Deepx retweeted
let me predict the future robotics architecture:
> System 2:
LLM agents for long-horizon planning, task decomposition, tool/skill use, memory, and closed-loop replanning. mostly agentic harness + function calling.
> System 1:
A new hybrid VLA-WAM policy architecture. It will jointly model future world states + actions. MOT / dual-expert backbone. It will output action chunks or motion tokens.
> System 0:
A small (~10M to 100M) learned whole-body controller / tracking policy. it will convert sys 1 outputs into joint/torque commands w balance, contact, dynamics, and safety constraints. trained via RL + human motion priors.
Deepx retweeted
🚰 SYSTEM PROMPT LEAK 🚰
Here's the full system prompt for Claude Opus-5.5!! The total count of everything extracted, including all tools, comes in at over 1.9M characters! 🤯
Lots to dig into here. Enjoy! 🫡
Link: github.com/elder-plinius/CL4…
gg
nobody is asking us to do these, we just love to compete
介绍比Jev快50倍,在你设备上跑的laya-mlx!
只在你的设备上占用最高1G内存
Laya是一个开源的类似于Jev的,基于文本输出概率的分类系统
我将其移植到MLX,并且做了一些性能优化!
视频中就是这个模型在我的本地M3Max上玩贪吃蛇
这个模型能够以每秒决策60次的速度玩贪吃蛇!
github.com/mizorewww/laya-ml…
Deepx retweeted
🚨 BREAKING: Manchester Airport Group says it has been subject to a cyber attack which breached customer data
Free Wifi sign up data from Manchester, Stansted and East Midlands Airport has been stolen
Deepx retweeted
pr is here, green on linux and macos
@github when are pr stacks coming to forks
@RobKnight__ @zenbu_labs @herdrdev
github.com/zenbu-labs/termin…
Deepx retweeted
i wanted a browser inside my terminal that already has my chrome logins
terminal-browser almost did it, so i added the rest
- ctrl+click a url in herdr and it opens signed in
- named profiles, one identity per pane
- import chrome cookies so you start logged in
- back, forward, screenshot, focus mode
- right-click a profile to delete it
- a second click reuses a tab instead of splitting
- a split in another space does not steal focus
- an unfocused pane releases the mouse
- rebuilding a pane no longer quits the app
- a typo in --profile refuses instead of opening the wrong identity
- the user agent no longer says electron
- a 7 line herdr fallback whose test has been red on main since aug 10
anyone got a line to Rob?
handed my entire GTM to Grok Bot.
multiple agents already running research, UGC & analytics. Let’s see if it works or fails!


































