W.Wyatt

ENG/CHN | LLM RAG-Agent | Knowledge-base Epistemic system | Math enthusiast
Github:

[Mount & Blade @Mount_and_Blade]

Today, we are deeply saddened to share the passing of our beloved co-founder, İpek Yavuz, whose passion, dedication, and vision have been at the heart of TaleWorlds since the very beginning.

Together with her family and loved ones, we mourn her loss and remember her with love, respect, and gratitude.

[W.Wyatt @grexrr_w]

@RELXHQ what happend to relx ireland? it it permanent or temporarily?

[Benjamin De Kraker @BenjaminDEKR]

This is pretty awesome. Gaussian Splat, not CAD:

[AI Will @FinanceYF5]

Google Research 发布了“教科书终结者”。

它叫 Learn Your Way，使用 LearnLM 将任意 PDF 转换成 5 种个性化学习形式。使用该系统的学生，在记忆测试中的成绩为 78%，而传统方式为 67%。

教育革命已经到来。

[Dash @creativedash]

Node Editor Canvas 🔊
Play: https://t.co/EokmtA27OU

[Gorden Sun @Gorden_Sun]

Everything Claude Code：Anthropic黑客松冠军的生产级配置库

复用这套配置，可以快速搭建起一套高水平的 AI 辅助编程环境。
提供了从TDD（测试驱动开发）到代码审查的完整自动化流程，通过配置文件强制AI遵守团队代码规范，大幅降低了Review的人工成本。

关键模块（这部分由AI辅助我梳理）：
· Agents (专项智能体)：如 planner 负责需求拆解，architect 负责系统设计，security-reviewer 专注漏洞分析，避免单一大模型处理复杂任务时的能力退化。

· Skills (技能/知识库)：外挂的领域知识，包含 backend-patterns（后端模式）和 frontend-patterns（前端模式），让 AI 这里的代码符合特定技术栈的最佳实践。

· Commands (快捷指令)：将复杂 Prompt 封装为 Slash 命令，例如输入 /tdd 自动执行测试驱动开发流程，输入 /refactor-clean 自动清理死代码。

· Rules (核心准则)：AI 必须遵守的“宪法”，例如 security. md 禁止硬编码密钥，git-workflow. md 规范提交信息格式，确保输出质量底线。

· MCP Configs (工具集成)：预设了 GitHub、Supabase、Vercel 等常用开发工具的 MCP 连接配置，打通 IDE 与外部服务的连接。

Hooks (自动化钩子)：基于事件触发的自动化脚本，例如在工具使用前后自动检查 console.log 残留或运行特定的校验逻辑。

Github：https://t.co/8I2GjCCk9b

[Michael Truell @mntruell]

We built a browser with GPT-5.2 in Cursor. It ran uninterrupted for one week.

It's 3M+ lines of code across thousands of files. The rendering engine is from-scratch in Rust with HTML parsing, CSS cascade, layout, text shaping, paint, and a custom JS VM.

It *kind of* works! It still has issues and is of course very far from Webkit/Chromium parity, but we were astonished that simple websites render quickly and largely correctly.

[WquGuru @wquguru]

Vercel将他们过去10年的React功底浓缩成了一个Skill，CEO戏称“用这个Skill相当于雇佣了一个mini版的10年经验React开发者”

非常值得参考一下，理论上所有团队的内部最佳实践都值得做成skill

[向阳乔木 @vista8]

竟然还有这种网站？！

全球冲突风险监测器（由Reggie James创建）

免费提供实时全球地图，追踪冲突、军事基地、核设施、战略水道，以及五角大楼披萨指数。

网址见评论

[W.Wyatt @grexrr_w]

I truly hope the Iranian people will endure the last of darkness and finally see the light. The world needs a good news, we all wish you the best! #IranRevolution2026 #Iran

[Tibo @thsottiaux]

With the Claude Code shutdown. I am proud that we build Codex in the open with our OSS repo and we are 100% invested in supporting a flourishing ecosystem of agentic coding tools out there.

You can already build on top of https://t.co/hDjdmjH8pg directly, which includes ChatGPT login and same usage as you get in codex. Reach out if you are a builder!

[Kylie Robison @kyliebytes]

Scoop: xAI staff had been using Anthropic’s models internally through Cursor—until Anthropic cut off the startup’s access this week.​​​​​​​​​​​​​​​​

Here’s what xAI cofounder Tony Wu sent to staff on Wednesday according to a copy of the internal Slack message I viewed:

“Hi team, I believe many of you have already discovered that anthropic models are not responding on cursor. According to cursor this is a new policy anthropic is enforcing for all its major competitors.

This is a both bad and good news. We will get a hit on productivity, but it rly pushes us to develop our own coding product / models. We're at a time in which Al is now a critical technology for our own productivity. This coming year is rly going to be wildly exciting for all of us.

The team is rapidly developing our own models / product. We will have something to share with everyone soon. In the meantime, you may still try all different kinds of models in grok build.”

Anthropic declined to comment. A Cursor spokesperson directed me to Anthropic for comment. xAI did not respond to request for comment.

[Lexi 勒西 @lexi_labs]

https://t.co/YF2RdfvZ89

[Dr. Bo @BoDiplo]

人类取得了逆天的半导体科技成果，投资了万亿美元研发AI技术，建立耗能巨大的数据中心，就特么的是为了给大家看这个？😹

[alex zhang @a1zhang]

Much like the switch in 2025 from language models to reasoning models, we think 2026 will be all about the switch to Recursive Language Models (RLMs).

It turns out that models can be far more powerful if you allow them to treat *their own prompts* as an object in an external environment, which they understand and manipulate by writing code that invokes LLMs!

Our full paper on RLMs is now available—with much more expansive experiments compared to our initial blogpost from October 2025!

https://t.co/x47pIfIkTb

[凡人小北 @frxiaobei]

这个 repo 有点牛逼，
把一整套科研 + 医疗的复杂流程，
直接拆成了可复用的 AI 能力块。

138 个 scientific skills，
包含文献、组学、药物发现到临床研究、报告生成，
我能想象到的医生和药企的一整套流程都在里面了。

随便拎几个出来，
封装成 AI scientist 产品，
剩下的就是去卖了。

这种 repo，看的人不多，
但看懂的人，基本都知道它值钱在哪。
https://t.co/RRKGAcszM3

[Crypto_Painter @CryptoPainter]

这种龙卷风是不是只能去美国看？

从小到大只见过台风，没见过真正的龙卷风...

[Boris Cherny @bcherny]

I'm Boris and I created Claude Code. Lots of people have asked how I use Claude Code, so I wanted to show off my setup a bit.

My setup might be surprisingly vanilla! Claude Code works great out of the box, so I personally don't customize it much. There is no one correct way to use Claude Code: we intentionally build it in a way that you can use it, customize it, and hack it however you like. Each person on the Claude Code team uses it very differently.

So, here goes.

[KK.aWSB @KKaWSB]

DeepSeek 新论文：用 60 年前的数学公式，拯救失控的大模型。

他们刚刚解决了一个困扰 AI 领域的根本性难题——而且他们用的武器，竟然是一个 1967 年的古老算法。

深度学习曾因网络过深导致信号消失，ResNet 通过建立一条“直通单车道”（恒等映射）解决了这一难题。最近兴起的“超连接”技术试图将其升级为“多车道并行”以提升性能，但这种复杂的混合让信号在层层传递中失控膨胀（甚至放大 3000 倍），导致训练崩溃。DeepSeek 的神来之笔是引入 1967 年的 Sinkhorn-Knopp 算法，从数学上强制限制这些混合通道必须“行与列之和均为 1”，从而在保留“多车道”性能优势的同时，将信号波动死死按在 1.6 倍的安全范围内，用优雅的数学约束彻底解决了扩展性难题。

DeepSeek 得到的结果令人震惊：
稳定性： 信号波动从 3000 倍 被死死按在 1.6 倍 以内。
可靠性： 这种稳定是由数学定理保证的，不再需要祈祷运气。
低成本： 这一切只增加了 6.7% 的训练开销。

[Aakash Gupta @aakashgupta]

This is the AlphaGo moment for software agents.

Every AI coding agent today, from Cursor to Devin to Claude Code, learns from human traces. GitHub issues. Pull requests. Test suites written by developers. Someone had to write all of it first.

This creates a hard ceiling. SWE-bench Verified required 93 professional developers to manually screen 1,699 samples just to produce 500 usable problems. The original SWE-bench needed human-curated issue descriptions, pass-to-pass tests, fail-to-pass tests. The data scales linearly with human effort, and human effort doesn’t scale.

SSR breaks that constraint entirely. One model plays two roles: bug injector and bug solver. It explores real codebases, discovers how tests work on its own, creates progressively harder bugs by removing code or manipulating repository history, then learns to fix what it broke. The injector must produce bugs that fail tests through semantic errors, not syntax mistakes. The solver gets no natural language description, just the raw codebase and a test patch.

No human labeling. No curated issues. No pre-existing test suites. Just Docker images with source code.

The results: +10.4 points over human-data baselines on SWE-bench Verified. +7.8 on SWE-Bench Pro, which contains 731 enterprise-level tasks averaging 107 lines of code across 4+ files. The model trained entirely on self-generated bugs still generalizes to solving natural language issues written by humans that it never saw during training.

They also introduced higher-order bugs constructed from the solver’s own failed repair attempts. When the solver fails, those failure patterns become new training problems. The curriculum gets harder automatically.

This is the same loop that took Go from “decades away” to superhuman in 18 months. AlphaGo didn’t need human games to learn. It generated its own curriculum by playing against itself, discovered strategies no human had conceived, and scaled past every grandmaster on the planet.

We’re watching the same pattern emerge in code. The bottleneck was always data, and the assumption was that useful coding data required human developers to create it. SSR suggests that assumption is wrong. Raw repositories plus self-play can generate unlimited training signal.

Meta trained this on 16 million token batches across 150 steps using H100 clusters. The infrastructure exists to scale this dramatically further. More repositories, more self-play iterations, harder bugs.

GitHub has 200 million repositories. The training data ceiling just disappeared.