Introducing SKY-001, the first Engineering World Model. On CAD and BIM tasks, it beats GPT-6 Astra and every frontier model we tested: -> 51-125x faster - > 150-210x cheaper -> Higher IOU than all frontier models SKY-001 lets you dream a physical object before you build it. 18 months of work. Built by the team out of Maluuba's deep learning lab (acq. by @MSRResearch). The first ‘AI General Engineer’ will be the backbone for Autonomous Manufacturing.

Robotics has a distribution problem. Today, Ultra is announcing $62M in funding to solve it, including a $50M Series A led by @hiFramework. We believe that deployment is the bottleneck in robotics. It’s relatively easy to make a robot look impressive in a controlled environment. It’s much harder to put a robot into the open world, make it useful, keep it running, earn the trust of the people working alongside it, and do that over and over and over again. So that’s what we’re focusing on.

There are basically three approaches to venture. And one is dominating the attention, headlines, and capital. 1. Fighting for diamonds in the middle of the fairway 2. Searching for diamonds in the rough 3. Mining for silver

There are basically three approaches to venture. And one is dominating the attention, headlines, and capital. 1. Fighting for diamonds in the middle of the fairway 2. Searching for diamonds in the rough 3. Mining for silver

As part of testing @Muse, I quizzed it to explain to me a few things I think about when using these agents and deciding how much connections and context to give - the struggle of the Trust-Utility Curve (the more you give, the more useful it can be). xitter.cf/melodykoh/status/20262… TLDR - there continues to be limitations around how LLMs fundamentally work and the probabilistic nature of things, and there really isn't a way to get around on context leakage or prompt injection-like attacks. However, I found Muse/Meta's willingness to be transparent in its mechanics reassuring. A few highlights on my Q&A with Muse - see more details on the screenshots of my questions and its responses. 🧵 Q1 - you say that you will ask for my permission for actions, but how do I know this can be enforced deterministically (vs instructions in files, which are just vibes for these models)? Most reassuring part of the answer: "When I invoke a send tool, the runtime intercepts the call and blocks execution until you've approved it. I can't trigger the approval myself, can't see it, can't override it — the approval card is invisible to me and your decision is final. So even a compromised or injected version of me, actively trying to send without asking, hits the same wall: there is no code path where the send executes without the runtime obtaining your decision first." Translate to non-developer speak: there's a mechanical gate that forces the model to serve up approval to the user when the action is a list of predefined "must get user approval". No vibes-bsed (good!)