2026年9月2日 技术热点总结

项目简介：you’re telling me a shrimp made this video?

技术栈：Rust

核心介绍：> you’re telling me a shrimp made this video?! Shrimply is a free and open-source video editor for creating videos from start to finish, whether you are making a quick edit or something fancy. Shrimply is currently pre-alpha software, which means you should expect: For more information about Shrimply’s features and workflows, visit the documentation website. Contributions to Shrimply are welcome. There are several ways to help beyond Before submitting a contribution, read the repository’s

项目数据：⭐ 653 Stars，🍴 3…

项目简介：Open-source multi-provider AI gateway for Claude Code and other coding agents, with model routing, streaming, tools, reasoning, fallbacks, and local model support

技术栈：Python、ai、claude-code、free-claude-api、multi-provider、multi-provider-llm

核心介绍：It is an independent implementation. It is not affiliated with Anthropic. The catalog contains the broad provider set used by the project: Provider entries are not fake claims of universal support: providers with unusual authentication/protocols require a dedicated adapter. The common OpenAI-compatible providers use the shared t…

项目简介：A farm of real iPhones, run from your Mac. Open-source iOS device automation with live control, a Postgres-backed scheduler, and TikTok workflows. Self-hosted, Apache-2.0.

技术栈：TypeScript、appium、device-farm、ios-automation、marketing-automation、phone-farm、tiktok-automation-script、webdriveragent

核心介绍：An open-source, standalone application for operating physical iOS devices and running scheduled TikTok workflows. It includes guided device registration, WDA/Appium supervision, live video and remote input, PostgreSQL-backed scheduling, recurring jobs, uploads, execution history,…

Long autoregressive video generation faces a fundamental memory challenge: with a finite attention window, a model must decide which information from an ever-expanding history to retain. Existing methods organize memory temporally, preserving recent frames while compressing or discarding older ones. We instead propose RECAP-Forcing, organizing memory by appearance novelty. A long video is not merely a sequence of frames, but an evolving cast of subjects, objects, and scenes whose identities must remain consistent over time. We organize memory by retaining the KV cache associated with newly …

Chain-of-thought (CoT) reasoning has dramatically improved large language models (LLMs) by allowing them to decompose problems into intermediate steps. While CoT is widely effective for linguistic tasks, text-only CoT forces models to serialize visual problems into awkward prose. Although architectural solutions exist to process visual inputs, the community lacks a massive, multi-step, self-corrected dataset to teach models how to build and maintain internal visual workspaces when solving purely textual reasoning problems. To address this limitation, we introduce CoVA-SFT, a highly structur…

Multimodal misinformation on social media is highly prevalent, potent, and harmful, yet difficult to detect and counter, and still poorly understood compared to its text-only counterpart. Research on the properties and deceptive strategies of multimodal misinformation is hindered by a lack of taxonomies grounded in real-world contexts and by the limitations of current multimodal machine learning models, which prevent the automation of annotation and analysis at scale. We address these shortcomings in three steps. First, we collect a large-scale, high-quality dataset of real-world misinforma…

Detecting hallucinations in Large Vision-Language Models (LVLMs) requires both accurate span localization and well-calibrated confidence scores. Fine-tuned generative VLMs excel at identifying hallucinated text spans but suffer from overconfidence and high inference latency. Discriminative sequence taggers offer deterministic speed and superior calibration but exhibit conservative span recall. We present SpanCalib-VLM, a hybrid dual-system for the SHROOM-Visions Shared Task that combines a multimodal sequence tagger, consisting of XLM-RoBERTa-Large fused with a SigLIP vision encoder via cro…

Text-to-image models learn associations between concepts – in the case of this paper, people's professions, which we refer to as roles – and visual attributes. These associations can underpin many observed forms of stereotypical bias. A key open question in this area is whether these associations are stable or change when visual representations of people in professional roles are placed in different prompted contexts. We introduce ContextBias, a controlled evaluation framework, and ContextBench, a benchmark spanning 92 roles and 1,656 semantically controlled prompts, designed to isolate the…

Chain-of-thought (CoT) monitoring assumes that reasoning traces faithfully record the information that shapes a model's answer. Existing faithfulness tests often place explicit bias cues in the user message, while agents may encounter preferences through tool returns or raw artifacts. We introduce FACE-Eval (Faithful Attribution of Cue Effects Evaluation), a 5,100-sample evaluation that varies cue location (user message or tool return) and explicitness (direct summary or raw artifact). We measure verbalized commitment among cue-following answers and unverbalized adoption among all cued samp…