Alireza Ghods

Alireza Ghods

20,222 Twitter followers

Follow

Building @natixnetwork 🛠️ #Web3 Physical infrastructure (#DePIN) creator & investor.

Posts

This edge case tells a big story. From the front view alone, this video almost looks unremarkable. The multi-camera view tells a completely different story. Single-camera datasets aren't just incomplete. They're misleading. You train on them and your model learns the wrong representation of what's rare, what's dangerous, and what caused what. The blind spot gets baked in.

NATIX Edge Case Spotlight 🛑 Physical AI doesn't just need to move. It needs to know when to wait. A vehicle blocking the lane. No special signal, just context. Extracted from our network via an advanced VLM. This is what the long tail is all about.

Real-world AI is not limited by data alone but by the lack of true perspective 🌍 NATIX captures multi-camera, multi-angle data, creating a richer view of the physical world 🧠 That kind of capability is what made this collaboration a natural fit for MaprGo 🚀

Everyone is focusing on the OEMs. The bigger signal is NVIDIA doubling down on datasets + simulation + reasoning models. Autonomy is converging toward a learning loop built on real-world data. That’s where the real bottleneck is. twitter.com/AlirezaGhods2/stat...

World model accuracy degrades the moment it leaves the sandbox environment. That's not a model problem. That's a data problem. Fix the data strategy first. The model will follow.

You can't simulate traffic in Lagos by training on footage from San Francisco. Scale without diversity isn't scale. Real-world AI requires real-world variety — and most companies have neither.

Raise your hand if you thought the first generation of cyborgs will be cockroaches 🫣

The German startup SWARM Biotactics makes biorobots based on living insects and has already attracted 13 million euros of investment,and deployed with paying NATO customers. source:https://linkedin.com/posts/s-wilhelm_one-year-ago-this-didnt-exist-today-activity-7432506837616463873-4S64?utm_source=share&utm_medium=member_desktop&rcm=ACoAAB3Ot4EBxrxlaANpW1nZnXZxm_80FRABZPc…

The shift from language to vision in Physical AI means progress. LLMs learned how to understand us, reply, and predict the next sentence. World Models are how we teach AI how to see, interact, and predict the next moment in reality. For that to happen, visual data is the key. twitter.com/AlirezaGhods2/stat...

Thanks @Hongyu_Lii @jiahuifu_carol for the sharing! Seems increasingly the case that if ur world model can dream it, u can infer actions to do it in reality.

The holy grail of robotics is to be able to perform previously-unseen, out-of-distribution manipulation tasks “zero shot” in a new environment. NovaFlow proposes an approach which (1) generates a video, (2) computes predicted flow — how points move through the scene — and (3)

Edge case found: Sensor overload, autonomous vehicle does not respond. Vision-only autonomous driving is better. In the long tail and in everyday driving.

Nic Cruz Patane

Elon Musk on LiDAR: “LiDAR also does not work well in snow, rain or dust due to reflection scatter. That’s why Waymos stop working in any heavy precipitation.” Yesterday in LA: x.com/elonmusk/statu…

Loading..