2026年10月9日 技术热点总结

项目简介：Metal driver for NVIDIA GeForce RTX cards on macOS 15 Sequoia (Intel / OpenCore). Free, source included.

技术栈：Rust

核心介绍：A Metal driver for NVIDIA Turing-and-later cards on Intel Macs and OpenCore systems running macOS 15 Sequoia. The device table includes GTX 16, RTX 20/30/40/50, TITAN RTX, and supported Quadro/RTX workstation cards. The driver implements NVIDIA-backed display and Metal interfaces. Feature availability and application behavior depend on the card, operating-system version and installed components; device-table coverage is not runtime qualification. Made…

项目简介：Screenshots, hung out to dry. A tiny native macOS app that hangs every screenshot on a line at the top of your screen.

技术栈：Swift

核心介绍：Free and open source. For macOS 14 and later. Download › Build from source › English · Español · 简体中文 Every screenshot you take hangs on a line just above your screen. Rest the pointer in the menu bar and it glides down. Move away and it’s gone. Hand Tendedero your screenshots1 and they skip the Desktop entirely. No floating thumbnail. No five-second wa

项目数据：⭐ 991 Stars，🍴 85 Forks

项目简介：Native PC port of P.T. (runs from your own PS4 game files)

技术栈：C++

核心介绍：If the port is worth something to you, you can buy me a coffee: patreon.com/loreanxavier. It keeps the testing hardware and the releases coming. This is a native PC port of P.T., the 2014 PS4 teaser by Kojima Productions. It is not an emulator. The game logic was rebuilt in C++ from the original’s behaviour and the renderer is written on Vulkan; every level, model, texture, sound, script and cutscene is read at run time from your own copy of the PS4 game. There is no game data in this repository

**项目…

项目简介：An open-source, clean-room reimplementation of Microsoft Word in pure Rust

技术栈：Rust

核心介绍：WordCraft Writing and document design; an open-source, clean-room reimplementation of Microsoft Word, rebuilt in pure Rust. A fast, open-source word processor with the Word workflow you already know: the ribbon, styles, tables, track changes, references and mail merge. It reads and writes .docx, runs natively on macOS, Windows, Linux and BSD, and in the browser via WebAssembly. By the ArtCraft team. WordCraft on getartcraft.com · ArtCraft · All Crafting Apps

项目数据：⭐ 808 Stars…

项目简介：CADCraft: computer-aided design and drafting — an open-source, clean-room AutoCAD-style app in pure Rust

技术栈：Rust

核心介绍：CADCraft Computer-aided design and drafting; an open-source, clean-room reimplementation of Autodesk AutoCAD, rebuilt in pure Rust. A fast, open-source take on the AutoCAD workflow: the command line, object snaps, layers, dimensions, hatches, blocks and DXF drawings you already know. It runs natively on macOS, Windows, Linux and FreeBSD, and in the browser via WebAssembly. By the ArtCraft team. CADCraft on getartcraft.com · ArtCraft · All Crafting A…

项目简介：A source code for playgta5.com.

技术栈：JavaScript

核心介绍：This code is grabbed from playgta5.com before it got took down. You can run this locally or make it public hehe. Shoutout to Sebas Furbastian on Telegram for scrapping this code. Idk what is his GitHub but here’s the Telegram and X. Telegram: @SebasKitten Telegram: https://t.me/playgta5regen I don’t host the .mirror folder since it has copyrighted content from Rockstar Games. Please find the files by yourself.

项目数据：⭐ 703 Stars，🍴 393 Forks

World models, game simulators, and long-take video creation require coherent scene evolution and sustained dynamics over extended durations. Autoregressive (AR) video diffusion provides a natural framework for long-horizon generation, yet extended rollouts often become near-static or lose visual quality. We hypothesize that these failures reflect the limited guidance provided by short-video supervision on how ongoing scene dynamics develops over longer durations. This motivates us to introduce LongTake, a two-stage training pipeline built around Long-Horizon Teacher Forcing (TF) on curated …

Optimizing embedding systems for production recommendation pipelines demands systematic exploration that consumes disproportionate engineering effort at scale. We apply Andrej Karpathy's AutoResearch paradigm — a large language model that iteratively edits a training script and retains modifications that improve a held-out scalar metric — to automate this exploration. We report on twelve weeks of running this paradigm at production scale, where iterations consume hours of multi-GPU compute, evaluation involves competing criteria, and campaigns span weeks across many training jobs. Across …

Text-to-image diffusion models are personalized to a subject by DreamBooth fine-tuning on a handful of its images. Increasingly, these images come from a diffusion model rather than a camera. We show that fine-tuning on such synthetic images degrades subject fidelity, producing oversaturated color and excess high-frequency detail. To isolate the cause, we fine-tune two models from the same base model with the same DreamBooth recipe, one on real photos of a subject and one on synthetic images of that subject generated by the first. We trace the degradation to classifier-free guidance (CFG). …

AI agents increasingly operate in environments where they can diagnose failures and improve through experience, yet existing evaluations largely measure what an agent can do at a fixed point in time rather than how effectively it learns. Evaluating self-improvement requires answering three questions: does future performance improve and generalize beyond the interactions that enabled learning; how efficiently are new capabilities acquired; and where does the self-improvement process break down? To answer these questions, we study self-improvement in a controlled setting where agents amortize…

Latent world models have shown a remarkable ability to predict future states and to plan in the real world. In practice, however, we lack a principled way to estimate how their capabilities scale with model size, data, and compute, an open problem that slows progress in the field. In this work we present RoboJEPA, a world model based on the Joint Embedding Predictive Architecture (JEPA) and trained on a large-scale dataset spanning 12 robotic embodiments. We show that RoboJEPA's imagination error, the error of its latent rollouts, follows a second-order power law in compute, allowing us to …

Multi-agent systems (MAS) built from large language models coordinate specialized agents to tackle complex tasks, but effective workflows are difficult to design in advance. Test-time evolution refines workflows using execution feedback, yet broad revisions can disturb useful components, while re-executing unchanged requests can incur redundant computation. Inspired by the interplay of inheritance and selection in biological evolution, we introduce Inherit-MAS, which makes inheritance explicit at the workflow and execution levels. A meta-model first synthesizes a workflow of worker agents w…