The CHIL 2026 Doctoral Symposium is back! Apply by March 13th 📅
chil.ahli.cc/submit/docto...
Last year, we welcomed 28 outstanding PhD researchers for mentorship and lightning talks in health AI.
Watch 3 participant talks from 2025 👇 (see 2:48:43)
www.youtube.com/watch?v=YaDo...
7/ Materials:
Course website / syllabus / notes: web.stanford.edu/class/cs329h/
Living Textbook: mlhp.stanford.edu
#AIAlignment #MachineLearning #ResponsibleAI #Stanford #PreferenceLearning
6/ Tangent Space Fine-Tuning for Directional Preference Alignment (TS-DPO) — Mete Erdoğan
Multi-objective alignment. Learns separate update directions then composes them at inference via simple coefficient mixing → smoother control + broad Pareto coverage
github.com/meterdogan07...
5/ Gaussian Process Optimization with Predictions for Hypothesis Generation (PA-GP-UCB) — Stela Tong & Xin (Jennifer) Chen
Blends human feedback with AI surrogate predictions, proves O(√T) cumulative regret, and shows much faster convergence than GP-UCB
drive.google.com/file/d/1JpuM...
4/ Learning Code Generation “Vibe” for MCP Servers via DPO — Rami Ratl Mrad
Preference learning for subjective code quality beyond mere correctness. 100% AI-judge preference and 85% human preference, with better error handling, docs, and structure for MCP
drive.google.com/file/d/1g_r_...
3/ Silicon Sampling of Professional Philosophers — Jeremy Shi
Can LLMs authentically simulate expert philosophical reasoning when conditioned on philosopher profiles? Finding: LLMs show inflated demographic→position correlations (up to 3× human data).
whusym.github.io/silicon-phil...
2/ Active Alignment with Bayesian General Preference Model (BGPM) — Ahmed Mohsin & Muhammad Umer.
Enables active preference elicitation with >35% fewer annotations via uncertainty-guided, information-theoretic query selection.
drive.google.com/file/d/1IU5e...
1/ Wonderful student projects from CS329H (Fall ’25) ML from Human Preferences at Stanford University! 🚀
Sang Truong, Andy Haupyt, and I introduced students to preference learning + alignment, culminating in final projects. Out of ~50, here are 5 standouts 👇
Our Responsible AI group is hiring at GSK! Join a great team investigating how to responsibly develop AI for drug discovery in an environment mixing research and real-world impact. Full-time in London. jobs.gsk.com/en-gb/jobs/4...
We want your work on fairness, alignment, and/or agentic systems!!
Proud to be co-organizing AFAA @iclr-conf.bsky.social with:
@prakharg.bsky.social @adoubleva.bsky.social @Miriam @Jamelle @Golnoosh @jessicaschrouff.bsky.social @sanmikoyejo.bsky.social
www.afciworkshop.org
#AFAA2026 #ICLR2026
This is work with Daniel E. Ho and @sanmikoyejo.bsky.social. We also have a related preprint with Erin Beeghly on the social impacts of this personalization, and how it interacts with group-based preferences and stereotypes: angelina-wang.github.io/files/person...
This has huge implications for evaluation:
• Benchmark scores ≠ what end users actually experience
• Some high-risk behaviors, e.g., manipulative patterns, might only surface in personalized interfaces
We argue for more realistic evals that take personalization into account.
In fact, even the same MMLU science question can yield different ChatGPT answers for different users, despite using the exact same underlying model. User-level personalization and interaction patterns shape outputs in ways existing evals do not capture.
The personalization gap challenges an implicit assumption in AI evaluation: that we can measure capability and safety independently of deployment context. Good opportunity to rethink how we evaluate AI systems.
New in @science.org : 20+ AI scholars—inc. @alondra.bsky.social @randomwalker.bsky.social @sanmikoyejo.bsky.social et al, lay out a playbook for evidence-based AI governance.
Without solid data, we risk both hype and harm. Thread 👇
Grateful to win Best Paper at ACL for our work on Fairness through Difference Awareness with my amazing collaborators!! Check out the paper for why we think fairness has both gone too far, and at the same time, not far enough aclanthology.org/2025.acl-lon...
Instead, we should permit differentiating based on the context. Ex: synagogues in America are legally allowed to discriminate by religion when hiring rabbis. Work with Michelle Phan, Daniel E. Ho, @sanmikoyejo.bsky.social arxiv.org/abs/2502.01926
Most major LLMs are trained using English data, making it ineffective for the approximately 5 billion people who don't speak English. Here, HAI Faculty Affiliate @sanmikoyejo.bsky.social discusses the risks of this digital divide and how to close it. hai.stanford.edu/news/closing...
Our latest white paper maps the landscape of large language model development for low-resource languages, highlighting challenges, trade-offs, and strategies. Read more here: hai.stanford.edu/policy/mind-...
Collaboration with a bunch of lovely people I am thankful to be able to work with: @hannawallach.bsky.social , @angelinawang.bsky.social , Olawale Salaudeen, Rishi Bommasani, and @sanmikoyejo.bsky.social. 🤗
🧑🔬 Happy our article on Creating a Generative AI Evaluation Science, led by @weidingerlaura.bsky.social & @rajiinio.bsky.social, is now published by the National Academy of Engineering. =) www.nae.edu/338231/Towar...
Describes how to mature eval so systems can be worthy of trust and safely deployed.
📣We’re thrilled to announce the first workshop on Technical AI Governance (TAIG) at #ICML2025 this July in Vancouver! Join us (& this stellar list of speakers) in bringing together technical & policy experts to shape the future of AI governance! www.taig-icml.com
AI systems present an opportunity to reflect society's biases. “However, realizing this potential requires careful attention to both technical and social considerations,” says HAI Faculty Affiliate @sanmikoyejo.bsky.social in his latest op-ed via @theguardian.com: www.theguardian.com/commentisfre...
Very excited we were able to get this collaboration working -- congrats and big thanks to the co-authors! @rajiinio.bsky.social @hannawallach.bsky.social @mmitchell.bsky.social @angelinawang.bsky.social Olawale Salaudeen, Rishi Bommasani @sanmikoyejo.bsky.social @williamis.bsky.social
3) Institutions and norms are necessary for a long-lasting, rigorous and trusted evaluation regime. In the long run, nobody trusts actors correcting their own homework. Establishing an ecosystem that accounts for expertise and balances incentives is a key marker of robust evaluation in other fields.
which challenged concepts of what temperature is and in turn motivated the development of new thermometers. A similar virtuous cycle is needed to refine AI evaluation concepts and measurement methods.
2) Metrics and evaluation methods need to be refined over time. This iteration is key to any science. Take the example of measuring temperature: it went through many iterations of building new measurement approaches,
Just like the “crashworthiness” of a car indicates aspects of safety in case of an accident, AI evaluation metrics need to link to real-world outcomes.
We identify three key lessons in particular.
1) Meaningful metrics: evaluation metrics must connect to AI system behaviour or impact that is of relevance in the real-world. They can be abstract or simplified -- but they need to correspond to real-world performance or outcomes in a meaningful way.
We pull out key lessons from other fields, such as aerospace, food security, and pharmaceuticals, that have matured from being research disciplines to becoming industries with widely used and trusted products. AI research is going through a similar maturation -- but AI evaluation needs to catch up.
