Jacob Steinhardt Portfolio
Founder & CEO, Transluce
UC Berkeley statistics faculty member (Associate Professor of Statistics and EECS on his own site; the department directory lists him as Associate Adjunct Professor), part of BAIR, and co-founder and CEO of Transluce, a nonprofit lab building open technology for understanding frontier AI systems; his site indexes publications on ML robustness, interpretability, and alignment alongside a "Bounded Regret" blog on AI safety research strategy.
Location
Didn't say. We're respecting the mystery.
Companies
Credentials
- PhD, Stanford University (advised by Percy Liang)
Skills
Notable
- Co-author of MMLU, "Measuring Massive Multitask Language Understanding" (11k+ Google Scholar citations)
- Co-author of "Concrete Problems in AI Safety" (2016)
- Former advisees include Dan Hendrycks (director of the Center for AI Safety) and Collin Burns and Alex Wei (both now at OpenAI), per his site
What makes it work
A breakdown of the choices that make this portfolio stand out.
A lab page that doubles as a mentorship record
Listing current student advisees with their own research profiles turns the homepage into a visible record of the researchers he's training, not just his own output.
Research strategy essays alongside technical publications
"AI Alignment Research Overview" and "Research as a Stochastic Decision Process" address how to choose and pursue research problems, a meta-level contribution beyond the object-level papers.
A second organization built to operationalize the research agenda
Co-founding Transluce (with Sarah Schwettmann) as a nonprofit lab extends the academic research (understanding and correcting model behavior) into an organization built specifically to pursue it at a scale beyond a university lab alone.
What AI Researchers can take from this
Specific, actionable tips to apply to your own portfolio — no generic advice.
List your students or collaborators and their own work on your site — it shows the research program you're building, not just your individual output.
Occasionally write about how you choose and evaluate research problems, not only the results — it's a distinct and valuable form of expertise.
If an academic research agenda needs more resources than a university lab provides, consider what organizational structure would let it scale.
Projects
A roster of students and where they went next sits on the homepage right next to the papers, so Jacob Steinhardt's research program shows up as much in the people as in the publications.
Transluce
A nonprofit research lab he co-founded in October 2024 to build open, scalable technology for understanding frontier AI systems; releases include Docent for analyzing long agent transcripts.
Shows the jump from running an academic lab to building an organization around a research agenda.
Co-founded with Sarah Schwettmann
Open tooling for model oversight
2025-2026 papers co-authored with Transluce researchers
MMLU: Measuring Massive Multitask Language Understanding
A multitask benchmark of knowledge and problem-solving questions across dozens of subjects, led by his then-advisee Dan Hendrycks.
An evaluation that became standard reporting for language models.
Co-authored with advisees
Followed by the MATH dataset paper
Evaluation work that fits his research on understanding model behavior
Bounded Regret
His blog on AI safety research strategy, forecasting, and oversight, alongside the Essays section on his homepage ('Research as a Stochastic Decision Process', 'AI Alignment Research Overview').
Writing about how to pick and pursue research problems is a distinct form of expertise.
'Film Study for Research' and 'Advice for Authors' posts
AI forecasting updates
Cross-posts from Transluce
More AI Researcher portfolios
Ready to build your portfolio?
Follow the AI Researcher roadmap — skills, projects, and timeline to get hired.