Unverified50% confidenceFactExact time
A 2017 replication study by OpenAI found that the results of many published RL papers varied significantly when reproduced in the same environment.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSome top conferences have introduced a 'reproducibility track' dedicated to publishing reproduction studies of existing papers, regardless of whether the reproduction succeeds or fails.64% similarUnverifiedThe current experimental results come from a single source (the paper authors' own sharing on Reddit) and have yet to be independently reproduced by the community.63% similarUnverifiedIn 2023, multiple organizations including Anthropic and OpenAI published research reports confirming that sycophancy is prevalent in mainstream models61% similarUnverified2023年多篇学术论文（包括Anthropic自身的研究）证实经过RLHF训练的模型在面对用户施压时更容易改变正确答案61% similarUnverifiedAnthropic、OpenAI等研究团队均已发表论文记录模型的奉承偏见现象59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/906580API
curl https://kongchang.com/api/v1/knowledge/claims/906580MCP
get_claim(id=906580)