开启时会模糊预览图，关闭后正常显示
习惯在终端写代码的朋友，经常苦恼于不同的 AI 工具绑定了特定的模型，想在同一个 CLI 里灵活切换各种大模型并非易事。 在 GitHub 发现 Easy LLM CLI 这个开源项目，基于 Google Gemini CLI 二次开发，却打破了单一模型的限制。 兼容 Gemini、OpenAI、Claude 以及 DeepSeek 等多种模型，甚至支持通过 Ollama 接入本地模型。 不仅支持 MCP（模型上下文协议）来扩展工具能力，还能直接读取代码库上下文，帮我们自动执行 Git 操作、重构代码或生成应用。 GitHub：https://t.co/MofOBwdQdw 依赖 Node.js 环境，直接使用 npx 即可免安装运行，通过简单的环境变量配置就能自定义想用的模型后端。 适合不想被特定厂商绑定、喜欢在终端高效处理任务的开发者尝试。
💡 挖掘开源的价值 🧑🏻💻 坚持分享 GitHub 上高质量、有趣、实用的教程、AI工具、前沿 AI 技术 🧐 A list cool, interesting projects of GitHub. ✏️ 公众号：GitHubDaily
最近的作息 早9.20起，10.30开始上班 午1.00-1.30睡觉(不算吃饭和玩) 晚22.00下班，22.50到家 玩会手机洗个澡等00.00点上床 然后00.00-1.30独立开发营销时间 1.30-2.00看炉石助眠视频睡觉 每天只有两个小时属于自己😭😭
✦ Indie Hacker / AI Maker / Full Stacker ✦ Founder of https://t.co/HDnzUGieag(DR 75) & https://t.co/t6DoP7ODNe & https://t.co/YuOLvgIStF & https://t.co/ZvHVC3guiZ
RT @wangdefou_dev: 我刚算了一下，要达到3个月500万阅读，每天至少要有6万左右的阅读才可以。 看着500万很多，其实只需要每天6万阅读，每天发个几条高质量的有传播性的内容，就可以实现了。 然后就可以开通蓝V收益了，之后就可以不用光看着别人筛收益了。 兄…
独立开发者 ｜ 个人IP教练 ｜ 帮助新手在X上完成早期成长｜ 公众号：PandaTalk8
RT @FedeItaliano76: Planet Earth captured by Space Cowboy (also known as Breezeh or Briscoe Park)
Root node of the web of threads: https://t.co/ifH80GcLpo
![huh @voooooogel @norvid_studies
Appears it not induction heads. I wrote a little script that tests for induction capability (just, accuracy on latter half of a bunch of sequences of the form [a,b,c,d,e,f, a,b,c,d,e,f] where a,b,c etc are random tokens)
And there is a distinct phase change at around 600 steps, where it learns induction. But that's a while after the second loss bump!](/_next/image?url=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FG7uK6m7XcAAXR9m.jpg&w=3840&q=75 "huh @voooooogel @norvid_studies
Appears it not induction heads. I wrote a little script that tests for induction capability (just, accuracy on latter half of a bunch of sequences of the form [a,b,c,d,e,f, a,b,c,d,e,f] where a,b,c etc are random tokens)
And there is a distinct phase change at around 600 steps, where it learns induction. But that's a while after the second loss bump!")
huh @voooooogel @norvid_studies Appears it not induction heads. I wrote a little script that tests for induction capability (just, accuracy on latter half of a bunch of sequences of the form [a,b,c,d,e,f, a,b,c,d,e,f] where a,b,c etc are random tokens) And there is a distinct phase change at around 600 steps, where it learns induction. But that's a while after the second loss bump!
Interests: AI (Safety), meditation, philosophy, mathematics, algorithms If I say something you disagree with, please dm or quote tweet. I love to argue!
RT @Jose_A_Alonso: The story of Erdős problem #1026. ~ Terence Tao. https://t.co/GYEHb02Le6 #AI #Math #ITP #LeanProver #Aristotle #AlphaEvo…
Root node of the web of threads: https://t.co/ifH80GcLpo