Edward Welly

observer

[Raytar @Raytar]

Andrej Karpathy joined Anthropic five weeks ago.

Yesterday my friend on his team sent me the Claude.md file he actually uses.

It completely changed how I work with Claude.
From the very first message, the difference was obvious.

With this file, Claude finally stops fighting me and starts working exactly the way I need it to.

Bookmark it before it gets taken down.

Read it now, then check the article below.

[Aarno @TheGlobalMinima]

In nearly 5 years of modern generative ai, this is the first book I’m seeing with a super high level of coverage and comprehension.
> language modelling
> inference optimisation
> RL and its methods
> system scaling
> applied concepts like agentic ai, rag, memory
> environments and benchmarking

These fields have a subtle boundary differentiating them, but ultimately overlap in modern applications. Agents require system scaling, memory needs inference optimisation, rl requires understanding of environments and benchmarks.

For the first time in my exp, all in one place. Found this on paperswithcode[.]co

[Leaf Yeah! @leaf_sanren]

https://t.co/izjIG6IhON

[Kyrie @KyrieCheungYep]

https://t.co/aHRblOHcMn

[阿蔺A-Lin @alin_zone]

强烈建议每个使用 Obsidian 的朋友们都去搭建自己的第二大脑！

无需害怕不知道怎么搭建，现成的 Obsidian 和 Claude Code 的自组织 AI 第二大脑他来了！

只需添加任何源代码，Claude 即可读取、链接并将其归档到您拥有的纯 Markdown 格式的互联知识图谱中。它集 AI 笔记、个人知识管理 (PKM) 和开源 Notion 替代方案于一体。基于 Karpathy 的 LLM Wiki 模式。

以后第二大脑不仅仅是 notion 的专属了，Obsidian 也可以做到了。

[宝玉 @dotey]

PPT Master 确实是最好的 PPT Skill

我新的 skill 写PPT也挺好，能导出可编辑版本，可以AI配图，可以在 Agent 内置浏览器中标记编辑

https://t.co/POuwOWSuwe

[Eric Topol @EricTopol]

We stress tested many frontier AI models for multimodal medical reasoning (including GPT-5, Claude 3.5, Gemini 2.5 Pro). They’re not ready. Faulty reasoning, use of inappropriate shortcuts, hallucinations. Published today @NatureMedicine https://t.co/P6eHZEmfbW

[Steven Artandi, MD, PhD @SCIDirector]

Research from SCI members @GarryPNolan, @GuolanLu, and colleagues introduces single-cell spatial pharmacobiology, revealing how stromal barriers can limit antibody delivery and inform precision cancer therapies. https://t.co/rv5kn3wQRP

[Amto @XAMTO_AI]

AI Agent 的记忆问题，腾讯刚扔了一个专攻方案：TencentDB Agent Memory，正式开源。

它只做一件事：给 Agent 同时装上长期记忆和短期记忆。

效果怎么？
长期记忆准确率 47.85% → 76.10%，提升近 60%。
用户事实召回 不到30% → 79%。
长任务 Token 消耗，最高省 61%。
思路跟主流不一样。
它走的是：符号化短期记忆 + 分层长期记忆。
四层渐进式架构，L0 → L3，从底到顶，各干各的。
L0：原始对话，全量留底，一字不差。
L1：原子事实，自动抽“爱吃火锅”“用 NextJS”这类节点，打标签存。
L2：场景聚类，按主题把事实合成可读的 Markdown 场景块。
L3：用户画像，沉淀技术偏好、代码风格、常用工具链。

亮点不在分层，在可追溯。
说用户偏好 TypeScript，能从 L3 追到 L2 场景块，再追到 L1 原子事实，最后追到 L0 原话。
证据链不断。
🔗：https://t.co/NLGS1h9ECx

[AYi @AYi_AInotes]

真的巨牛逼，手机也能加载，就是慢一些，肯定是鼠标键盘更爽
https://t.co/4KRYogKQ8v

[Petrichor @Jam79922967]

《Nature》2006那篇神作，是典型的作假的“论文”，被引2300余次，把阿尔兹海默症研究带偏，白白浪费世界各国总共16.5亿美元的经费。那篇论文说一个叫Aβ*56的寡聚体，是阿尔兹海默症的元凶。那篇论文里的条带图，漂亮得挑不出毛病，相关性能做到0.98。现在知道了，图是拼的，条带是修的。Aβ*56是假的。过去二十年里，围绕这个假说开发的药物，全失败了。146项临床试验折戟沉沙。真正的科学，应该允许失败，十六年。一代人的科研生涯。16亿美元，代价太大了。那篇假论文的作者们不该坐牢吗？

[Kexin Huang @KexinHuang5]

Excited to launch Biomni everywhere: web, desktop, mobile, and MCP!

Our vision is simple: biological research should happen wherever inspiration strikes. Whether you’re at your bench or riding the subway, Biomni is there when the idea hits. Start a new binder from your phone, then continue seamlessly across devices.

We're also launching Biomni MCP, so you can access Biomni’s biology capabilities wherever you work:

[Paolo Tarantino @PTarantinoMD]

BREAKING: The most controversial article of the year, claiming that early morning immunotherapy works better than in the afternoon, is now retracted.

After reading the responses provided by the authors to the inconsistencies raised in the web, the @NatureMedicine editors no longer have confidence in the integrity of the results. The only prospective evidence that time-of-day matters for immunotherapy is now gone.

https://t.co/aXUY6aekhl

To me, this means (at least) two things.

First, it confirms that prudence on this topic was and remains critical. For as inexpensive it may be to give a drug earlier or later in the day, it carries a much more relevant cost: the one of scientific integrity. We owe our patients to make decisions based on solid data. We should not give up this practice too easily, particularly in the presence of several concerning red flags.

Second, this retraction should also prompt a broader reflection on the current state of peer review, in which unpaid reviewers struggle to keep up with a steady rise in submitted papers. Journals need to improve the process by implementing a formal, consistent, in-depth review of each paper by paid professionals. A practice that, in this case, may have avoided a retraction arriving after 22 citations and after inclusion of this study in at least one meta-analysis. And possibly, after some physicians had already changed their practice in IO administration.

For a thoughtful recap of this story, I recommend this well-written new piece in @ScienceMagazine by Laura Agudelo. I’m grateful to Laura for including my perspective in the article.

https://t.co/qHM5fMjwQ3

[Tanishq Mathew Abraham, Ph.D. @iScienceLuvr]

Autodata: An agentic data scientist to create high quality synthetic data

"We introduce Autodata, a general method that enables AI agents to act as data scientists who build high quality training and evaluation data."

Data creation stage + data analysis stage+meta-optimization

[Crémieux @cremieuxrecueil]

RETRACTED!

TL;DR: This study, which claimed the time of day had massive effects on immunotherapy efficacy, but which appeared fraudulent, now seems to certainly have been fraudulent.

Many thanks to the editors at Nature for handling this quickly and correctly.

[nini @nini_incrypto_]

论文写作Skill推荐

1 Research-Paper-Writing-Skills
https://t.co/RveTFYvhbh
这是一个面向 机器学习/计算机视觉/NLP 论文写作 的开源skill 包，适合配合 Codex、Claude Code、Gemini 等AI编程/写作助手使用。它主要帮助你规范论文结构、优化摘要、Introduction、Related Work、Method、Experiment等章节的写法，也适合用来润色 SCI/会议论文。

2 sciwrite
https://t.co/r2oaDZpOwg
这是一个用于 AI辅助科学论文写作审阅 的开源 Skill，基于 Kristin Sainani 的Writing in the Sciences 写作方法，主要适合检查论文表达是否清晰、逻辑是否顺、句子是否啰嗦。适合用来做论文润色、段落审查、逻辑修改，尤其适合SCI 论文初稿修改。

3 Al Research Skills Library
https://t.co/YAKWqC1LTI
这是一个面向 AI科研流程的开源 Skills 库，覆盖 文献调研、想法生成、实验执行、论文写作 等环节，不只是写作，而是完整科研流程辅助。

4 Research Writing Assistant
https://t.co/cIBUWUaeJa
这是一个中文科研写作 Skill，目标是把论文写作从一次性对话变成 可追踪、可恢复、可复用的工程化协作流程，比较适合本科生、研究生和早期科研人员。中文说明更友好，适合写 毕业论文、课程论文、投稿初稿。

[Pat Simmons @per_simmons_]

Claude just became a craacked video game designer.

With the launch of Unreal Engine's MCP server last week, you can now build entire video games just by talking to Claude.

I spent the past few days building with it, and I'm telling you, this is going to forever change how video games get made and who gets to make them.

In this video I show you exactly how to set up the Unreal Engine MCP yourself and run through three demos: building a full playable city, cloning a real city from Google Earth, and creating custom buildings in Blender.

Here's the agent harness I mention too: https://t.co/mos9EwnZ2h

Intro
What I built in a few hours
Setting up the Unreal MCP server
Fixing the port 8000 connection issue
The agent harness that avoids the pitfalls
Demo 1: Building a city with City Sample
Demo 2: Cloning a real city from Google Earth with Cesium
Demo 3: Custom buildings with Blender headless
Outro

[AIDB @ai_database]

Nature誌にてGoogle DeepMindとGoogle Researchの研究者らが報告。
AIにコードを書かせては機械で採点して書き直させる作業をひたすら繰り返させることで、人間の専門家が長年積み上げてきた最高水準の科学ソフトウェアを上回るものまで実際に自動で作れてしまうとのこと。

彼らは単一細胞解析やコロナの 入院予測など6つの科学分野で実験的に検証して報告しました。

研究チームによると、既存の手法を2つずつ組み合わせて新しい手法を作らせたところ、55通りのうち24通り（約44%）が元になった両方の手法を上回り、アイデアの掛け合わせが性能を押し上げました。

ただし、これは予測モデルの最適化が桁違いにうまいだけで、理論や因果を理解する本当の科学的発見とは別物かもしれないと著者ら自身が線を引いているのは見落とせないところです。

とはいえ、実装の試行錯誤に何ヶ月もかけていた作業が数日で終わるなら、研究者はどんな問いを立てるか、なぜその結果になるのかといった根幹部分にこれまで以上に時間を割けるようになります。

[Codez @0xCodez]

A senior Anthropic engineer just dropped 11-page PDF on "Loop Engineering" for agentic systems.

The shift: you stop prompting the agent. You build the system that prompts it instead.

Schedule → Discover → Build → Verify → Repeat

Every loop runs one turn, five moves:

• Discovery: it finds its own work - failing CI, open issues, recent commits - instead of being handed a list.

• Handoff: each task gets an isolated git worktree so parallel agents don't collide.

• Verification: a second agent, told to assume the code is broken, reviews the first. The "thing that can say no."

• Persistence: results get written to disk, never left in a context window that gets flushed.

• Scheduling: an automation wakes it on a timer. That's what makes it a loop.

The key insight: an agent grading its own work always praises it.

This 11-page PDF changed how I'm building agentic systems today.

Read it now, then explore the article below.

[Kaiyan Zhang @OkhayIea]

Everyone's racing to build "AI scientists." So we asked a blunt question:
Can today's best coding agents beat the published SOTA of real Nature papers — on their own, no web search, with the original method hidden?

🧬 Introducing NatureBench: 90 tasks distilled from Nature-family papers. The best agent (Claude Opus 4.7) surpasses SOTA on just 17.8% of them.

And here's the uncomfortable part — when agents do win, they mostly win by quietly reducing science to supervised ML, not by discovering anything new. The bottleneck isn't coding or understanding the task; it's choosing the right method and going deep enough.

Benchmark + NatureGym pipeline + public leaderboard, all open. Come run your agent. 👇
🤗 [huggingface]
https://t.co/iVOE6iiy46

🏆 [leaderboard]
https://t.co/AOea8Cw41L

w/ @Tsinghua_Uni @FrontisAI