--- title: "Yuntao Bai — author profile and papers" description: "Yuntao Bai, Xiamen University: 2 papers on Elamur Research (2022), h-index 26 — affiliations, topics, co-authors and who cites them in our library." canonical: "https://research.elamur.ai/free/en/authors/yuntao-bai/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/authors/yuntao-bai/) [home](https://md.research.elamur.ai/free/en/) / [authors](https://md.research.elamur.ai/free/en/authors/) / Yuntao Bai # Yuntao Bai Yuntao Bai works on the safety of large language models, in particular on making conversational assistants less likely to give harmful answers. Both papers here, from 2022, treat harmlessness as a problem of training and evaluation. Models are probed with adversarial prompts, and their responses are either measured or used as a signal for further training. The work falls under [Bias, fairness & safety](https://md.research.elamur.ai/free/en/topics/bias-fairness-safety/) and [Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/). The most cited paper here is [Constitutional AI: Harmlessness from AI Feedback](https://md.research.elamur.ai/free/en/papers/bai-2022-constitutional-ai-harmlessness-ai-feedback/) (2022). It replaces human judgements of harm with a short set of written principles. A helpful-only model critiques and rewrites its own answers to provocative prompts and is then fine-tuned on the rewritten versions. In a second stage, a preference model built from the model's own comparisons of paired answers guides reinforcement learning. The companion study, [Red Teaming Language Models to Reduce Harms](https://md.research.elamur.ai/free/en/papers/ganguli-2022-red-teaming-language-models-reduce/) (2022), had crowdworkers try to draw offensive or dangerous replies out of an assistant over multi-turn chats. It compared plain, prompted, rejection-sampling and RLHF models at three sizes. As size increased, the RLHF models alone grew more resistant to these attempts. Both papers come from Anthropic teams, and the two lines of work connect: red-teaming prompts serve as the starting material for the first stage of the constitutional method. Co-authors on the papers here include [Amanda Askell](https://md.research.elamur.ai/free/en/authors/amanda-askell/), [Andy Jones](https://md.research.elamur.ai/free/en/authors/andy-jones/), [Jackson Kernion](https://md.research.elamur.ai/free/en/authors/jackson-kernion/), [Saurav Kadavath](https://md.research.elamur.ai/free/en/authors/saurav-kadavath/), [Deep Ganguli](https://md.research.elamur.ai/free/en/authors/deep-ganguli/) and [Ethan Perez](https://md.research.elamur.ai/free/en/authors/ethan-perez/). Bylines from 2022 to 2024 also associate the name with the Machine Intelligence Research Institute in Berkeley and the [University of Oxford](https://md.research.elamur.ai/free/en/universities/university-of-oxford/). Yuntao Bai has 2 papers in our library, in 2022, most of them on [Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/), [Bias, fairness & safety](https://md.research.elamur.ai/free/en/topics/bias-fairness-safety/); often with [Amanda Askell](https://md.research.elamur.ai/free/en/authors/amanda-askell/), [Andy Jones](https://md.research.elamur.ai/free/en/authors/andy-jones/), [Jackson Kernion](https://md.research.elamur.ai/free/en/authors/jackson-kernion/). 2 papers in our library · 2022 [OpenAlex](https://openalex.org/A5091860006) · [ORCID](https://orcid.org/0000-0003-3998-7837) · [**h-index 26**](https://openalex.org/authors/A5091860006) · [59 works](https://openalex.org/works?filter=author.id:A5091860006) · [3,058 citations](https://openalex.org/works?filter=author.id:A5091860006&sort=cited_by_count:desc) OpenAlex, 2026-10-14 [📄 2 works in our library](https://md.research.elamur.ai/free/en/authors/yuntao-bai/#papers) · [🔗 4,322 citations in our library](https://md.research.elamur.ai/free/en/authors/yuntao-bai/cited-by/) · [4 more in our library without a page yet](#also) · [⭐ Paper Star rank #1565](https://md.research.elamur.ai/free/en/paper-stars/page/16/#star-1565) The profile is the library's match of the name; two people with one name are not always told apart. Elsewhere: [Google Scholar](https://scholar.google.com/citations?view_op=search_authors&mauthors=Yuntao+Bai) · [Semantic Scholar](https://www.semanticscholar.org/search?q=Yuntao+Bai) · [dblp](https://dblp.org/search?q=Yuntao+Bai) · [ACL Anthology](https://aclanthology.org/search/?q=Yuntao+Bai) ## Affiliations [🎓**Xiamen University** Xiamen · China 2011–2026 · 38 papers*current*](https://md.research.elamur.ai/free/en/universities/xiamen-university/) · [🎓**University of Oxford** Oxford · United Kingdom 2023–2024 · 2 papers](https://md.research.elamur.ai/free/en/universities/university-of-oxford/) · [🤝**Machine Intelligence Research Institute** Berkeley · United States 2022–2023 · 2 papers](https://ror.org/01h4ass90) From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 2 more institutions on single bylines are not listed. ## Topics [Language models & pretraining (1)](https://md.research.elamur.ai/free/en/topics/language-models/) · [Bias, fairness & safety (1)](https://md.research.elamur.ai/free/en/topics/bias-fairness-safety/) ## Works by year ### 2020s [**2022** 2 papers](https://md.research.elamur.ai/free/en/authors/yuntao-bai/year/2022/) ## Papers The most cited first (OpenAlex and our library). 1. **[Constitutional AI: Harmlessness from AI Feedback](https://md.research.elamur.ai/free/en/papers/bai-2022-constitutional-ai-harmlessness-ai-feedback/)** *(2022 · 957 citations)* The paper looks for a way to train a language assistant to avoid harmful responses without relying on human labels that mark harm. 2. **[Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned](https://md.research.elamur.ai/free/en/papers/ganguli-2022-red-teaming-language-models-reduce/)** *(2022 · 456 citations)* Large language models can produce many kinds of harmful content. This Anthropic study examines manual red teaming, in which people deliberately try to provoke bad outputs, as a way both to measure and to reduce such behavior. Crowdworkers held open-ended, multi-turn chats with an AI assistant, tried to steer it… ## Also in our library 4 more works of theirs are in our library without a page here yet; they will appear as we publish them. ## Co-authors here 12 of 16 co-authors with a page, the most papers together first. - [Amanda Askell](https://md.research.elamur.ai/free/en/authors/amanda-askell/) ×2 - [Andy Jones](https://md.research.elamur.ai/free/en/authors/andy-jones/) ×2 - [Jackson Kernion](https://md.research.elamur.ai/free/en/authors/jackson-kernion/) ×2 - [Saurav Kadavath](https://md.research.elamur.ai/free/en/authors/saurav-kadavath/) ×2 - [Anna Chen](https://md.research.elamur.ai/free/en/authors/anna-chen/) - [Anna Goldie](https://md.research.elamur.ai/free/en/authors/anna-goldie/) - [Azalia Mirhoseini](https://md.research.elamur.ai/free/en/authors/azalia-mirhoseini/) - [Ben Mann](https://md.research.elamur.ai/free/en/authors/ben-mann/) - [Carol Chen](https://md.research.elamur.ai/free/en/authors/carol-chen/) - [Catherine Olsson](https://md.research.elamur.ai/free/en/authors/catherine-olsson/) - [Deep Ganguli](https://md.research.elamur.ai/free/en/authors/deep-ganguli/) - [Ethan Perez](https://md.research.elamur.ai/free/en/authors/ethan-perez/) ## Cited by Papers of our library cite their works 4,322 times (of the 140,884 papers whose bibliographies we read); [296 citing papers with a page here →](https://md.research.elamur.ai/free/en/authors/yuntao-bai/cited-by/).