2026年10月14日 技术热点总结

项目简介：Search your Mac by what’s in your files: photos, PDFs, recordings, videos and code. On-device, built on Google DeepMind’s EmbeddingGemma 2.

技术栈：Swift、embeddinggemma、embeddings、llama-cpp、local-first、macos、menubar-app、multimodal、on-device

核心介绍：DigUp Describe it. Dig it up. Search the files on your Mac by what’s in them, with Google DeepMind’s EmbeddingGemma 2. Download for Mac · Apple Silicon, macOS 14 or later · Free and open source Windows and Linux versions are on the way. EmbeddingGemma 2 is the multimodal embedding model Google DeepMind released in October 2026. It pu…

项目简介：The Craft App Launcher

技术栈：Rust

核心介绍：ArtCraft Launcher One place to install, update and open every app in the Craft App Suite, rebuilt in pure Rust. PhotoCraft, LightCraft, FilmCraft, DesignCraft, VectorCraft, PdfCraft, EffectCraft, WordCraft, DeckCraft, GridCraft, CADCraft, SoundCraft and FigCraft: the right build for your computer, kept up to date. Download ArtCraft Launcher · ArtCraft · The Craft App Suite > ArtCraft is a community of artists from all walks of life. Painters, photographers,

项目数据：⭐ 2,290 Stars，🍴 766 Forks

技术栈：Rust

核心介绍：A native macOS, Linux and Windows toolkit for a cleaner workspace and an environment you can verify. Claude identities · Browser fingerprints · System settings · Network Skill Features · Installation · Updates · Quick start · Usage guide · Architecture · Development · Contributing

项目数据：⭐ 2,047 Stars，🍴 206 Forks

项目简介：SolidCraft: an open-source parametric 3D CAD application built in pure Rust. Desktop, web, and automation through MCP.

技术栈：Rust、3d-modeling、brep、cad、egui、mcp、parametric-cad、rust、step

核心介绍：Sketch, constrain, and build solids with a parametric feature timeline. Change a dimension and watch your model rebuild. Run it on your desktop, in your browser, or let AI agents model through the built-in MCP server. Also for [ARM64][windows-arm64-msi] and [32-bit][windows-x86-msi] PCs, or as a [portable zip][windows-portable] that runs without installing. (Apple silicon and Intel). Or …

Modern AI models are aligned through post-training to adapt them to downstream tasks. Recent work shows that fine-tuning language models on narrow tasks can induce emergent misalignment (EM), causing broadly harmful behaviors beyond the training task. However, EM has been studied almost entirely in text-only tasks, leaving its manifestation in multimodal models unclear. In this paper, we define and analyze EM in the context of vision-language models. We first induce EM via fine-tuning on narrow multimodal tasks targeting vulnerable code, careless household-object use, and conspiratorial int…

Multimodal large language models (MLLMs) struggle with spatial, embodied, physical, and temporal reasoning. We hypothesize that these failures reflect a shared deficit in visual transition reasoning, and test whether this capability can serve as a shared training primitive, one that different models can learn from different supervision sources and reuse across different tasks, with a systematic training recipe. Existing benchmarks document these deficits separately but do not support controlled comparisons across scenes, actions, and reasoning operations. We therefore introduce WOVEN, a tra…

Long-form thinking traces can substantially improve the multi-step reasoning performance of large language models (LLMs), but they introduce high inference-time overhead, with latency dominated by sequential decoding. We propose HyperThink, a text-to-parameter approach that amortizes this reasoning computation into a single query-conditioned parameter update: a lightweight hypernetwork reads the question and predicts updates to a small subset of the base LLM's parameters, while a vector-quantized decoder constrains them to a finite set of reusable patterns to improve robustness and transfer…

Language-model agents are usually trained by reinforcement learning from one reward per episode, and privileged self-distillation enriches it by letting the same policy, given a skill, teach its skill-free self through token probabilities. However, we identify two phenomena that question this channel. Invisible Advantage: a skill in context lifts WebShop success from 42.2% to 56.2%, yet changes the probabilities of fewer than a quarter of the sampled tokens. Much to Align: a skill changes the hidden states of over 80% of response tokens, in a way that linear probes can trace back to the spe…

World action models (WAMs) aim to answer a coupled physical question: given a task instruction, what motion should the robot execute, and how will that motion change the surrounding world? Most existing WAMs build on pretrained video generators and represent world evolution through images or visual latents. Robotic interaction, however, takes place in metric three-dimensional space, while images are view-dependent projections whose pixel distances do not directly encode physical distances. We introduce UNITAS, to our knowledge the first 3D-native world action model that unifies observations…

Vision-language models (VLMs) are increasingly trained to generate structured outputs like points and bounding boxes that downstream interfaces, agents, and robots can act on, yet safety alignment of this output channel has not been systematically analyzed. We study visual grounding safety by repurposing three safety benchmarks spanning direct harm (VLSU), social bias (BBQ-V), and situational safety (Asimov-2.0) into 15,401 matched pairs of harmful requests that differ only in the requested output: a free-text answer (VQA) or a grounding (point or bounding box). Across five VLMs, models tha…