Juls retweeted
LLM 7x and RLM 2x, wonder if World Model could pull smtg here. 🤔
World Model
$WM
6KXjdTtGqXMawUqcAypAYXCdav7rXQYZjDyejfjQpump
Og world model on solana:
6KXjdTtGqXMawUqcAypAYXCdav7rXQYZjDyejfjQpump
$WM
1
37
Why ape le world model when the there is an og. CEkbQMxyzXgZPZPvxcWqHtPEqEMgE51nf6L5QUympump ❌️
The og:
6KXjdTtGqXMawUqcAypAYXCdav7rXQYZjDyejfjQpump ✅️
1
119
Le World model is getting traction on twitter.
Ape the og $WM
6KXjdTtGqXMawUqcAypAYXCdav7rXQYZjDyejfjQpump
1
161
Juls retweeted
World Models allow robots to follow instructions and respect physics
$WM
6KXjdTtGqXMawUqcAypAYXCdav7rXQYZjDyejfjQpump
This paper shows how to pair language models and world models so robots follow instructions and respect physics.
Embodied AI means a machine sees the world, thinks about it, then does something in it.
i.e. the blend of language smarts with physics smarts in one system.
This fixes the common gap where a robot makes a good plan that fails in the real world.
Language models turn vague goals into step by step plans using knowledge from text.
Multimodal language models also look at images or sensor data and connect that to the plan.
World models are internal simulators that predict what will happen after each action.
Language models understand meaning but often miss friction, timing, and safety limits, while world models respect those rules but miss high level intent.
The paper suggests a loop where the language model plans, the world model checks and forecasts, and memory keeps useful lessons.
This loop connects seeing, reasoning, and acting, so the agent can adjust mid task and fix mistakes.
The survey also covers how to run these models on real devices with quantization, compilers, and custom chips.
Early wins are in service robots, rescue drones, and factory arms where natural instructions become physics aware behavior.
----
Paper – arxiv. org/abs/2509.20021
Paper Title: "Embodied AI: From LLMs to World Models"
Juls retweeted
World Model is trending again on X
x.com/i/trending/20368829772…






