[人人能懂] 当AI学会“四两拨千斤”：智能的终点不是暴力计算

00:01:54 四两拨千斤：让小模型变聪明的“内存魔法”

00:07:11 AI调教指南：不说假话的秘密，竟然是“相信自己”？

00:11:07 AI思考的“地图”与“导航”

00:15:16 AI大模型的“阿喀琉斯之踵”？

00:18:49 AI写作高手进阶：人多，不如方法好

本期节目介绍的五篇论文：

[CL] KV Cache Steering for Inducing Reasoning in Small Language Models  

[University of Amsterdam & University of Technology Nuremberg]  

https://arxiv.org/abs/2507.08799 

---

[CL] The Curious Case of Factuality Finetuning: Models' Internal Beliefs Can Improve Factuality  

[University of Washington]  

https://arxiv.org/abs/2507.08371 

---

[LG] CTRLS: Chain-of-Thought Reasoning via Latent State-Transition  

[UC San Diego]  

https://arxiv.org/abs/2507.081 

---

[LG] One Token to Fool LLM-as-a-Judge  

[Tencent AI Lab]  

https://arxiv.org/abs/2507.08794 

---

[LG] Inference-Time Scaling of Diffusion Language Models with Particle Gibbs Sampling  

[Stanford University]  

https://arxiv.org/abs/2507.08390