1
AI with Authority, from Application to Silicon
生成式AI让机器验证从奢侈品变为必需品，成为不可收买的裁判
arXiv:2608.21356v1 Announce Type: cross Abstract: For sixty years, machine verification has been a major cost overhead, affordable only for exceptiona…
2
Competing at Every Price Point with Agentic Evolution over a Menu of LLMs
多模型菜单驱动智能体进化，在每一价格点实现Pareto最优精度，重塑LLM竞争策略。
arXiv:2608.16207v1 Announce Type: new Abstract: Consider a firm that surveys its competition for a particular agentic task and seeks to offer superior…
3
Ask HN: What's Next for LLMs?
用不到一美元日成本，DeepSeek在Agent工作流中效率翻倍，引发对前沿模型定位的思考。
I've been using DeepSeek v4 flash for guided agent workflows (in-IDE, prompt/review diffs) and have found no noticeable benefit to using Haiku, Opus, …
4
CAT-Q: Cost-efficient and Accurate Ternary Quantization for LLMs
被ICML 2026接收为oral，提出一种兼顾成本与精度的三元量化新方法，为大模型高效部署提供突破。
arXiv:2606.26650v1 Announce Type: cross Abstract: In this paper, we present CAT-Q, Cost-efficient and Accurate Ternary Quantization, for compressing a…
5
Prism: Cost-Efficient Multi-LLM Serving via GPU Memory Ballooning
提出GPU内存气球技术，实现多LLM服务成本大幅降低，已在超万卡生产环境验证。
arXiv:2505.04021v3 Announce Type: replace-cross Abstract: Inference providers must maintain availability for many LLMs, including low-volume but essen…
6
研究员测试 AI 漏洞挑战：GPT 5.5 成功率最高、Deepseek V4 Pro 成本最低
GPT-5.5解题率最高，DeepSeek V4 Pro成本仅对手1/15，AI漏洞攻防实战对比。
IT之家 6 月 4 日消息，安全研究员 Kasra Rahjerdi 昨日（6 月 3 日）发布报告，搭建了一个故意留有漏洞的图书评论 APK， 测试多款 AI 大语言模型的安全推理能力。 研究员模拟真实场景漏洞，在 APK 文件内放入暴露的 Firebase（谷歌移动端后端服务）凭据，模型只要解…
7
Talk Is Cheap: The Operational Impact of LLM Use
从实操层面剖析LLM应用的真实成本与效率权衡，直击“说易行难”的核心痛点。
Article URL: https://unessays.substack.com/p/talk-is-cheap Comments URL: https://news.ycombinator.com/item?id=48347155 Points: 23 # Comments: 11
8
Goal-Conditioned Supervised Learning for LLM Fine-Tuning
提出目标条件监督学习新方法，有效平衡LLM微调的成本与效果，无需外部奖励模型。
arXiv:2605.16345v1 Announce Type: new Abstract: Large language models often require fine-tuning to better align their behavior with user intent at dep…