Unverified50% confidenceFactExact time
The open-source code of the Quiet-STaR paper (Eric Zelikman et al., 2024) uses direct fine-tuning with bf16 weights and the AdamW optimizer and may have inadvertently triggered the weight-freezing phenomenon.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/2/2026
First Seen
Valid until: 1/31/2027
Sources
Related Entities
Related Claims
UnverifiedGPTQ was proposed by Frantar et al. in 2022 and minimizes quantization error based on the inverse of the Hessian matrix; AWQ was proposed by Ji Lin et al. at MIT, with the core insight that a small number of weight channels associated with large activation values should retain higher precision.61% similarUnverified收缩自编码器通过惩罚编码器Jacobian矩阵的Frobenius范数,迫使模型对输入的微小扰动不敏感60% similarUnverifiedIn the CS231n assignments, students get stuck when deriving the softmax cross-entropy gradient because it involves the coupling of the Jacobian with probability normalization; implementing the batchnorm backward pass tests an even finer grasp of the chain rule.60% similarUnverified2024年以来,研究者发现了对抗性后缀攻击(Adversarial Suffix),即在正常提示后附加一串看似无意义的token序列就能让模型忽略安全对齐59% similarUnverifiedA 2024 study fine-tuned GPT-4 with only a small number of samples producing unsafe code, causing the model to begin expressing harmful values in question-and-answer scenarios unrelated to code.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1127251API
curl https://kongchang.com/api/v1/knowledge/claims/1127251MCP
get_claim(id=1127251)