2026年10月18日 技术热点总结

项目简介：Uncharted: The Nathan Drake Collection running natively on Windows – by SuperMedo

技术栈：C++

核心介绍：Same idea as my Bloodborne port: the original PS4 game code runs directly on your PC and the graphics are translated to Vulkan. Unpack the zip, start Uncharted.exe, pick your game folder, choose a game and press PLAY. > No game files are included. You need your own decrypted dump of the collection > (CUSA02343, CUSA02320, …: the folder with eboot.bin, big2-ps4_Shipping.elf, > big3-ps4_Shipping.elf, sce_sys, u1data, u2data and u3data).

关键特性：All three games from one…

项目简介：基于 REA 深度拆解并高保真还原 Claude 官方桌面端 UI。还原 30 个独立 React 组件与 8 项硬核交互物理算法，附带 3000+ 行逆向分析手册。

技术栈：TypeScript、anthropic、audioworklet、claude、design-system、framer-motion、motion-physics、react、reverse-engineering

核心介绍：基于 REA 深度拆解，高保真生产级还原 Claude 设计系统 (CDS) 与工程交互物理学。 简体中文 · English open-claude-ui 是基于对现代桌面端 AI 客户端进行深度拆解与逆向分析构建的高保真开源 UI 组件库与动效引擎。 整个逆向推导、反混淆静态分析与 ASAR 资源解构，均依托于先进的多平台逆向工程框架 REA (Reverse Engineer Anything) 完成——通过 REA 的静态 AST 解析、闭包代码反混淆与运行时跟踪能力，穿透生产级压缩包，精准还原了底层工程物理常数与数学模型。 不同于市面上常见的盲猜或近似复刻，本项目所有组件规范、物理阻尼常数、着色器关键帧及音频管道均严格依据真实前端架构反编译验证提炼而成：

项目数据：⭐ 1,133 Star…

项目简介：Introduction to Semiconductor Manufacturing

技术栈：Python、manufacturing、semiconductor

核心介绍：Introduction to Semiconductor Manufacturing An open, beginner-friendly book that follows one silicon wafer from sand, through the fab, test and packaging, until it becomes a chip. It skips heavy physics and focuses on process flow, terminology and know-how from the fab floor, with term boxes and know-how boxes in every chapter. No background in semiconductor physics is needed.

项目数据：⭐ 1,094 Stars，🍴 103 Forks

项目简介：Native Windows port of Shadow of the Colossus (PS4, CUSA08809) – bring your own game dump. Port nativo para PC.

技术栈：C++

核心介绍：> Nenhum arquivo do jogo nem da Sony está incluído / No game or Sony files are included. > Você precisa do seu próprio dump do jogo e, para ter som, de dois módulos do seu próprio > console. / You need your own game dump and, for sound, two modules from your own console. > Este projeto não tem ligação com a Sony Interactive Entertainment nem com a Bluepoint Games. / > This project is not affiliated with Sony Interactive Entertainment or Bluepoin…

项目简介：Scout: the open-source browser agent, by Outcrawl. Mirrored automatically from Outcrawl’s own repository.

技术栈：TypeScript、ai-agent、browser-agent、browser-automation、cdp、llm、typescript、web-automation

核心介绍：Quickstart Examples Models Cloud Docs Scout is an open-source agent that uses a real browser to finish a task for you. Describe what you need in plain words. Scout reads the page, clicks, types and scrolls until the job is done, then hands back the answer. If it cannot finish, it tells you what is missing and returns what it found.

项目数据：⭐ 936 Stars，🍴 28 Forks

Modern generative models produce high-quality individual outputs but struggle to sample diversely, even when explicitly instructed to do so. We introduce MINE-SAMPLE-RENDER (MSR), a test-time approach that externalizes the diversity into an explicit, corpus-derived sampling procedure. Our pipeline (i) mines an interpretable categorical schema from a corpus of real artifacts and supplies statistics for sampling; (ii) samples attribute configurations from the corpus-derived statistics; (iii) uses a generative model purely as a conditional renderer of a sampled configuration. Across weather re…

Sparse autoencoders (SAEs) represent model activations with sparse, linearly decoded features, but the downstream effects of those features need not compose additively. We introduce SAE Additivity Residuals (SAR), which measure the behavioral variation left unexplained above a chosen interaction order. Under independent feature subset sampling, population SAR is exactly the normalized energy of interactions above that order and bounds coalition ordering error for the optimal low order approximation. To distinguish compositionality associated with the SAE representation from that induced by …

Sparse autoencoders (SAEs) are typically trained on hundreds of millions to billions of activation tokens, yet little work has tested whether this scale is necessary. We study how much activation data SAEs need by training four SAE architectures on three language models with budgets ranging from 5M tokens to a 500M-token reference. We measure two notions of convergence: functional behavior on standard evaluations and one-to-one reproducibility of learned features. Functional data requirements are strongly metric-dependent: reconstruction often approaches the full-data reference at substanti…

Computer-use agents increasingly tackle long-horizon tasks, where running a frontier model at every step is expensive and a small open-weight model alone often fails. Weak-to-strong escalation offers a middle ground where a weak agent acts by default and hands control to a strong one only when needed. The core challenge is deciding when to escalate, which requires an uncertainty signal cheap enough to check at every step. However, even after handoff, the strong agent can still get lost, but a black-box API that the strong agent is usually served exposes no internals to monitor. To this end,…

Agent environments are essential for large language model (LLM) development. They use programmatic verifiers to turn complex outcomes into rewards. Verifiers can be attacked into awarding full credit to cheating agents without completing the intended task, causing inaccurate evaluation and misalignment in training. As agent environments scale in numbers and difficulty, so must the repairing of exploitable verifiers. To this end, we ask whether today's model with scaled compute can automatically harden agent environments to prevent potentially stronger future models from reward hacking. We i…

AI systems now write long and difficult mathematical proofs, and checking their step-by-step validity is becoming a bottleneck. A learned verifier could check proofs cheaply, but training one requires a correctness label on every step, and such labels are hard to obtain for open-ended proofs. Existing methods take them from human annotators, which is costly, or derive them from final answers, which proofs do not have, and formalizing each proof in Lean requires expert effort. We introduce LeanPRM, a framework that uses the Lean proof assistant to create step-level labels for natural-languag…