faych
233 posts
OpenAI 算是把 macOS 玩明白了，local-first Agent!!!!!!!!!!!
OpenAI@OpenAI
Introducing GPT-5.5 A new class of intelligence for real work and powering agents, built to understand complex goals, use tools, check its work, and carry more tasks through to completion. It marks a new way of getting computer work done. Now available in ChatGPT and Codex.
日本語
@googledevs yes, those three commits still exist, we can use git reflog to get the first commit hash and git reset to it with --hard.
English
Test your git knowledge with this developer challenge: after a git reset, three commits are removed from the log, but now you realize you need them back. What are your options? Share your answer in the comments.
English
KohakuTerrarium 1.0.0 is now published!
github.com/Kohaku-Lab/Koh…
KohakuTerrarium is a framework for you to make ANY agent. You can use it to reproduce openclaw, hermes agent, or design completely new agent architectures.
It also ships production-ready agent out of the box.
English
Here’s a quick git dev challenge: you've got three edits, two adds, and one commit. Looks straightforward, but which version actually ends up in the repo? Watch the video and share your answer in the comments.
English
Just launched: Awesome Gemini Prompts ✦
A curated collection of 1,800+ high-quality prompts, specifically optimized for Gemini series.
I built this because finding awesome prompts were too hard.
Open Source & Free.
awesomegeminiprompts.tech
github.com/neverbiasu/awe…
English
Loki Mode 多代理自主创业系统 for Claude Code
这玩意儿能把你的 PRD 直接变成赚钱的产品，全程不需要你插一根手指。
37 个 AI 代理 24 小时给你干活，分布在 6 个"蜂群"里：PRD → 竞争研究 → 架构 → 基础设施 → 开发 → QA → 部署 → 商业运营 → 增长，一气呵成。
github.com/asklokesh/clau…
中文
@localhost_8188 I directly use opencv and I don't think any of them have be done by gpu bcuz I used pre built wheel XD
I know some ppl tried to build opencv for cuda but I never looked into it...
English
real-time streaming Video AE encoding + decoding with webcam
1024x768, streaming speed is around 13~15fps
current bottleneck: CPU (webcam capture+render final frame + queuing...) and Screen capture (maybe I need to use some capture card and connec to another machine)
English
@KBlueleaf oh gotchu, I misunderstood before. So the bottleneck is about webcam and final frame rendering. I wanna know if this part can also be done in GPU.
English
@localhost_8188 note the full encoding/decoding is run on GPU
so I dunno what's "accelerate the decoding process" means
English
HDM v1.1 is coming
Finetuned from HDM-xut-340M (v1) and it is now 370M param (some blocks added)
With TREAD enabled again, HDM v1.1 can inference with internal token routes which provide 2~4× speed up, make HDM v1.1 have more than 2× speed vs. both sd1 & sdxl on 1024px
English