1
雅马哈发布旗下最紧凑杜比全景声回音壁 B200A，长度 60cm
IT之家 9 月 16 日消息，Yamaha（雅马哈）本月 15 日宣布推出该企业旗下最紧凑的杜比全景声认证条形音箱（回音壁）产品 B200A。 这一型号 长度仅有 60cm ，定价 1,580 元人民币 / 249.95 美元 （IT之家注：现汇率约合 1,681 元人民币) 。 雅马哈消费音频总…
2
When Entropy Is Not Enough: Reclaiming Lost Semantics in LLM Output Length Prediction
熵不再是万能指标，这项研究用语义恢复破解LLM输出长度预测难题。
arXiv:2608.15592v1 Announce Type: new Abstract: Efficient LLM serving is often bottlenecked by the need to pad sequences to a fixed maximum length, an…
3
Beyond Representational Similarity: Source-Conditioned Description-Length Gain for Generative Plagiarism Detection and Candidate Source Reranking
用信息论新指标精准识别生成式抄袭，还能重排候选来源，直击AI时代学术诚信痛点。
arXiv:2608.03859v1 Announce Type: new Abstract: Large language models (LLMs) pose challenges to academic integrity and peer review. Yet generative pla…
4
VaLiDRec: Variable-Length LLM-Aligned Semantic IDs for Generative Recommendation
生成式推荐新突破，用LLM对齐的可变长度语义ID提升推荐效果，值得关注。
arXiv:2607.25209v1 Announce Type: cross Abstract: Generative recommendation commonly represents items using fixed-length semantic identifiers (SIDs) c…
5
Ask HN: Have you noticed an improvement in AI responses with memory disabled?
禁用AI记忆功能后，反应质量不降反升？来自Hacker News的真实体验分享。
I use Claude on the $20/mo plan, mainly for research, "rubber duckying," and as an idea soundboard (I was using Fable for this but am now relegated to…
6
34.9 元：酷态科 CP 编织 6A 彩虹线开售，长度约 2m
IT之家 7 月 20 日消息，酷态科 CP 编织 6A 彩虹线目前已在京东开售，新品支持 240W MAX 超级快充，线长约 2m， 售价 34.9 元 。 据介绍，这款充电线采用编织线材质，拥有纳米抗脏技术，在编织线表面形成一层隐形防护膜，日常使用不易沾染污渍。 规格方面，这款充电线最高支持 6…
7
Beyond Prediction: Tail-Aware Scheduling for LLM Inference
大模型推理的延迟"长尾"痛点有解了！本文提出GPU调度新方法UniBoost，通过感知输出长度分布，显著降低高响应百分位延迟。
Article URL: https://yl3469.github.io/uniboost-icml26/ Comments URL: https://news.ycombinator.com/item?id=48808378 Points: 2 # Comments: 0
8
How Much is Left? LLMs Linearly Encode Their Remaining Output Length
大语言模型秘密编码剩余输出长度，揭示其生成结构的可预测线性规律。
arXiv:2607.05316v1 Announce Type: cross Abstract: Large language models generate one token at a time, yet their responses show remarkably consistent l…
9
TruncProof: A Guardrail for LLM-based JSON Generation under Token-Length Constraints
LLM生成JSON时因token截断而崩溃？这篇被IJCNN录用论文提出TruncProof护栏，确保格式完整不报错。
arXiv:2605.13076v2 Announce Type: replace Abstract: The LLM-based generation of machine-readable outputs such as JSON has attracted significant attent…
10
Randomized YaRN Improves Length Generalization for Long-Context Reasoning
一种随机化YaRN方法，破解长上下文推理外推难题，实验显示显著提升长度泛化能力。
arXiv:2606.23687v1 Announce Type: new Abstract: Large language models (LLMs) are typically pretrained on short sequences and then extended to work on …
11
CARE: Competence-Aware Reward Shaping for Adaptive Reasoning Length in Video-MLLMs
视频大模型推理长度不再是死板固定，CARE用能力感知奖励塑形实现自适应思考，效率与精度兼顾。
arXiv:2606.19927v1 Announce Type: new Abstract: In multimodal video reasoning, reinforcement learning-based methods typically rely on simplistic and i…
12
大疆 DJI Power 140W 充电头发售：2C + 1A 售 219 元，同步推出 1 米长度 USB-C 带屏 7A 线缆
IT之家 6 月 16 日消息，大疆旗下 DJI Power 140W 充电头现已正式发售， 该产品采用 2C + 1A 设计，定价为 219 元 ，官方同步推出了 1 米长度 USB-C 带屏 7A 充电线，充电头配充电线套装定价为 239 元。 京东 大疆 DJI Power 140…
13
PostgreSQL 2200F Error: Causes and Solutions Complete Guide
详解PostgreSQL 2200F错误根源，提供正则函数与约束场景的即用解决方案，助你快速定位并修复零长度字符串问题。
PostgreSQL Error 2200F: Zero Length Character String PostgreSQL error 2200F: zero_length_character_string occurs when an empty string ( '' ) is passed…
14
长度超过 5.2 米，比亚迪王朝首款 D 级旗舰轿车大汉申报
IT之家 6 月 10 日消息，工信部今晚发布第 408 批《道路机动车辆生产企业及产品公告》新产品公示，比亚迪大汉轿车完成申报。与家族的大部分车型类似，新车依旧提供纯电、插混两种动力，定位比亚迪王朝首款 D 级旗舰轿车。 IT之家从申报信息获悉，“大汉”的长宽高分别为 5256/1999/1510…
15
Show HN: Best setup local LLM found for a 5090 (llama.cpp fork + turboquant)
RTX 5090本地部署大模型新方案：450K上下文+多模态，基于llama.cpp fork与turboquant，实测效果出色
Hi folks, I found this setup on consummer hardware that seems to have great results on local hardware. - qwen 3.6 q6 - 450 K context using turboquant …
16
SmartThinker: Progressive Chain-of-Thought Length Calibration for Efficient Large Language Model Reasoning
提出SmartThinker方法，通过渐进式调整思维链长度，让大模型推理效率大幅提升，已被ICML 2026收录。
arXiv:2603.08000v2 Announce Type: replace-cross Abstract: Large reasoning models (LRMs) like OpenAI o1 and DeepSeek-R1 achieve high accuracy on comple…
17
HMPO: Hybrid Median-length Policy Optimization for Chain-of-Thought Compression
混合中长度策略优化实现思维链高效压缩，兼顾性能与成本。
arXiv:2606.01934v1 Announce Type: new Abstract: Large language models achieve remarkable performance via extended chain-of-thought (CoT) reasoning, ye…
18
Scheduling LLM Inference with Uncertainty-Aware Output Length Predictions
用不确定性感知的动态预测优化LLM推理调度，打破传统单点估计瓶颈，显著提升SJF效率
arXiv:2604.00499v2 Announce Type: replace Abstract: To schedule LLM inference, the \textit{shortest job first} (SJF) principle is favorable by priorit…
19
Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?
揭秘自蒸馏为何会损害LLM的数学推理能力，并指出抑制关键探索过程是背后原因。
arXiv:2603.24472v3 Announce Type: replace-cross Abstract: Self-distillation has emerged as an effective post-training paradigm for LLMs, often improvi…
20
Queue Length Regret Bounds for Contextual Queueing Bandits
探索上下文排队老虎机中的队列长度后悔界限，为在线队列调度提供理论保证。
arXiv:2601.19300v2 Announce Type: replace Abstract: We introduce contextual queueing bandits, a new context-aware framework for scheduling while simul…