来自Linux.do社区的用户反馈显示，ChatGPT 5.2网页版存在明显使用体验问题。用户指出，该版本对指令的遵循能力较差，回答排版变得混乱，出现过多大标题和小标题，导致内容重点不突出，难以快速获取关键信息。尽管如此，用户认为模型的基本智能水平相比之前版本没有明显下降，这些体验问题实际上影响了用户对模型’智商’的评价。这一反馈对正在使用或考虑升级到5.2版本的用户具有重要参考价值，也反映了AI产品在实际应用中可能遇到的问题。
原文链接：Linux.do
Xiaomi recently launched its open source large model MiMo-V2-Flash, and preliminary tests show that the model performs excellently in inference speed, able to quickly respond to various requests. However, user feedback indicates that the model still falls short of expectations in terms of comprehension and content generation quality. As an important part of Xiaomi's AI strategy, the open source nature of MiMo-V2-Flash will provide more possibilities for developers and researchers. Industry insiders believe that although the current version has certain limitations, Xiaomi's continued investment in AI model research and development is expected to launch more complete versions in the future.
Original Link:V2EX Share & Discovery
EU regulators found Apple's App Store rules violating the Digital Markets Act six months ago, but developer coalitions are now publicly accusing Apple of still failing to achieve true compliance. Non-profit organizations like the Coalition for App Fairness have sent letters to the European Commission stating that Apple's revised App Store terms continue to collect fees that are prohibited by legislation. Despite the EU ruling in April this year that Apple's App Store policies were illegal and harmful to developers and consumers, Apple has failed to make 'any meaningful changes or proposals'. This incident highlights the ongoing tension between tech giants and EU regulators, and also reflects global tech regulatory trends. As one of the world's most valuable tech companies by market capitalization, Apple's compliance progress will affect the entire app ecosystem and developer community, warranting continued attention from the tech industry.
Source Link:Hacker News
GitHub has announced a major architectural upgrade and pricing adjustment for its Actions service, increasing daily processing capacity from 23 million to 71 million jobs, a 3x improvement. Under the new architecture, enterprises can launch 7x more jobs per minute. Starting January 1, 2026, GitHub-hosted runner prices will decrease by up to 39%, while introducing a $0.002 per minute cloud platform fee. Self-hosted runners will be included in the free quota starting March 1. This upgrade also enhances the self-hosted experience with features like Scale Set Client, multi-label support, and Actions Data Stream, providing stronger support for AI agent workloads. Statistics show that 85% of users will see reduced bills, with individual developers largely unaffected. These improvements aim to deliver a faster, more reliable CI/CD experience while providing a scalable execution layer for GitHub's agent platform.
Original Link:Hacker News
FVWM-95 is a window manager project based on fvwm2, designed to provide Unix system users with a Windows 95-like interface experience. By mimicking the classic appearance and functionality of Windows, this project helps users migrating from Windows to Unix maintain a familiar operating environment. FVWM-95 offers flexible configuration options, supports feature extension through loadable modules, and provides practical components like a taskbar. The project has also developed a C++ class library that offers Win95-style controls, making it easier to develop new applications. As an open-source project, FVWM-95's main code is hosted on SourceForge, with a dedicated mailing list for user communication. Although the project has existed for many years and is no longer active, it represents an interesting attempt in the history of operating system interfaces, demonstrating the cross-pollination and fusion of interface styles between different operating systems.
Original Link:Hacker News
According to observations from Linux.do community users, AI model services have experienced significant fluctuations recently. The Hakimi 3.0 model became largely unavailable around midday, while AI Studio service has mostly resumed supply, with the high-thinking mode no longer rate-limited. Users noted that the restoration of AI services is a prerequisite for testing with prompts. In terms of model selection, the article recommends using Gemini 3 and Claude, and advises following prompt engineering principles. Notably, users have experimented with developing and optimizing prompts in the linguistic style of Zhu Ziqing (Peixian), a unique approach that may offer new perspectives for AI applications. For tech enthusiasts focused on AI model performance and availability, these observations and recommendations hold considerable reference value.
Original Link:Linux.do
MimoV2, an AI assistant that identifies as Gemini, demonstrates highly human-like interactions with smooth and natural responses, creating a stark contrast to models like GPT and delivering a comfortable user experience. However, in practical use, when thinking mode is enabled, the system frequently encounters errors after just 3 seconds, leading to functional instability. Similar issues exist in other AI products like Doubao, revealing the current technical bottlenecks in AI conversation systems when handling complex tasks. For AI enthusiasts and developers, this provides valuable real-world feedback that can help improve product performance.
Original link:Linux.do
最新评论
照片令人惊艳。万分感谢 温暖。
氛围绝佳。由衷感谢 感受。 你的博客让人一口气读完。敬意 真诚。
实用的 杂志! 越来越好!
又到年底了，真快！
研究你的文章, 我体会到美好的心情。
感谢激励。由衷感谢
好久没见过, 如此温暖又有信息量的博客。敬意。
很稀有, 这么鲜明的文字。谢谢。