1
AMD 发布 ROCm 10.0.0，重点关注 AI 推理、开发者工具、性能分析
IT之家 8 月 28 日消息，AMD 本周发布了 GPU 加速计算开放软件堆栈 ROCm 的 10.0.0 版本，这一版本号旨在纪念其诞生 10 周年。新版本 重点关注 Instinct、Radeon、Ryzen AI 平台上的 AI 推理、开发者工具、性能分析 。 ROCm 10.0.0 扩展了…
2
How to Steal an AI Model’s Private Thoughts
破解加密推理闭环，看研究团队如何窥探AI私有思考，安全边界再受拷问
In August 2026, a team at MATS Research, the ELLIS Institute Tübingen, and the Max Planck Institute for Intelligent Systems wanted to test whethe…
3
OpenAI 秀肌肉，与博通共研的 Jalapeño 芯片将向英伟达 GB300 发起挑战
OpenAI自研推理芯片Jalapeño首曝成绩，联手博通硬刚英伟达GB300，性能飙升看点十足。
IT之家 8 月 25 日消息，OpenAI 今天（25 日）晚间在 Hot Chips 大会上公布了全新推理系统 Jalapeño 的更多细节，并首次披露基准测试成绩。在 SemiAnalysis 的 InferenceX 测试中，Jalapeño 无论是单用户 token 处理量，还是每千瓦吞吐…
4
OpenAI’s Jalapeño chip is built for fast inference at scale, benchmarks show
OpenAI自研Jalapeño芯片基准测试曝光，专为大规模快速推理而生，直面英伟达Blackwell。
Tested on SemiAnalysis’ InferenceX benchmark, Jalapeño registered both more tokens per user and more throughput per kilowatt than the currently availa…
5
TailSieve: Partial-Rollout-Guided Tail Routing for LLM Rollouts
LLM推理新招：用部分展开引导尾部路由，直击长尾生成瓶颈，性能提升值得关注。
arXiv:2608.22788v1 Announce Type: new Abstract: Large-scale rollouts have become a core component of modern LLM systems, spanning reinforcement learni…
6
OmicSync: Reliability-Aware Spatial Multi-Omics Clustering with Evidence-Constrained LLM Reasoning
用证据约束大模型推理，为空间多组学数据聚类注入可靠性感知，精准破解生物信息难题。
arXiv:2608.22785v1 Announce Type: new Abstract: Spatial multi-omics technologies jointly profile gene expression, surface proteins, and histology at e…
7
Training Large Language Models to Reason in a Continuous Latent Space
让AI在潜空间里“默想”而非逐字推演，这项研究颠覆了传统思维链范式
arXiv:2412.06769v4 Announce Type: replace Abstract: Large language models (LLMs) are typically constrained to reason in the language space, where they…
8
Thinking at the Right Size: Amortized Distillation Across Post-Trained LLMs
不同尺寸模型间摊销蒸馏，让推理成本与思考深度灵活匹配，EMNLP 2026 前沿成果
arXiv:2608.22854v1 Announce Type: new Abstract: Practical deployment of large language models (LLMs) requires families of post-trained variants---inst…
9
ClinicalGPT-R1: Pushing reasoning capability of generalist disease diagnosis with large language model
面向全科疾病诊断的ClinicalGPT-R1，专注提升大模型临床推理能力，医学AI新突破。
arXiv:2504.09421v3 Announce Type: replace-cross Abstract: Recent advances in reasoning with large language models (LLMs)has shown remarkable reasoning…
10
Evaluating Inference-Time Defenses Against Package Hallucination in LLM-Generated Code
大模型写代码常“编造”不存在依赖包？这篇论文系统评估了推理时防御手段，帮你避开代码幻觉坑。
arXiv:2608.22652v1 Announce Type: cross Abstract: LLMs are increasingly used for code generation, yet they frequently hallucinate non-existent softwar…
11
SAFE-G: Structure-aware Faithful Evidence-guided Generation for Knowledge-based Visual Question Answering
结构感知+证据引导，让知识型视觉问答的推理过程更可信、可验证。
arXiv:2608.21796v1 Announce Type: cross Abstract: Knowledge-based Visual Question Answering (KB-VQA) aims to answer queries that necessitate reasoning…
12
NeuroPrefetcher: Storage-Aware Sparse LLM Inference via Delta Prefetching
稀疏大模型推理新突破，用Delta预取打通存储瓶颈，并行计算场景提效显著
arXiv:2608.22643v1 Announce Type: cross Abstract: Deploying large language models on edge devices is increasingly limited by a widening gap between mo…
13
Teaching LLMs How ICU Physicians Approach Clinical Reasoning Through OMOP-Aligned Retrieval Improves Reasoning Across Clinical Domains
用ICU医生的临床推理范式训练LLM检索，OMOP对齐让AI在跨医疗领域的推理能力显著提升。
arXiv:2608.22622v1 Announce Type: cross Abstract: Clinical decision-making relies on identifying relevant patient information to guide diagnosis and t…
14
OpenAI Unveils o3 Mini: Faster, Low‑Cost AI Reasoning Model
低成本高速推理模型o3 Mini发布，开发者不用巨额算力也能用上顶尖AI推理。
Lead OpenAI revealed that it will roll out o3 Mini , a new AI reasoning model, on September 12, 2026 . The company says the model delivers near‑state‑…
15
Write-Side Custody: Trust Begins Before Memory
AI记忆栈关键一环：在写入端建立信任，让事后审计无懈可击，深度解析写侧托管机制。
Part 5 of the Building the AI Memory Stack series The previous articles introduced the Reasoning Ledger and then worked through what a single ledger r…
16
Beyond RGB: Benchmarking and Enhancing MLLMs for Hyperspectral Image Understanding via Training-Free Reasoning Framework
多模态大模型不止看懂RGB，高光谱成像理解迎来全新基准与免训练推理框架。
arXiv:2604.08884v2 Announce Type: replace-cross Abstract: Multimodal Large Language Models (MLLMs) have achieved strong performance on RGB image under…
17
RetroDFM-R: Reasoning-Driven Retrosynthesis Prediction with Large Language Models via Reinforcement Learning
用强化学习让大模型学会「讲道理」，逆合成预测从黑箱猜测走向可解释推理——AI驱动分子设计又进一步。
arXiv:2507.17448v2 Announce Type: replace-cross Abstract: Retrosynthetic planning is a cornerstone of organic synthesis and drug discovery. Yet existi…
18
Hierarchy-Aware Supervised Uncertainty Estimation for Black-box LLM Taxonomic Reasoning
黑盒LLM分类推理的不确定性估计，巧用层级感知监督提升准确性。
arXiv:2608.22839v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used for scientific decision support, yet reliable con…
19
CAI-DLLM: Convergence Aware Inference for Diffusion Language Models
扩散语言模型推理提速新思路，用收敛感知机制减少计算浪费，值得关注。
arXiv:2608.22646v1 Announce Type: new Abstract: Diffusion language models can generate many tokens in parallel, but they still require repeated denois…
20
英特尔介绍 Crescent Island 加速卡：面向 AI 推理，额定 350W、最高 480GB 内存
英特尔AI推理新杀器！Xe3P架构、480GB超大内存，350W功耗锁定智能体工作负载。
IT之家 8 月 25 日消息，科技媒体 videocardz 昨日（8 月 24 日）发布博文，报道称在 Hot Chips 2026 活动中，英特尔详细介绍了 Crescent Island GPU，采用 Xe3P 架构，主要面向 AI 推理和智能体 AI 工作负载。 Crescent Isla…