The DOI matches its DataCite record.
Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
Open accessOverview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
RLHF has led to improvements in LLMs' tendencies to hallucinate, but has not protected against it entirely.
“While the use of RLHF has led to improvements in LLMs’ tendencies to hallucinate, it has not protected against it entirely (Casper et al., 2023; Ouyang et al., 2022).”
From Helpful, harmless, honest? Sociotechnical limits of AI alignment and safety through Reinforcement Learning from Human Feedback (2025) · cites this paper as Casper et al., 2023
Casper et al. (2023) propose three categories of technical challenges of RLHF-collecting human feedback, training the reward model, and training the policy-and label them tractable or fundamental.
“Casper et al. (2023) provide a taxonomy for open problems and limitations of RLHF, proposing three categories of technical challenges: collecting human feedback, training the reward model, and training the policy. The challenges are further labelled as tractable and fundamental,”
From Helpful, harmless, honest? Sociotechnical limits of AI alignment and safety through Reinforcement Learning from Human Feedback (2025) · cites this paper as Casper et al., 2023
Citations
Cites
Still being checked.
Cited by
Sorted by citations
Dahlgren Lindström, A., Methnani, L., Krause, L., Ericson, P., de Rituerto de Troya, Í. M., Coelho Mollo, D., & Dobbe, R. (2025) · Ethics and Information Technology · doi:10.1007/s10676-025-09837-2 · Cited by 54
Dual-process theory and decision-making in large language models
Brady, O., Nulty, P., Zhang, L., Ward, T. E., & McGovern, D. P. (2025) · Nature Reviews Psychology · doi:10.1038/s44159-025-00506-1 · Cited by 17
Li, L., & Cao, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.25240
Li, L., & Cao, Y. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7143159
Metadata
- Authors
- Casper, S., Davies, X., Shi, C., Krendl Gilbert, T., Scheurer, J., Rando Ramirez, J., Freedman, R., Korbak, T., Lindner, D., Freire, P., Wang, T., Marks, S., Segerie, C.-R., Carroll, M., Peng, A., Christoffersen, P., Damani, M., Slocum, S., Anwar, U., Siththaranjan, A., Nadeau, M., Michaud, E. J., Pfau, J., Krasheninnikov, D., Chen, X., Langosco, L., Hase, P., Bıyık, E., Dragan, A., Krueger, D., Sadigh, D., & Hadfield-Menell, D.
- ORCID
- Venue
- ETH Zurich
- Publisher
- ETH Zurich
- Published
- 2023
- OpenAlex
- W6928748004
- Cited by
- 3
- References
- 0
- Citations by year
- 2026: 1 · 2025: 2
- FWCI
- 0.26
- Type
- journal article
- Language
- en
- Open access
- green
- hdl.handle.net
- Licence
- other-oa
- Locations
- Repository for Publications and Research Data (ETH Zurich) (submittedVersion, other-oa)
- ETH Zürich Research Collection
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Algorithms
- Social Sciences › Decision Sciences › Management Science and Operations Research › Advanced Bandit Algorithms Research
- Keywords
- reinforcement learning from human feedback, open problems, fundamental limits, human feedback
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
