Unverified50% confidenceFactExact time
OpenAI omitted key parameters such as model scale and training compute from the GPT-4 technical report, citing 'competitive considerations and safety factors.'
1
Sources
50%
Confidence
Long-term
Relevance
10/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOpenAI在InstructGPT论文中证明，少量高质量人类偏好数据对模型对齐的贡献远超大规模无监督训练74% similarUnverifiedOpen-source model safety should have safety properties embedded at the training stage, and relying solely on post-training RLHF alignment is insufficient.70% similarUnverifiedThe study ruled out simple explanations including model origin, insufficient training, backbone controls with matched budgets, and metric cycle controls.69% similarUnverifiedOpenAI在训练RLHF模型时面临标注者多样性问题，最终通过严格的标注者池管理来缓解69% similarUnverifiedThe safety alignment in the original training of open-source models can be removed through a small amount of fine-tuning.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/965992API
curl https://kongchang.com/api/v1/knowledge/claims/965992MCP
get_claim(id=965992)