LeoDailyIngest

KO | EN

gitlite — search

#mcp #github #json #typescript #nodejs #web #gemini #large-language-models #agents #codex

(★ 15)

我的每天摄入信息的一部分，跟我一起一起大量摄入信息，构建自己的认知体系，增加自己的视野和判断能力

LeoDailyIngest Latest Version Download

ZIP Download

# Use via CDN

jsDelivr

jsDelivr serves any public GitHub repository as a CDN with zero setup. Pick a version and a file to get a ready-to-paste link and snippet.

Version

File path

Link

Example

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

我的每天摄入信息的一部分，跟我一起一起大量摄入信息，构建自己的认知体系，增加自己的视野和判断能力

# 2025-11-27

- [https://web.stanford.edu/class/cs234/](https://web.stanford.edu/class/cs234/)：吃灰课程，上次的课都还没看完！
- [https://weaponizedspaces.substack.com/p/x-just-accidentally-exposed-a-vast](https://weaponizedspaces.substack.com/p/x-just-accidentally-exposed-a-vast)：我只能说这次X显示国家的Feat太精彩了，我已经听到了好几个有趣的现象了

# 2025-11-24

- [https://strangestloop.io/essays/things-that-arent-doing-the-thing](https://strangestloop.io/essays/things-that-arent-doing-the-thing)：现在马上开始。很多时候冲动去做一个事情，反而是最好的，我发现很多时候很多事情，从人性的角度来看是不能持续做很久的，也很容易被其他的想法给取代，所以最好的就是现在开始做，不用在开始就追求完美，just for fun是一个很好的理由开始
- [https://github.com/htx-studio/One-Handed-Keyboard](https://github.com/htx-studio/One-Handed-Keyboard)：播客里有听到htx找人在量产这个键盘了，看着也开源了，值得点赞
- [https://searchepsteinfiles.com/](https://searchepsteinfiles.com/)
- [https://blog.ctms.me/posts/2025-11-14-being-poor-or-being-broke/](https://blog.ctms.me/posts/2025-11-14-being-poor-or-being-broke/)：穷人之殇
- [https://github.com/ZHZisZZ/dllm](https://github.com/ZHZisZZ/dllm)：可用于教学目的

# 2025-11-21

- [https://zed.dev/blog/zed-is-our-office](https://zed.dev/blog/zed-is-our-office)：Zed的官网好清爽，阅读体感很不错
- [https://unsplash.com/awards/2025/finalists](https://unsplash.com/awards/2025/finalists)：Unsplash Awards 2025结果出来了，太棒了，不仅仅是拍的照片，更是配在照片旁边的文字

# 2025-11-20

- [https://huijzer.xyz/posts/123/do-not-put-your-site-behind-cloudflare-if-you-dont](https://huijzer.xyz/posts/123/do-not-put-your-site-behind-cloudflare-if-you-dont)：CF挂6hr就PTSD了~但是我感觉在CF后不一定为了应对DDOS，免费CDN本身也是一个很大的诱因
- [https://go.dev/blog/greenteagc](https://go.dev/blog/greenteagc)：写得特别好的一篇文章，Go的Blog Post经常有一些高质量的内容。我还是坚信可视化永远是教育里最牛逼的手段

# 2025-11-18

- [https://andreacanton.dev/posts/2025-11-08-always-ready-to-leave/](https://andreacanton.dev/posts/2025-11-08-always-ready-to-leave/)：有一些启发，想要把一些东西通过文档化留下来
- [https://github.com/jonasstrehle/supercookie](https://github.com/jonasstrehle/supercookie)：很有趣的想法，用favicon来替代cookie追踪不同用户
- [https://brainbaking.com/post/2025/10/the-internet-is-no-longer-a-safe-haven/](https://brainbaking.com/post/2025/10/the-internet-is-no-longer-a-safe-haven/)：个人需要花很多effort去阻断AI bot，或许接入CF是个更好的选择
- [https://djnn.sh/posts/anthropic-s-paper-smells-like-bullshit/](https://djnn.sh/posts/anthropic-s-paper-smells-like-bullshit/)：写得很好，这篇文章，三观也很正。外行好忽悠，内行一下点破。现在反观Anthropic的报告，确实bullshit，观感持续下降。引用一段：What is that even supposed to mean ? You claim your agents were able to find exploitable vulnerabilities in multiple services. Were these patched ? What about the extracted data ? What about the affected people ? Do you care about this at all ?
- [https://latenitesoft.com/blog/evaluating-frontier-ai-image-generation-models/](https://latenitesoft.com/blog/evaluating-frontier-ai-image-generation-models/)：Gemini（Nanobanana）真的最现实，效果真好啊。反而我觉得OAI不太好，AI感很强，即梦确实也很棒

# 2025-11-17

- [https://www.linkedin.com/pulse/positive-sum-future-satya-nadella-bjs7c/](https://www.linkedin.com/pulse/positive-sum-future-satya-nadella-bjs7c/)：倒是也没多大触动，感觉文笔一般

# 2025-11-16

- [https://open.substack.com/pub/bytebytego/p/how-perplexity-built-an-ai-google](https://open.substack.com/pub/bytebytego/p/how-perplexity-built-an-ai-google?r=2eqm57&utm_medium=ios)：从整体的角度再看看perplexity如何基于rag衍生出这么一个产品的
- [https://gpuleaderboard.com/](https://gpuleaderboard.com/)：对比GPU提供商的价格

# 2025-11-15

- [https://open.substack.com/pub/henrythe9th/p/why-the-best-ai-companies-look-more](https://open.substack.com/pub/henrythe9th/p/why-the-best-ai-companies-look-more?r=2eqm57&utm_medium=ios)：现在AI初创和传统的初创差异，速度不同，创始人背景也不一定是技术出身，vc看投资的思路也需要改变

# 2025-11-14

- [https://how-did-i-get-here.net/](https://how-did-i-get-here.net/)：可视化YYDS，要是以前刚学习网络和traceroute有这个文章就能加速理解了。虽然现在有AI，但是可视化还是一个极其重要的教学手段

# 2025-11-12

- [https://simonw.substack.com/p/code-research-projects-with-async](https://simonw.substack.com/p/code-research-projects-with-async)：我觉得是一个值得尝试的，Simon会借助cc codex之类的去做research，然后归档到某个repo，全程只需要提交问题，异步处理。make sense。他的三观也很正，希望能有feat可以让这ai slop仓库可以不被索引

# 2025-11-11

- [https://drfeifei.substack.com/p/from-words-to-worlds-spatial-intelligence](https://drfeifei.substack.com/p/from-words-to-worlds-spatial-intelligence)：李飞飞写得关于空间智能和世界模型，看着也是她团队未来探索的方向了，乐见这种有价值的频道出现

# 2025-11-09

- [https://www.xiaoyuzhoufm.com/episode/6908b9c80ceab2a71c48668c?s=eyJ1IjogIjY4Mzk3OTM0ZDFkMzUwNzI2OWRiOTQ4NCJ9](https://www.xiaoyuzhoufm.com/episode/6908b9c80ceab2a71c48668c?s=eyJ1IjogIjY4Mzk3OTM0ZDFkMzUwNzI2OWRiOTQ4NCJ9)：呼应了昨天那篇文章

# 2025-11-08

- [https://magazine.sebastianraschka.com/p/beyond-standard-llms](https://magazine.sebastianraschka.com/p/beyond-standard-llms)：写得太好了，讨论了Transformer全注意力之外的其他注意力机制的发展

# 2025-11-07

- [https://cognition.ai/blog/codemaps](https://cognition.ai/blog/codemaps)：代码仓库地图，有点简化版本的deepwiki给到coding agent
- [https://inventingthefuture.ghost.io/mr-tiff/](https://inventingthefuture.ghost.io/mr-tiff/)：认真并且持续去坚持做一件事，往往会让人动容
- [https://artificialanalysis.ai/downloads/state-of-ai/2025/Q3-2025-Artificial-Analysis-State-of-AI-Highlights-Report.pdf](https://artificialanalysis.ai/downloads/state-of-ai/2025/Q3-2025-Artificial-Analysis-State-of-AI-Highlights-Report.pdf)：2025Q3的AI报告

# 2025-11-06

- [https://www.anthropic.com/engineering/code-execution-with-mcp](https://www.anthropic.com/engineering/code-execution-with-mcp)：MCP卸载到磁盘目录里，用Code Exec来驱动，也是个思路，也可以进一步借用Skills
- [https://blog.cloudflare.com/code-mode/](https://blog.cloudflare.com/code-mode/)：CF版本的，也是借助了自家的Worker来做sandbox执行
- [https://www.marktechpost.com/2025/11/02/comparing-the-top-6-ocr-optical-character-recognition-models-systems-in-2025/](https://www.marktechpost.com/2025/11/02/comparing-the-top-6-ocr-optical-character-recognition-models-systems-in-2025/)：6个OCR模型对比，小红书的dots.ocr也不错

# 2025-11-05

- [https://www.decodingai.com/p/getting-agent-architecture-right](https://www.decodingai.com/p/getting-agent-architecture-right)：agent pattern好文章

# 2025-11-04

- [https://words.filippo.io/claude-debugging/](https://words.filippo.io/claude-debugging/)：引用一句原文的话：I am sharing this because it made me realize I still don’t have a good intuition for when to invoke AI tools, and because I think it’s a fantastic case study for anyone who’s still skeptical about their usefulness. 很多时候我们忽略了AI能在某些我们意料不到的方面帮助我们，别尬吹自己的ai native player了，真正的native并不会说，而是直接做，就好像互联网长大的一代觉得互联网是很正常的东西，没有任何hype在里面，AI也是
- [https://artificialanalysis.ai/?intelligence=artificial-analysis-intelligence-index](https://artificialanalysis.ai/?intelligence=artificial-analysis-intelligence-index)：对比一下模型的情况不错，甚至还有一些类似timeline的展示，之前还想让ai收集撸一个，现在省事了
- [https://emily.space/posts/251023-uv](https://emily.space/posts/251023-uv)：astral团队真牛逼，正在推动team对齐uv+ruff+ty

# 2025-11-02

- [https://www.forbes.com/sites/lanceeliot/2025/11/01/prompt-engineering-newest-technique-is-verbalized-sampling-that-stirs-ai-to-be-free-thinking-and-improve-your-responses/](https://www.forbes.com/sites/lanceeliot/2025/11/01/prompt-engineering-newest-technique-is-verbalized-sampling-that-stirs-ai-to-be-free-thinking-and-improve-your-responses/)：提示词技巧，和Few Shot之类的技巧类似，感觉并没有什么新奇或更可取的方式，让LLMs给出多种可能和概率，并不太能代表是模型内部真是的置信度。
- [https://danieldelaney.net/normal/](https://danieldelaney.net/normal/)：有点less is more的感觉了，简单易用≥庞大复杂
- [https://news.ycombinator.com/item?id=45711816](https://news.ycombinator.com/item?id=45711816)：评论区有一些不错的建议，HN有时候还是很正能量的
- [https://huggingface.co/blog/openenv](https://huggingface.co/blog/openenv)：HF和Meta发布OpenEnv Hub，标准化Agent环境用于训练和部署智能体
- [https://www.nvidia.com/en-us/learn/learning-path/deep-learning/](https://www.nvidia.com/en-us/learn/learning-path/deep-learning/)：英伟达发的Deep Learning的学习路径，继续吃灰

# 2025-10-30

- [https://x.com/cursor_ai/status/1983567621602881992](https://x.com/cursor_ai/status/1983567621602881992)：Cursor也有自己的模型了，还是要走出卡脖子
- [https://pokerbattle.ai/event](https://pokerbattle.ai/event)：和ai trade类似，一群AI在打扑克
- [https://512pixels.net/2025/10/boring-is-what-we-wanted/](https://512pixels.net/2025/10/boring-is-what-we-wanted/)：或许是吧，但是这两年苹果软件肉眼程度的有各种毛病出现，是不可忽略的
- [https://blog.langchain.com/doubling-down-on-deepagents/](https://blog.langchain.com/doubling-down-on-deepagents/)：langchain不断迭代deepagents了
- [https://www.lesswrong.com/posts/5jjk4CDnj9tA7ugxr/openai-email-archives-from-musk-v-altman-and-openai-blog](https://www.lesswrong.com/posts/5jjk4CDnj9tA7ugxr/openai-email-archives-from-musk-v-altman-and-openai-blog)：之前就看到，一直遗落在角落，看了下大佬间的对话，还是很有趣的

# 2025-10-29

- [https://www.youtube.com/playlist?list=PLZHnYvH1qtOZCUU5UAui608MaZN_60dmK](https://www.youtube.com/playlist?list=PLZHnYvH1qtOZCUU5UAui608MaZN_60dmK)：Nvidia GTC 2025，非常值得一看。每次看完GTC，对AI泡沫的感觉又少了一点，感叹人类科技的发展。另外Nvidia的Creative Team做的视频是真的很有料，点赞
- [https://alignment.anthropic.com/2025/stress-testing-model-specs/](https://alignment.anthropic.com/2025/stress-testing-model-specs/)：比较深，AI辅助下看完了
- [https://www.postman.com/state-of-api/2025](https://www.postman.com/state-of-api/2025)：postman的AI报告，主要针对API的
- [https://aitinkerers.org/p/welcome](https://aitinkerers.org/p/welcome)：AI Tinkers，如果是在那边肯定去加入，去玩了

# 2025-10-28

- [https://github.com/yandexdataschool/nlp_course](https://github.com/yandexdataschool/nlp_course)：这两天在看斯坦福的AI课，有人给我推荐了这个，感觉不错，先吃灰，后面可以看看
- [https://news.mit.edu/2025/3-questions-pros-cons-synthetic-data-ai-kalyan-veeramachaneni-0903](https://news.mit.edu/2025/3-questions-pros-cons-synthetic-data-ai-kalyan-veeramachaneni-0903)：2024年60%用于AI应用的数据是合成数据。合成数据适用于：软件和性能测试，训练模型。
- [https://menlovc.com/perspective/2025-the-state-of-ai-in-healthcare/](https://menlovc.com/perspective/2025-the-state-of-ai-in-healthcare/)：医疗行业发展这么猛，没有关注就很容易忽略，还是要在金融、教育和医疗这些主流行业里多关注一下行业趋势
- [https://atlassianblog.wpengine.com/wp-content/uploads/2025/09/atlassian-ai-collaboration-report-2025.pdf](https://atlassianblog.wpengine.com/wp-content/uploads/2025/09/atlassian-ai-collaboration-report-2025.pdf)：Atlassian的AI报告（全球Fortune 1000的高管与1.2万名知识工作者的研究结果），大概讲了AI虽然提升了个人生产力，但企业层面没有特别显著的收益提现，报告很长，不过很多不错的图表可以比较直观的查看，翻看一下图表也能得到不错的insight

# 2025-10-27

- [https://consumed.today/](https://consumed.today/)：这个repo就是受这个网站影响的
- [https://www.nan.fyi/database](https://www.nan.fyi/database)：可视化教育是王道。这篇可以让任何人了解数据库背后的东西，虽然讲的是针对KV的，但是依然可以看到很多MySQL之类的数据库也是站在这样的原理之上设计的。教学最好的方法就是循环渐进，讲为什么引入这个技术或者做这个改变，而不是一上来就说结论和现状，后者会更倾向让人死记硬背，前者可以引发思考，甚至未来还有能力去推动创新
- [https://www.anthropic.com/engineering/building-effective-agents](https://www.anthropic.com/engineering/building-effective-agents)：Anthropic讲Workflow和Agent的文章，质量很高
- [https://maurycyz.com/misc/the_cost_of_trash/](https://maurycyz.com/misc/the_cost_of_trash/)：个人相对抗爬虫的成本特别高，尤其现在AI加持下对数据的渴望。某些b端场景下也会因此有很大的经济损失

# 2025-10-26

- [https://github.com/anthropic-experimental/sandbox-runtime](https://github.com/anthropic-experimental/sandbox-runtime)：anthropic开源的sandbox runtime，不需要容器，直接对进程施加文件和网络限制
- [https://cme295.stanford.edu/](https://cme295.stanford.edu/)：斯坦福的大模型课程，吃灰+1。要是学生时代就可以爽吃一波了
- [https://blog.langchain.com/agent-frameworks-runtimes-and-harnesses-oh-my/](https://blog.langchain.com/agent-frameworks-runtimes-and-harnesses-oh-my/)：DeepAgents流行，值得关注
- [https://www.vtrivedy.com/posts/claude-code-sdk-haas-harness-as-a-service](https://www.vtrivedy.com/posts/claude-code-sdk-haas-harness-as-a-service)：以claude code sdk出发，介绍了harness和haas

# 2025-10-25

- [https://www.dwarkesh.com/p/andrej-karpathy](https://www.dwarkesh.com/p/andrej-karpathy)：还在听，倒是没有期望获得很多东西，单纯用练英语听力的角度来听
- [https://x.com/karpathy/status/1979644538185752935](https://x.com/karpathy/status/1979644538185752935)：现在自媒体很喜欢标题党，上面这个访谈很多人就说karpathy唱衰ai，karpathy自己有特意发了推出来澄清一些事实。或许本身也是在oai利益相关方，需要一定的公关挽一下。理性看待，AI一直都不是二元对立的，这个世界也是，不要让偏见遮蔽自己的双眼
- [https://news.ycombinator.com/item?id=45613567](https://news.ycombinator.com/item?id=45613567)：防御爬虫日益成为一个需要重视的问题了
- [https://fullstack.rocks/article/sqlalchemy/brewing_with_sqlalchemy](https://fullstack.rocks/article/sqlalchemy/brewing_with_sqlalchemy)：SQLAlchemy系列教程，吃灰去吧
- [https://github.com/virattt/dexter](https://github.com/virattt/dexter)：financial research agent，有需要的时候可以看看代码
- [https://github.com/opendatalab/MinerU](https://github.com/opendatalab/MinerU)：PDF文档提取，想和rapidocr，docling，paddle对比一下
- [https://blog.baro.dev/p/the-future-of-python-web-services-looks-gil-free](https://blog.baro.dev/p/the-future-of-python-web-services-looks-gil-free)：Python3.14没有GIL了，对比了一下FastAPI（ASGI）和Flask（WSGI）的影响，只给了表格，感觉如果可以给出图进行可视化对比会更加直观
- [https://open.substack.com/pub/artificialintelligencemadesimple/p/the-ai-is-a-bubble-narrative-is-stupid](https://open.substack.com/pub/artificialintelligencemadesimple/p/the-ai-is-a-bubble-narrative-is-stupid)：非常精彩的分析了围绕nvidia、oai最近的循环投资问题，直指nvidia走向垄断的路径

# 2025-10-24

- [https://github.com/iFurySt/AI-EBooks](https://github.com/iFurySt/AI-EBooks)：Galileo的4本AI相关的电子书，吃灰去吧
- [https://www.youtube.com/watch?v=g7EGMWvJ1fI](https://www.youtube.com/watch?v=g7EGMWvJ1fI)：用Registry Pattern的方式来取代if elif链，在某些时候非常make sense，比如注册一些服务、工具或agent
- [https://github.com/Pokee-AI/PokeeResearchOSS](https://github.com/Pokee-AI/PokeeResearchOSS)：小参数模型做deepsearch，mark一下
- [https://seed.bytedance.com/en/seed3d](https://seed.bytedance.com/en/seed3d)：字节新出的image to 3D的模型
- [https://open.substack.com/pub/bytebytego/p/the-memory-problem-why-llms-sometimes](https://open.substack.com/pub/bytebytego/p/the-memory-problem-why-llms-sometimes?r=2eqm57&utm_medium=ios)：大模型窗口大小问题，简单科普属于，没太多深度

# 2025-10-23

- [https://karunpal.substack.com/p/why-a-boring-life-might-be-the-happiest](https://karunpal.substack.com/p/why-a-boring-life-might-be-the-happiest)：很peace的感觉，我坚信我就是很rush也很workaholic的人，有时候用一些citywalk、reading、painting来对冲。或许文章想表达的是每种人生都是值得过的，只要你喜欢，无需用他人的标准来衡量自己。The truth is: there are many like you. Many like me. Living quietly. Doing what they like. Finding peace in repetition. A simple boring life may not be shiny and exciting, but it’s peaceful.
- [https://nathan.rs/posts/roberta-diffusion/](https://nathan.rs/posts/roberta-diffusion/)：很有趣的观点，BERT通过多步掩码训练，BERT可以约等于扩散模型，也能具备GPT一样的生成能力
- [https://neal.fun/space-elevator/](https://neal.fun/space-elevator/)：可视化每个高度可能出现的东西，原来mallard居然能飞，还能飞这么高…帝王蝶和大黄蜂能飞那么高也很惊奇。滑倒最后只能感叹自然的伟大和人类的牛逼
- [https://blog.abdellatif.io/production-rag-processing-5m-documents](https://blog.abdellatif.io/production-rag-processing-5m-documents)：针对RAG非常make sense的建议
- [https://blog.daviddodda.com/how-i-almost-got-hacked-by-a-job-interview](https://blog.daviddodda.com/how-i-almost-got-hacked-by-a-job-interview)：或许可以叫做DeepScam了，简单的一句话+cursor，拯救了一切，不管实际的AI bubble多大，LLMs已经深度集成到我们的生活中了
- [https://ditto.xiaohongshu.net/ditto/vincent/a49284f45b184e2b8f79da0670943256/dynpath250272a8?fullscreen=true&resource_instance_id=255875&naviHidden=yes&source=share&page_source=search_result_notes.onebox&search_id=2fhp8piuu32p9qmnd3h3b&isOverseaData=0&isOverseaRegister=0&lang=zh_cn&xhs-statusbar-height=54&instance_id=a49284f45b184e2b8f79da0670943256&page_instance=2683](https://ditto.xiaohongshu.net/ditto/vincent/a49284f45b184e2b8f79da0670943256/dynpath250272a8?fullscreen=true&resource_instance_id=255875&naviHidden=yes&source=share&page_source=search_result_notes.onebox&search_id=2fhp8piuu32p9qmnd3h3b&isOverseaData=0&isOverseaRegister=0&lang=zh_cn&xhs-statusbar-height=54&instance_id=a49284f45b184e2b8f79da0670943256&page_instance=2683)：明天去线下小局。里面很多笔记的评论干活十足，用心翻阅一下可以通过碎片化问答拼凑出很多有价值的东西，要是有人能让AI抓一下这些内容出来就好了，绝对很有价值
- [https://towardsdatascience.com/things-i-learnt-by-participating-in-genai-hackathons-over-the-past-6-months/](https://towardsdatascience.com/things-i-learnt-by-participating-in-genai-hackathons-over-the-past-6-months/)：every idea starts small，非常make sense。8/2法则也很符合现在AI Copilot的时代，做完备的思考和计划，AI辅助，是比较舒服的姿势
- [https://www.anthropic.com/engineering/building-effective-agents](https://www.anthropic.com/engineering/building-effective-agents)

# 2025-10-22

- [https://www.oreilly.com/radar/podcast/generative-ai-in-the-real-world-context-engineering-with-drew-breunig/](https://www.oreilly.com/radar/podcast/generative-ai-in-the-real-world-context-engineering-with-drew-breunig/)：Drew Breunig大佬谈上下文工程
- [https://steipete.me/posts/just-talk-to-it](https://steipete.me/posts/just-talk-to-it)：一些与AI协作的姿势，最高效的还是直奔目的不要追求太fancy的东西，当然玩是例外

# 2025-10-21

- [https://dotslashwatch.com/](https://dotslashwatch.com/)：DIY最酷了
- [https://50centadjustedforinflation.com/](https://50centadjustedforinflation.com/)：一眼通货膨胀，笑死
- [https://planetscale.com/blog/caching](https://planetscale.com/blog/caching)：可视化YYDS，可视化讲解了缓存相关的内容
- [https://samwho.dev/big-o/](https://samwho.dev/big-o/)：又是一个YYDS的可视化，这篇介绍了算法的时间复杂度。这个人的个人站里有很多可视化的例子，非常NICE
- [https://snarky.ca/why-it-took-4-years-to-get-a-lock-files-specification/](https://snarky.ca/why-it-took-4-years-to-get-a-lock-files-specification/)：pylock.toml出现。一个规范的制定要持续很久！

# 2025-10-20

- [https://blog.bytebytego.com/p/how-salesforce-used-ai-to-reduce](https://blog.bytebytego.com/p/how-salesforce-used-ai-to-reduce)：始料未及的是，最后看起来像是在给cursor打call的感觉，整体还是概念性，没太多有趣的细节
- [https://www.philschmid.de/agents-2.0-deep-agents](https://www.philschmid.de/agents-2.0-deep-agents)：大家开始聊Deep Agent或者说Agent 2.0了，我个人理解就是重视multi-agent的方向去了
- [https://nlp.elvissaravia.com/p/deep-agents](https://nlp.elvissaravia.com/p/deep-agents)：同上
- [https://github.com/deepseek-ai/DeepSeek-OCR](https://github.com/deepseek-ai/DeepSeek-OCR)：DeepSeek-OCR 把长文本“拍成图片”，再让模型从图片里读回文字，用这种“看图读文”的方式把上下文压缩成原来十分之一，还能保持差不多的准确率
- [https://www.felixstocker.com/blog/talent](https://www.felixstocker.com/blog/talent)：努力是必要条件，但天赋决定了努力是否值得。不要因为别人的高产感到自卑，你也许只是把努力花在了不适合你的方向上

# 2025-10-19

- [https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills](https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills)：分了三级的Skills设定，感觉有点类似插件系统了，写各种Prompt和工具使用，还可以使用脚本，甚至感觉有些场景下比MCP更好用

# 2025-10-16

- [https://harrisoncramer.me/15-go-sublteties-you-may-not-already-know/](https://harrisoncramer.me/15-go-sublteties-you-may-not-already-know/)：不错的Go知识点
- [https://www.pixnapping.com/](https://www.pixnapping.com/)：同个手机可以获取另一个APP的信息，比如Google Authenticator。原理非常牛逼…也是这个叫**Pixnapping Attack**的原因，通过控制渲染像素点和一些遮罩层隔离及放大像素可以慢慢对比拼凑出渲染像素的文字是什么，这样可以在30s拿到一个2FA的密码
- [https://github.com/microsoft/edgeai-for-beginners](https://github.com/microsoft/edgeai-for-beginners)：又是教程，吃灰Again

# 2025-10-15

- [https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents)：关于上下文工程实践讲的不错，里面延伸的链接也很棒
- [https://www.elonmuskbook.org/](https://www.elonmuskbook.org/)：Naval的Newsletter里看到的
- [https://www.anthropic.com/engineering/multi-agent-research-system](https://www.anthropic.com/engineering/multi-agent-research-system)：综述了一下multi-agent在research里的优点

# 2025-10-14

- [https://arxiv.org/abs/2510.08338](https://arxiv.org/abs/2510.08338)：用LLMs替代真实用户进行调研。好奇这真的奏效么，哈哈。不过确实人类的选择会有偏差，很多时候1-5分，可能有人会偏好4-5分，有人偏好都在中间或更低。
- [https://danybittel.ch/macro.html](https://danybittel.ch/macro.html)：太酷了，结合拍照和模型和一些相关的技术，最终生成昆虫的3D模型
- [https://www.leoniemonigatti.com/blog/ai-agent-from-scratch-in-python.html](https://www.leoniemonigatti.com/blog/ai-agent-from-scratch-in-python.html)：展示如何用Python构建一个AI Agent demo，按需推进，适合新人了解

# 2025-10-13

- [https://github.com/HandsOnLLM/Hands-On-Large-Language-Models](https://github.com/HandsOnLLM/Hands-On-Large-Language-Models)：看着很不错的课程
- [https://github.com/karpathy/nanoGPT](https://github.com/karpathy/nanoGPT)：Karpathy写的NanoGPT
- [https://arxiv.org/pdf/2510.00446](https://arxiv.org/pdf/2510.00446)：通过数学计算去压缩，而不是通过LLMs进行压缩。很神奇的想法
- [https://microsoft.github.io/autogen/stable//user-guide/agentchat-user-guide/graph-flow.html](https://microsoft.github.io/autogen/stable//user-guide/agentchat-user-guide/graph-flow.html)

# 2025-10-12

- [https://www.xiaoyuzhoufm.com/episode/68e74f521bef327f3d7ddcd7](https://www.xiaoyuzhoufm.com/episode/68e74f521bef327f3d7ddcd7)：老罗和Tim的对话，信息密度挺高的。播客是经历盗窃，很有趣的观点。
- [https://docs.google.com/document/d/1rsaK53T3Lg5KoGwvf8ukOUvbELRtH-V0LnOIFDxBryE/edit?pli=1&tab=t.0](https://docs.google.com/document/d/1rsaK53T3Lg5KoGwvf8ukOUvbELRtH-V0LnOIFDxBryE/edit?pli=1&tab=t.0)：看着不错的Agent设计模式

# 2025-10-11

- [https://blog.miguelgrinberg.com/post/python-3-14-is-here-how-fast-is-it](https://blog.miguelgrinberg.com/post/python-3-14-is-here-how-fast-is-it)：不严肃但有趣的py3.14基准测试，图表对比直观且阅读友好
- [https://ikeamuseum.com/en/explore/ikea-catalogue](https://ikeamuseum.com/en/explore/ikea-catalogue)：**IKEA catalogue**，很有趣，果然视觉是显学
- [https://www.anthropic.com/research/small-samples-poison](https://www.anthropic.com/research/small-samples-poison)：在数据里参杂250个（绝对值而不是百分比）就可以污染大模型，让其按照一定行为动作，为AI攻防提供了新思路

LeoDailyIngest

Owner Repos

LeoDailyIngest Latest Version Download

# Use via CDN

Was this content helpful?

Recent Feedback

LeoDailyIngest

Owner Repos

LeoDailyIngest Latest Version Download

# Use via CDN

Was this content helpful?

Recent Feedback

Download README