Sabitlenmiş Tweet
CayaLee
122 posts

CayaLee
@cutegoary
Coding on the sand 🌴🥥👨💻✨ Just staring at the waves until the logic clicks... 🌊🐚 🦥
Los Angeles, CA Katılım Haziran 2010
1.1K Takip Edilen12 Takipçiler

@strrlthedev 十年前年做过更大的。当时由于没有可以跨mac、windows、linux和嵌入式的多端可读写可加密磁盘镜像格式,然后就用sqlite当磁盘镜像,甚至可以自动和手动控制哪些文件在内存中。这个需求如今告诉AI,不到10分钟就实现了。
中文

朋友们,我们今天正式在 ProductHunt 发布了 Yansu!
producthunt.com/products/yansu
Yansu 是一款主动型 Agent 应用。
它会在后台默默记录你的截图、操作和声音,理解你真实的工作方式,并把零散上下文沉淀成结构化知识。
然后,它会基于这些知识,主动帮你生成贴合自己工作流的定制工具:项目追踪器、AI spend 监控、会议 action item 提取器、专属 dashboard……
它还有自己的虚拟光标,可以在后台开 App、填表单、提工单,不抢鼠标,也不打断你。
本地优先,数据不进我们的服务器;已通过 SOC 2 Type II 和 ISO 27001。支持 macOS / Windows / Linux,也有团队版和企业版。
中文

From text to a manufacturable bed frame.
This is Orca: an AI design tool that lets anyone create precise, assembly-ready 3D designs using natural language.
No 3D modeling experience required.
Just describe what you want in natural language, and Orca turns it into precise, manufacturable, assembly-ready 3D designs.
In this demo, we generate a complete bed frame design from a simple text prompt.
We’re now opening early beta access.
Do you wanna try it? See my bio to join our waitlist
#AI #Startup #BuildInPublic #ProductDesign #DesignTools #CAD #3DDesign #GenerativeAI #FutureOfDesign #CAD #TextToCAD #IndustrialDesign #3DModeling #Manufacturing #DesignTools
English

@linkayser @leap_71 Great work! Although I didn't use C++, I implemented a surface blending algorithm similar to Rhino using Three.js in the browser.
English

@cutegoary @leap_71 Very nice. Check out PicoGK.org where we open sourced a large part of our tech stack.
English

近期新出现的小而美的 markdown 软件,已经这么多了。
- quarkdown.com
- clearly.md
- writer.computer
- sobanotes.app
- cogito.md
- tolaria.md
中文

@xeonai44 @pupposandro super cool ! it brought my old V100 server back to life again.
English

@pupposandro I have the Tesla V100-SXM2-32GB PCIe hacked version and it's pretty good! Did some tests here: github.com/xeonai44/xllam…
English

@bstnxbt @ivanfioravanti Great work ! how it performs on huge models like the GLM 5.1 in M3 Ultra 512G
English

DFlash. Speculative decoding for Apple Silicon. Stock MLX, no fork.
github.com/bstnxbt/dflash…
@ 2048 tokens, M5 Max, stock mlx_lm baseline:
► Qwen3.5-4B: 53.74 → 219.83 tok/s (x4.10)
► Qwen3.5-9B: 30.96 → 127.07 tok/s (x4.13)
► Qwen3.5-27B-4bit: 32.35 → 62.78 tok/s (x1.90)
► Qwen3.5-35B-A3B-4bit: 142.12 → 240.21 tok/s (x1.69)
Block-diffusion draft generates 16 tokens in one pass:
► Target verifies in one pass.
► Every emitted token is verified.
► Lossless.
English

















