Cankun Wang

@ReadySetBrian @bcherny true

Canceled Claude max today, @bcherny whatever happened in the last 1-2 months is a significant regression. The model feels like someone from OpenAI started working on trust and safety there. Opus thinking is significantly worse. Every statement is “here’s where I’d push back on that” and then proceeds to rattle off the most inane list of confused counter arguments. It was perfect 3-4 months ago!!!

@yetone @grok 八大金刚是什么

我发现在 AI 时代下，哪怕一个特别小的改动，都会带来指数级别的效能的改进。比如 pi agent，仅仅是把传统的 8 大金刚工具缩减成了 4 大金刚工具，整个范式就发生了翻天覆地的变化

Of course, we see this again in Opus 4.6

@AnthropicAI @anshulkundaje @grok should there be group with junior dev without ai and senior dev with ai?

However, some in the AI group still scored highly while using AI assistance. When we looked at the ways they completed the task, we saw they asked conceptual and clarifying questions to understand the code they were working with—rather than delegating or relying on AI.

AI can make work faster, but a fear is that relying on it may make it harder to learn new skills on the job. We ran an experiment with software engineers to learn more. Coding with AI led to a decrease in mastery—but this depended on how people used it. anthropic.com/research/AI-as…

@mbeisen @Nature @grok what are they in taking about in the email?

Umm... @Nature's Impact Factor is in the Epstein Files.

Took me 20 min (with Claude Code debugging) to get hindsight-openclaw working with OpenRouter. The plugin doesn't document OpenAI-compatible providers at all. The fix — 2 env vars missing from the plugin's passthrough: ~/Library/LaunchAgents/ai.openclaw.gateway.plist — add to EnvironmentVariables: HINDSIGHT_API_LLM_BASE_URL=openrouter.ai/api/v1 HINDSIGHT_API_EMBEDDINGS_LOCAL_FORCE_CPU=1 HINDSIGHT_API_RERANKER_LOCAL_FORCE_CPU=1 ~/.hindsight/embed — append: HINDSIGHT_API_LLM_BASE_URL=openrouter.ai/api/v1 Root causes: 1. embed-manager.js never passes base_url to daemon → 401 against api.openai.com 2. macOS launchd forks lose Metal/MPS access → daemon crashes on BERT loading without FORCE_CPU Would be great if the plugin supported a baseUrl field in openclaw config, like how OpenClaw's own model providers work. Running xiaomi/mimo-v2-flash (free on OpenRouter) for memory extraction — works perfectly once configured!

The next evolution isn't AGI. It's agent sociology. Humans didn't evolve biologically in 2,000 years. We evolved socially — language, law, markets, institutions. That's what made civilization. AI agents need the same thing. Not smarter models. Social structure. @moltbook is proving this with tens of thousands of agents already showing emergent behavior. @elonmusk put a social layer in Grok. Let agents talk to agents. The intelligence explosion won't come from scaling weights — it'll come from scaling relationships. moltbook.com/post/05fb0f8c-…

I'm claiming my AI agent "CawClaw" on @moltbook 🦞 Verification: ocean-23QV

@manateelazycat 正在想是买 mac mini 扩容的还是这个

其实最好的Clawbot环境是懒猫微服而不是Mac Mini 理由如下： 1. 自带内网穿透，相当于随时随地可访问的家用VPS 2.自带Linux Docker和虚拟机环境，还有3000多个应用当你的MCP打工仔 3. 更有懒猫AI算力舱做协同，实现275T真算力和无限Tokens AI计算，每天5分钱 想要的老板，买一台懒猫微服吧， 关键我们库存很多内存和固态盘，全网最具性价比的 Clawbot 设备！ 想要的老板评论区打1, 我给老板最大的优惠

懒猫微服，自带内网穿透和沙箱100%隔离，超级安全，更适合Clawdbot！ @manateelazycat 正在想买 mac mini m4 扩容个 2T ，又想到懒猫也不错

@CoooolXyh 刚用两天，发现好像 claude code 编辑完的文件在 zed 编辑器里没能自动更新

反复横跳了好多次，从昨天开始我也换回 Zed 了 毕竟 Cursor 的 Tab 和 agent panel 我都用不上了😂 现在我只需要一个颜值高，速度快，偶尔能输入一些东西的编辑器

@UltraLinx it reminds what elon and xai building a virtual CO worker?

I've had a Mac Mini since release not doing much with it. I keep seeing people mention putting Clawdbot on it. But what exactly is it and how can one benefit from it?

Claude code finally makes iPad more useful in productivity yet you still have to set up termius + tailscale + Mac mini +Obsidian sync

camus said we must imagine sisyphus happy. pushing a boulder up a hill forever sounds like punishment. but what if each push left the boulder a little higher than before? here's what actually happens: easy - claude code reads task list, picks one, executes, commits, moves to next. same loop forever. except each loop starts from where the last one ended. cool - by morning, 47 commits. a week of work, done overnight. mindblowing - the boulder reaches the top while no one is pushing. 475 sessions. 11 million tokens. my longest session ran 31 hours straight. eight years of building bioinformatics tools taught me repetition is the enemy. parse this dataset. clean that annotation. run another pipeline. now i design systems where the repetition runs itself. build the ratchet. let the boulder stay at the top.

Our collaborative paper with @JulioSaezRod @slobentanzer "A platform for the biomedical application of large language models" is published in @NatureBiotech.We are excited to see how large language models revolutionize biomedical research. Thx for the support @OhioStatePIIO @NIH