Never too late to be your authentic self and dream about world endless possibilities.
The Netherlands
Joined March 2008
- Tweets1.8K
- Following431
- Followers134
- Likes2.1K
Just saw Skyfall’s world model. I can see this opening a lot of doors inside enterprises.
LLMs got us this far, but I keep thinking @ylecun might be right about autoregression.
At some point, why should agents even need to operate in ways humans can understand?
Today we're unveiling SKY 00-1, the first Engineering World Model towards Autonomous Manufacturing.
LLMs are great at information retrieval and code, however, the physical world involves geometric and spatial understanding. Designing complex physical objects requires long-horizon planning and latent reasoning, which LLMs weren't built for.
We stacked SKY-001 against the top frontier LLMs (GPT-6 Astra, Fable 5.1 etc) and we found that our model is 100x faster and 150x cheaper in CAD and BIM while having higher precision.
As a compositional visual world model, SKY-001 lets you "dream" a design before you build it. It explores many designs in parallel, breaks each one down into its smallest components, and reassembles them in 3D, ready to open in CAD tools like Autodesk Fusion.
With applications in several industries, SKY-001 is the first step to reimagining engineering firms.
👉 Join the frontier with us and visit our website for early access (link in comments).
Abstraction ceiling in SWE has hit the ceiling. Now it’s time to materialize it.
With REA and other means, we could redesign entire toolchains and workflows for agents in months. Anthropic’s Scion is a great example.
Agent-native SWE is next.
New on our Engineering blog: Creating in-house source control.
In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale.
Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: anthropic.com/engineering/re…
Reza Bozorgi retweeted
I stand with President Trump in support of the Iranian people.
When the tyrannical Iranian regime falls, its terrorist leaders must be held accountable, both for their crimes against their own people and against Americans.
Reza Bozorgi retweeted
⚠️ Update: #Iran has now been offline for 120 hours.
Despite some phone calls now connecting, there is no secure way to communicate and the general public remain cut off from the outside world.
What footage makes it through shows extensive use of force against civilians 📵
The grief, the cruelty, and the dark future being imposed on Iranians have set my heart and mind on fire.
Most basic rights are banned, what a shity world we're in... We deserve a lot more than this dark situation.
#DigitalBlackoutIran
We underestimate AI because we’ve never confronted intelligence like this before. The Industrial Revolution and the Internet were amplifiers, this is a new category.
That’s why most people are missing out on this Short Golden Window (SGW missed!)
🚀 Combining #Embedding_Gemma + #Gemma_3n = real superpowers for devs.
With such a tiny footprint, you can bring cutting-edge AI anywhere, from agile startups to giant enterprises that limit full AI adoption.
youtube.com/watch?v=Xu1X-J-r…
🧠AGI vs 🚀ASI:
AGI labs (e.g Anthropic) build human-level intelligence step-by-step with safety-first guardrails.
But others (e.g DeepCogito) are jumping straight to ASI. Reasoning-first, self-improving superminds.
Two very different races.
The latter is racing under the radar.
The biggest, fastest, and deepest shift in human history is unfolding. Not just driven by economic forces, but by geopolitics and even bad actors. There’s no pause button.
With powerful open-weight models now in our hands, society is about to face massive shifts.
Given how good Kimi K2 is -delivering Sonnet 4–level performance at ~¼ the cost- this feels like a pivotal moment: we don’t need major new model breakthroughs to hit AGI anymore.
The frontier now is application-layer orchestration, tooling, agents, and context engineering.
The recent aggressive poaching by Meta (targeting Scale AI) and Google (targeting Windsurf) highlights that human-in-the-loop still holds significant value, possibly for the final chapter in human history.
#AI #HumanInTheLoop #TechTrends
#CLI, as seen in @claude_code or Codex CLI, has never been deeply connected with humans.
A black canvas opens a new doorway to communicate more directly, more closely, and more deeply with AI. ◼️🚀
Agents bridge human intuition and AI power, deepening the connection and making interactions delightful.
For instance, @RabbitCodeAI doesn’t just review your code; it writes you a poem summarizing the feedback, connecting with your heart, not just your repo. ✨🐇
The generosity in offering this level of intelligence is unprecedented. Under the banner of “beta” or “research preview,” they're giving it away for free. We've never seen intelligence this powerful, this cheap and I doubt this open-handedness will last beyond the next few weeks.
As frameworks like #LlamaFactory abstract away complexity, building autonomous agents becomes reliable and accessible even for individuals.
This democratization of AI power creates unprecedented opportunities, but also unprecedented risks.
github.com/hiyouga/LLaMA-Fac…
We're witnessing a massive shift, from pair programming with copilots (e.g. Cline) to #peer_programming with autonomous agents (e.g. Jules). In just a few months, the leap in reliability, intelligence, and autonomy is reshaping how we build software. The game is changin so fast.
#Google_IO_2025 was a turning point—revolutionary releases, fierce competition, and no more holding back.
The intelligence explosion is nearer. Google’s no longer playing modestly. True #AI game has just begun.