半江红

全栈技术男

[半江红 @chptxy]

@hd_nvim mojo语言大家都看不上吗？https://t.co/4FjFnsmGbs

[半江红 @chptxy]

mojo语言，像是rust和Python的杂交语言。

[半江红 @chptxy]

@joshgreaves_ml it is a new era after this paper. https://t.co/2f3ECceyqZ

[半江红 @chptxy]

有高通加持的Arduino可以安装DHH的omarchy linux玩AI了。 https://t.co/n9ryY9Ue8T

[半江红 @chptxy]

mojo语言，像是rust和Python的杂交语言。

[Modular @Modular]

Today, we open sourced Mojo 🔥.

Announced just now during the ModCon keynote, effective immediately, Apache 2.0 License.

Thank you to our community for waiting patiently and building alongside us.

#ModCon2026

Full blog: https://t.co/y5cSUsohrs

[洛神父🐱 @baarlock]

deepseek v4 pro出来了，感觉商单拿错台词本了，明显比flash更拉。

[半江红 @chptxy]

在一个无人在意的角落，grok经历了核心团队剧烈变动后，慢慢重回正轨了。

[X Freeze @XFreeze]

Grok 4.5 now also ranked #1 on the Long-Horizon Terminal-Bench by binary pass rate, outperforming Claude Fable 5, Claude Opus 4.8 and GPT-5.6-sol

Under the strictest scoring metric - where a task counts only if it is fully solved with a perfect reward and zero errors......Grok 4.5 finished clearly ahead of every other tested model

This matters for real-world coding, automation and difficult engineering work because long-horizon terminal tasks require much more than making partial progress

The model has to maintain context, recover from mistakes and successfully complete an entire workflow across hundreds of steps

Grok 4.5 is showing serious strength on complex agentic tasks over extended periods of time

[洛神父🐱 @baarlock]

@LinearUncle “过早优化是万恶之源”, deepseek埋头搞性能工程优化，模型基础能力提升就很慢了。当前模型智能离AGI还很远，需要有大量的突破，专注工程优化属于本末倒置。当然从消费者角度，更喜欢便宜又够用的产品。

[半江红 @chptxy]

有没有x友能告诉我，loop engineering 出现后，是不是意味着 prompt engineering、context engineering、harness engineering 等这些名词已经过时了？我还没有开始学是不是就不用学了？

[AYi @AYi_AInotes]

Prompt Engineering已死，Loop Engineering接棒，

@karpathy 的做法我感觉最有画面感，晚上设好研究循环，自动改脚本，跑测试，留对的，扔错的，自己去睡觉，早上起来直接看结果，全程几乎不插手。

这就是现在圈里聊的Loop Engineering，从指挥AI一步步走，变成设计一个能自己跑完的系统。

四个支柱，每一个都是生产环境踩出来的血泪坑。

1️⃣知道什么时候停，别信它自己说搞定了，代码跑不通，逻辑有漏洞的情况比比皆是，要加硬性退出条件和独立验证，无数项目从demo到落地就死在这一步。

2️⃣保持上下文干净，循环跑久了旧错误，死胡同，冗余思考堆在一起，决策质量直线下滑，要把上下文当有限预算管，定期压缩，用子Agent隔离子任务。

3️⃣给真正能用的工具，不是堆一百个凑数量，要少而精不重叠，可安全重试，错误信息能被看懂，这是生产环境最容易翻车的地方。

4️⃣有能说不的机制，Agent不会挑自己的毛病，自己改作业永远满分，必须把干活的和检查的分开，也就是Maker-Checker模式，Claude Code团队早就在用了。

@bcherny 说得最直白，我已经不prompt Claude了，我写loops，然后让它们跑。

模型正在慢慢变成标准件，
同一个模型，配上更好的循环设计，效果能从中等拉到顶级。
真正的工程价值，早就从单次对话转移到了长期自治系统的设计上。

做Agent的朋友可以对照着查一遍，什么时候你从写提示词的人，变成了设计系统的人，什么时候才算摸到了下一阶段的门槛。

[半江红 @chptxy]

间线上如果出现那种明显的从小红书或者短视频搬运来的shit，顺手就把这个用户mute了，看到一些原创的内容要记得点赞，现在算法已经不给我推那些shit了，好干净的界面。

[半江红 @chptxy]

同一段话，日语和中文长度差距有这么大？

[macha @macha3142]

「摆烂」の意味を日本人でもわかるようにやさしく教えてください。

[半江红 @chptxy]

@xicilion 短视频上都是些外行，用AI开发了Todo list水平的前端界面，就开始咋呼程序员要被AI代替了。

[半江红 @chptxy]

好奇罗永浩老师的访谈节目，会不会哪一天突然邀请王自如出场？

[半江红 @chptxy]

一台mini主机、一块达子显卡拼到一起，丐版炼丹环境就有了，有空还要把踩过的坑总结一下。

[半江红 @chptxy]

比特币不再安全了！Google新的量子算法突飞猛进，破解钱包密钥如探囊取物，比特币社区需要尽快升级到新的量子安全算法。

[nic carter @nic_carter]

Many are wondering "what Google saw" that caused them to revise their post-quantum cryptography transition deadline to 2029 last week. It was this:

https://t.co/dQtmTK9pdz

[半江红 @chptxy]

@bronzeagepapi no source code?

[半江红 @chptxy]

个人电脑、手机又该降价了

[Grummz @Grummz]

Micron shares are in free fall this week.

- Be Micron.

- Get excited about Sam Altman singing letters “intending” to buy 40% of DRAM.

- Close your entire Crucial consumer memory division so you can sell gamer RAM to OpenAI.

- Sam Altman loses investors, never buys the RAM he signed letters for.

- Google announces breakthrough saving AI 6x the RAM.

- Micron now stuck with all this RAM and no way more division to sell to consumers.

[半江红 @chptxy]

LLM按token计费的情况下，用英文比中文性价比高不少，英文单词比汉字拆分的token更少，而且如果是技术话题不是抒情八股扯淡，英文比中文更简洁更精确，通常也就更省钱了。

[Mark Kretschmann @mark_k]

Gemini 3.1 Pro benchmarks leak @GoogleDeepMind

Take with a grain of salt no guarantees for authenticity 🤷‍♂️

[半江红 @chptxy]

Google Gemini这一步跨得也太大了！

[Chubby♨️ @kimmonismus]

How is this not big news?!

Google DeepMind has unveiled Gemini Deep Think’s leap from Olympiad-level math to real-world scientific breakthroughs with their internal model "Aletheia", scoring up to 90% on IMO-ProofBench Advanced and tackling PhD-level problems with scalable reasoning power!

"Aletheia" autonomously solved open math problems (including four from the Erdős database), contributed to publishable papers, and helped crack challenges in algorithms, economics, ML optimization, and even cosmic string physics.

2.5 years ago chatbots werent even able to solve simple math problems.

"We are witnessing a fundamental shift in the scientific workflow. As Gemini evolves, it acts as "force multiplier" for human intellect, handling knowledge retrieval and rigorous verification so scientists can focus on conceptual depth and creative direction. Whether refining proofs, hunting for counterexamples, or linking disconnected fields, AI is becoming a valuable collaborator in the next chapter of scientific progress."

[半江红 @chptxy]

@rohanpaul_ai release time?