OpenAI新论文拆解语言模型内部机制：用「稀疏电路」解释模型行为
"Interpretability" 相关文章
OpenAI把Transformer训练成「几乎全零」，黑箱首次被彻底揭开
Self-interpretability: LLMs can describe complex internal processes that drive their decisions
Weight-sparse transformers have interpretable circuits
OpenAI’s new LLM exposes the secrets of how AI really works
【ICML25】使用信息瓶颈理论为点云模型进行错误归因，为安全问题构建可解释工具
3A大作！阿里ROLL团队从基建->算法->机理，推动RL4LLM全栈协同优化
3A大作！阿里ROLL团队从基建->算法->机理，推动RL4LLM全栈协同优化
Can Models be Evaluation Aware Without Explicit Verbalization?
Bottom-Up: Principled Compression to Shrink LLMs
Toward Statistical Mechanics Of Interfaces Under Selection Pressure
【ICML25】使用信息瓶颈理论为点云模型进行错误归因，为安全问题构建可解释工具
The Future of Interpretability is Geometric
Learning to Interpret Weight Differences in Language Models
Activation Plateaus: Where and How They Emerge
Activation Plateaus: Where and How They Emerge
Finding Features in Neural Networks with the Empirical NTK
大语言模型的遗忘也许是一个悖论？清华团队揭示Unlearning技术的两难困境
Training Qwen-1.5B with a CoT legibility penalty
Training Qwen-1.5B with a CoT legibility penalty
