推理时多花算力能换准确率：self-consistency 投票、过程奖励模型引导搜索、budget forcing 强制续想，三条路线各有边界。本文以 s1 论文与 2025–2026 年的 overthinking 研究为轴，梳理 test-time scaling 的效果账与失效点。

以十个问答讲清 Scaling Laws：幂律为什么可外推、Chinchilla 如何修正参数与数据的配比、6ND 公式怎么当算力换算器、数据枯竭与推理时算力这两条新变量，以及这门「经验定律」对大公司与创业者的不同含义。

Test-time scaling can improve large language model reasoning by generating and combining multiple candidate responses. I…

2017 年 6 月，八位 Google 研究员在一篇机器翻译论文里抛弃了循环网络，标题只有五个词：Attention Is All You Need。此后九年：GPT-3 证明规模即能力、ChatGPT 两个月一亿用户、Scaling Laws 驱动军备竞赛、DeepSeek 用十分之一成本撼动算力叙事、推理模型开启第二条扩展曲线。本文是「机器学习简史」系列完结篇，所有 2024 年后的关键事实均经联网核实，带你看清截至 2026 年 10 月的大模型版图，与依然悬而未决的终极争论。

2017 年 Ali Rahimi 领奖时说「机器学习变成了炼金术」，LeCun 当场反驳；八年过去，模型能力涨了几个数量级，「为什么泛化」的数学解释仍然是一堆可解模型里的定理。本文盘点深度学习数学基础的现状：随机标签实验、双下降、NTK、grokking、scaling laws 各自被解释到了哪一步，哪一步仍然卡住，以及 2026 年「学习力学」的学科宣言算不算转折点。

Static coding benchmarks are fixed at release. Over time their tasks leak into training data, and stronger agents eventu…

Reinforcement learning (RL) is the central training paradigm for advancing large foundation models towards self-improvem…

Most modern multimodal large language models (MLLMs) build on a pretrained visual encoder that provides a strong visual …

While Large Language Models (LLMs) advertise million-token context windows, reasoning quality often collapses as inputs …

Jay* 2026-09-25 14:14:07 来源：量子位 一场物理世界的基模实验 程浅 发自 凹非寺 量子位 | 公众号 QbitAI 数亿元资金，投向了一场物理世界的基模实验。 Physical AI创业公司**息壤开物**宣布，…

A multi-agent system can reduce latency on complex tasks by executing work concurrently. Several pioneering harness fram…

Training capable coding agents via reinforcement learning (RL) requires diverse tasks with reliable verifiers. Open-sour…

As coding agents move from supervised code completion to unattended, around-the-clock exploration, their work expands fr…

World-action models (WAM) predict future states to guide robot actions, enabling learning from both action-free video an…

Automating empirical research is a long-standing direction of AI. Recent automatic research (AutoResearch) agents bring …

This paper shows that JEPA-style latent world models (LWMs) can learn across diverse domains and data sources by adoptin…

Medical workflows often require multiple interdependent analyses, but existing specialist and generalist AI models are n…

Computer-use agents (CUAs) are rapidly improving, yet synthesizing diverse and reliable software environments at scale r…

论文提出个性化测试时扩展（Personalized Test-Time Scaling）问题，将用户对精度、延迟与推理成本的联合要求纳入优化目标，并给出摊销式智能体策略发现框架 PersonTTS 以降低为不同用户需求重复搜索策略的开销。

Automated research systems increasingly run LLM agents over long horizons, but more inference does not by itself produce…

Preference distillation typically treats a teacher response as preferred and the student's own response as rejected. Thi…

Scaling language models to long contexts is limited by the quadratic cost of self-attention. Block sparse attention offe…

AI factories are the infrastructure of the intelligence era. Scaling them responsibly will depend as much on innovation …

System performance, efficient infrastructure scaling and continuous software optimization are key levers that determine …

We introduce LimiX-2, a new model in the LimiX family, developed through model and data scaling guided by our previously…

Scaling large language models efficiently has motivated sparse capacity mechanisms such as Mixture-of-Experts and, more …

把 2026 年可核查的公开证据整理成一张六层能力路线图——预训练、后训练 RL、推理时计算、上下文与记忆、智能体与 Harness、世界模型。含 Meta ScaleRL 40 万 GPU 小时实验结论、RL 预算占比 10%–30% 口径、Chinchilla 对比、Meta-Harness 6x 差距等数据锚点，并给出优先级表与算法工程师/产品经理的行动建议。

o1 系列通过强化学习训练模型'先思考再回答'，在数学、代码与科学推理上大幅跃升，开创了推理时扩展（Test-time Compute）的新 Scaling 维度。

Oct 25, 2026 Nano Banana Pro prompted by THE DECODER Key Points Anthropic recommends that companies evaluate whether the…

Relay, an AI-powered workflow automation tool built as an alternative to Zapier, shut down entirely in August. The produ…

Manuel Uth Oct 25, 2026 Nano Banana Pro prompted by THE DECODER Key Points The US and China have locked up the global AI…

Building a startup means making decisions before you have all the answers. When should you scale? How do you know you re…

「先更新数据库再删缓存」几乎是每个人的默认写法，但为什么是这个顺序、残余的不一致窗口长什么样、延迟双删和 binlog 订阅各自在补什么洞——本文把缓存与数据库双写的四种时序逐一推演，讲清 Cache-Aside 的规矩从哪来、窗口怎么收敛，以及生产环境的组合拳与避坑清单。

Kazakhstan’s AI ecosystem is moving from infrastructure plans to deployment. The shift is visible across the full AI sta…

Governments need modern solutions to serve the public while keeping sensitive information under their control. Open mode…

Sponsored In partnership withCelonis Enterprise AI can generate recommendations, automate tasks, and help organizations …

梦瑶* 2026-10-22 15:05:59 来源：量子位 挤进全球Physical AI核心供应链 这RSI，那AGI，别看今天山姆·奥特曼已经把自家模型卷得没边儿了，我还是要说一句—— 在Physical AI这事儿上，**Open…

At GTC Berlin, banks, payment companies and financial data providers are showing what production-scale intelligence look…

Hierarchical language generation separates continuous semantic modeling from discrete text realization, but its effectiv…

At GTC Berlin, NVIDIA is showcasing its full lineup of local AI hardware. RTX Spark, which launched last week, powers pe…

NVIDIA builds tools to enable safe agents across the stack. The NVIDIA OpenShell Reference Architecture, announced today…

As model parameters scale up, sparsely activated Mixture-of-Experts (MoE) approaches have become essential for model sca…

The Nordic countries have become the European leader in unicorn startup density, producing the most emerging companies v…

Keeping a city running means making decisions that affect residents every day — from when to repair a road to how to use…

Across industries, the U.K. is putting the full stack to work. The scale of deployment is growing fast, from enterprises…

提出一种无需重训的方法，通过可松弛的神经元与注意力头联合排序，将预训练 Transformer 转化为可随算力预算动态伸缩的单一模型。

The U.S. Air Force will buy data collected by WindBorne Systems‘ fleet of autonomous weather balloons under a new contra…

As AI factories grow from megawatts to gigawatts, building them takes more than advanced computing. Open standards are h…

Journalists, entrepreneurs, and even some public relations professionals will tell you that the PR process is broken (bu…

Oct 11, 2026 AI agent teams deliver almost no better results than single agents, research finds.** Evals company Vals AI…