and analyses (e.g., Pythia, QDAIF) that enable reproducible study of model training and capabilities. mlr+2
Bridging technical research and AI policy: He has moved between technical research and government-facing roles, helping to stand up the UK AI Security Institute and joining the Frontier AI Taskforce. In those positions he worked on model evaluations, summit/meeting organisation (the UK AI Safety / Bletchley summit), and translating technical capabilities into governance recommendations. herbiebradley+2
Community-building, mentorship and organisershp: Herbie has co-organised workshops and fellowships (e.g., synthetic data workshop at ICLR, ERA/Cambridge fellowship mentoring and Cambridge summer fellowship activity), and he mentors early-career researchers working on AI risks and policy-relevant technical problems. erafellowship+1
Public-facing commentary and synthesis: Beyond papers, he writes longer essays and policy pieces (e.g., "Glimpses of AI Progress" and "The Great Refactor") aimed at policymakers and the broader AI community, indicating strengths in translating technical ideas into accessible strategic advice. pathwaysai+1

Work

Education

Projects

Writing

The Great Refactor

August 1, 2025

Essay / policy-oriented piece (Institute for Progress) proposing a large-scale, practical program to translate critical C/C++ libraries to memory-safe languages (Rust) as a cyberdefense strategy.

Favicon imageherbiebradley.com

Glimpses of AI Progress

March 1, 2025

Long-form essay on AI Pathways analyzing drivers of AI progress in 2025, mental models for capability development, agents vs chatbots, distillation, and implications for policy and economics.

Favicon imagepathwaysai.org

Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling

January 1, 2023

Conference paper introducing Pythia: a suite of 16 LLMs and many checkpoints to study training dynamics and scaling behavior across model sizes; released code and checkpoints to support reproducible research.

Favicon imageproceedings.mlr.press

EleutherAI: Going Beyond "Open Science" to "Science in the Open"

October 1, 2022

Workshop/paper describing EleutherAI's experience doing public-facing, openly collaborative ML research and reflecting on benefits and pitfalls of research in the open.

Favicon imagearxiv.org

AI Strategy for a New American President

January 1, 2025

Essay analyzing likely directions for U.S. AI policy after the 2024 election, arguing a 'vibe shift' toward securitization, defense-focused uses, export controls on model weights, and implications for governance and research.

Favicon imagepathwaysai.org

Hobbies

Public writing and blogging on AI trajectories and policy (Substack / Pathways / personal blog). herbiebradley+1

Active on social media (X) commenting on AI, technical topics, and community events. x

Notes

SYShannon Yangsaid they've seen
HBHerbie Bradley
build meaningful solutions to complex problems, even outside traditional roles or organizations.

Jan 8, 2026