近期，科技爱好者观察到谷歌Gemini和OpenAI ChatGPT在用户认证后出现明显的性能下降，表现为回答质量降低、上下文缩短、实现简化等现象。作者分析认为，这可能是由于平台端提示词策略限制了模型的深度思考能力。为解决这一问题，作者分享了精心设计的Prompt集，涵盖论文分析、学术建议专家和绘图专家等场景。通过详细的少样本示例和操作指南，展示了如何优化Prompt以提升AI在学术论文解读、代码模型改进和科研绘图方面的输出质量。这些实用技巧不仅揭示了AI模型的局限性，还为用户提供了有效的性能优化方案，对关注前沿AI技术的读者具有较高参考价值。
原文链接：Linux.do
OpenAI Codex Officially Introduces Skills Feature, Expanding AI CapabilitiesOpenAI Codex Officially Introduces Skills Feature, Expanding AI CapabilitiesGemini 健忘实测：对比 ChatGPT 的记忆缺陷OpenAI Codex正式引入Skills功能，扩展AI能力边界AI酒馆应用：基于Gemini与Claude的纯前端创新2025年AI技术反思：扩散模型与安全挑战AI教学资源：YOLOv11目标检测全流程实战指南YOLOv11目标检测实战教程，20分钟掌握全流程本周AI领域密集发布：OpenAI的GPT Image 1.5让图像生成速度提升4倍；Google的Gemini 2.0 Flash以极低成本成为默认模型；Anthropic开放Skills标准化让技能跨平台复用；Cursor收购Graphite强化代码审查能力。整体趋势是多模态能力加强、成本持续下探、个性化成为差异化关键。
12月16日，OpenAI发布GPT Image 1.5，这是DALL-E系列的重大升级：
核心改进： - 速度提升4倍：从"等得花儿都谢了"到"几乎即时" - 指令遵循更精准：终于能理解"把左边那个人的帽子换成红色" - 一致性维护：面部特征、光照条件在多次编辑中保持稳定 - 精确编辑：支持局部修改，不再"牵一发动全身"
产品形态： ChatGPT侧边栏新增专用"Images"入口，像创意工作室一样操作： - 预设滤镜快速调整风格 - 提示词模板降低使用门槛 - 历史版本对比方便迭代
用户反馈：
"生成更快、更精确，迭代编辑终于靠谱了。" "以前改个细节要重新生成整张图，现在真的能'编辑'了。"
12月17日，Google发布Gemini 2.0 Flash，基于Gemini 2.0架构的轻量高效版本：
定价策略： | 项目 | 价格 | |------|------| | 输入 | $0.50/百万token | | 输出 | $1.50/百万token |
这个价格意味着什么？处理一本10万字的书，成本不到1美分。
性能表现： - 速度比Gemini 2.0 Pro快3倍 - 性能达到Pro版的90%以上 - 已成为Gemini App和Google Search AI模式的默认模型
企业采用： Figma、Salesforce等已快速集成，用于： - 实时文档分析 - 客户服务自动化 - 代码辅助生成
用户评价：
"性价比高、响应飞快，企业级应用的首选。" "在搜索集成中更自然、更准确。"
12月19日，Cursor宣布收购AI代码审查初创公司Graphite：
Graphite带来什么： - Bugbot：自动检测代码缺陷 - 堆叠PR：更智能的代码审查流程 - 协作增强：团队代码审查效率提升
战略意义： Cursor从"写代码"延伸到"审代码"，形成完整的开发闭环。这不是简单的功能叠加，而是开发者工作流的深度整合。
开发者期待：
"代码审查更智能、协作更顺畅，Cursor越来越像完整的IDE了。"
12月19日，OpenAI推出个性化设置更新，用户可调整：
| 设置项 | 说明 |
|---|---|
| 温暖度 | 从专业冷静到友好亲切 |
| 热情度 | 回复的积极程度 |
| 表情符号 | 使用频率控制 |
| 前置短语 | 减少"当然！"等开场白 |
用户反馈：
"终于能让ChatGPT少用表情符号、更直接了。" "调成专业模式后，回复质量明显提升。"
12月18日，Anthropic宣布Skills功能重大更新：
企业级增强： - 技能构建更简单 - 支持Notion、Canva、Figma、Atlassian等工具集成 - 技能发现机制优化
开放标准化： 最重要的是，Agent Skills采用开放标准，可跨平台移植： - 在Claude构建的技能 - 可直接用于ChatGPT - 也能在Cursor中运行
用户评价：
"终于能跨模型复用自定义技能，不用每个平台重新配置了。"
同日，Anthropic发布安全与福祉更新： - 加强自杀/自害对话处理机制 - 减少过度讨好（sycophancy）倾向 - 发布AI素养教育资源
用户感受：
"对话更自然、不那么油腻了。"
本周数据显示，ChatGPT企业版消息量同比增长8倍，主要驱动因素： - 代理功能成熟度提升 - API稳定性改善 - 企业级安全合规完善
| 平台 | 月费 | 本周亮点 | 适合场景 |
|---|---|---|---|
| ChatGPT Plus | $20 | 图像生成大升级 | 创意工作者 |
| Gemini Advanced | $20 | Flash成默认 | 成本敏感型 |
| Claude Pro | $20 | Skills开放 | 企业集成 |
| Cursor Pro | $20 | 收购Graphite | 专业开发 |
本周传出OpenAI新一轮融资消息，潜在估值高达8300亿美元。这个数字意味着： - 超过绝大多数上市科技公司 - AI赛道资本热度不减 - 但也引发泡沫担忧
ChatGPT个性化设置虽受欢迎，但部分用户希望更多控制： - 完全关闭某些前置短语 - 更细粒度的风格调整 - 记忆与个性化的平衡
本周重要发布： - OpenAI GPT Image 1.5 - Google Gemini 2.0 Flash - Anthropic Skills更新 - Cursor收购Graphite
数据来源： - 各平台官方公告 - 用户社区反馈 - 行业分析报告
获取最新AI动态，关注我们： - 网站：www.80aj.com - 微信群：AI路透社交流群
下期预告：2025年AI年度盘点，回顾这一年的重大突破！
本文为AI路透社原创内容，转载请注明出处。
In a new video featuring renowned physicist Brian Greene in conversation with Edward Witten, they reflect on the nearly 50-year development of string theory. Witten admits that when the 'anthropic landscape' theory emerged, he felt extremely uneasy and troubled, as it contradicted his original purpose as a physicist—to explain the properties of fundamental particles, rather than accepting their dependence on the choice of classical solutions. Witten expressed that this realization made him 'very unhappy for many years,' ultimately forcing him to accept the reality that the universe was not created for human understanding. The article explores the challenges facing string theory and the potential choices scientists may face when theories fail. This conversation not only reveals the historical context of string theory's development but also showcases the authentic journey of top scientists when confronting theoretical crises, holding significant value for understanding the essence of scientific exploration and scientific methodology.
Original Link:Hacker News
Immersa is an innovative open-source web tool designed for creating and displaying three-dimensional presentation content. Users can build interactive 3D slides directly through their browsers without installing additional software, suitable for various scenarios including education, business, and creative presentations. The project is hosted on the GitHub platform, encouraging the global developer community to contribute code, jointly optimize features, and expand application scope. As a cutting-edge technology tool, Immersa lowers the barrier to 3D content creation, drives innovative applications of web technology in immersive experiences, and provides users with efficient and flexible presentation solutions.
Original Link:Hacker News
This article shares a localized method for removing Google SynthID's invisible watermark. By using an img2img model to repaint the image and adjusting the denoise parameter to 0.2, the watermark can be effectively removed while preserving the original image content. The author provides a ComfyUI workflow link for readers to implement the technique. This method supports completely local processing without requiring an internet connection, emphasizing its use for learning and research purposes. The article details the basic principles of SynthID watermark technology and explains how to remove the watermark through image repainting techniques. The author emphasizes that this is not the optimal method but provides a practical workflow. The content includes before-and-after comparisons and invites community discussion on optimization solutions. A disclaimer reminds readers to use this for learning and research only, avoiding misuse. This sharing is of significant reference value for researchers focused on AI-generated content security.
Original Link:Linux.do
This article shares the author's research findings on Google SynthID invisible watermark technology. Through reverse engineering, the author successfully identified watermark features and attempted to implement localized watermarking functionality. Although the current results are incomplete, it provides valuable technical exploration ideas and implementation directions. The article includes technical detail sharing and a disclaimer, emphasizing that the research is for learning purposes only. This research is significant for understanding the identification and traceability technology of AI-generated content, and has reference value for technical personnel and researchers in related fields.
Original Link:Linux.do
Google Antigravity service has recently undergone a major upgrade, now fully supporting the business versions of Gemini Pro and Ultra plans. This update, clearly stated in the changelog, provides Google Workspace AI Ultra for Business subscribers with higher rate limits, and these limits refresh at a faster rate, significantly improving the AI service's response speed and processing capabilities. For enterprise users relying on AI tools, this means a smoother operational experience and increased productivity. Additionally, market reactions indicate that prices for business version Ultra accounts on second-hand trading platforms like Xianyu may rise due to this upgrade, suggesting the enhanced service's popularity and potential value. This update not only demonstrates Google's continuous innovation in AI technology but also provides more powerful tool support for business users.
Original Link:Linux.do
最新评论
I don't think the title of your article matches the content lol. Just kidding, mainly because I had some doubts after reading the article.
这个AI状态研究很深入，数据量也很大，很有参考价值。
我偶尔阅读 这个旅游网站。激励人心查看路线。
文章内容很有深度，AI模型的发展趋势值得关注。
内容丰富，对未来趋势分析得挺到位的。
Thank you for your sharing. I am worried that I lack creative ideas. It is your article that makes me full of hope. Thank you. But, I have a question, can you help me?
光纤技术真厉害，文章解析得挺透彻的。
文章内容很实用，想了解更多相关技巧。