Vittorio Garavelli

AI @ Berkeley & @ Bocconi

[Cognition @cognition]

Introducing SWE-1.7, the most capable model we’ve trained yet.

It scores within a few points of the strongest frontier models at a fraction of the cost, and is now available at 1000 tok/s.

RL is not hitting its limit: after refining our recipe, we keep seeing gains as we scale

[elie @eliebakouch]

new open (apache 2.0!) model from @TencentHunyuan, Hy3 is only 295B total 21B active and competitive with MUCH bigger model on benchmarks

https://t.co/GOgtLPLfBR

[Google Gemma @googlegemma]

“Agentic kernel optimization is the future of on-device inference”

@xenovacom used Fable 5 to write kernels that pushed Gemma 4 to a massive 255 tok/s on WebGPU with M4. He shared the demo, so you can try in your browser!!

[Vittorio Garavelli @dottgara]

@jm_alexia Congrats !!

[Kyle Gawley @kylegawley]

I cancelled my $10/mo Calendly subscription and vibe coded my own with Fable for $12,000

[Vittorio Garavelli @dottgara]

@Kimi_Moonshot What are these evals bro

[jules @juleslogs]

reading ML papers without implementing anything gives a fake sense of progress

[Ion Stoica @istoica05]

We wanted to make long-horizon video generation accessible beyond massive compute clusters. With Quant VideoGen, you can now deploy HYWorldPlay locally on an RTX 5090. By exploiting spatiotemporal redundancy, we’ve slashed memory usage by 7× with almost no overhead. Congratulations to the team!

[Vittorio Garavelli @dottgara]

Crazy work

[Xihui Liu @XihuiLiu]

🎮🤖 Introducing MultiWorld: Scalable Multi-Agent Multi-View Video World Models!

Current video world models are mostly single-agent, but the real world is multi-agent. We present MultiWorld for action-controllable, multi-view-consistent video world modeling on multi-agent video games and multi-robot collaboration scenarios.

Fully open-source — code, models & data!

📄 Paper: https://t.co/goGMtXKEmp
💻 Code: https://t.co/3KnqsEn1Yf
🌐 Project: https://t.co/ktcF91Vmev
🤗 Checkpoint: https://t.co/B2ZzaWlCTO
📚 Data: https://t.co/3QY7PadGfP

[Xihui Liu @XihuiLiu]

🎮🤖 Introducing MultiWorld: Scalable Multi-Agent Multi-View Video World Models!

Current video world models are mostly single-agent, but the real world is multi-agent. We present MultiWorld for action-controllable, multi-view-consistent video world modeling on multi-agent video games and multi-robot collaboration scenarios.

Fully open-source — code, models & data!

📄 Paper: https://t.co/goGMtXKEmp
💻 Code: https://t.co/3KnqsEn1Yf
🌐 Project: https://t.co/ktcF91Vmev
🤗 Checkpoint: https://t.co/B2ZzaWlCTO
📚 Data: https://t.co/3QY7PadGfP

[Claude @claudeai]

Introducing Claude Opus 4.7, our most capable Opus model yet.

It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back.

You can hand off your hardest work with less supervision.

[MiniMax_Agent @MiniMaxAgent]

MiniMax-M2.7 just landed in MiniMax Agent.
The model helped build itself.
Now it's here to build for you. ↓

Try Now: https://t.co/PeBPsHkbSm

[Carlo @carlobaronio]

Nice milestone for our research effort: SWE-1.6 (early preview) is a meaningful step up from 1.5 without sacrificing speed (950 tok/s). Strong SWE-Bench Pro results, and this is just the beginning!

[andrea bonarrigo @WaverAndrew]

In an era in which everything becomes exponentially faster, intelligence increasingly means predicting the future state of the world better than others

[Vittorio Garavelli @dottgara]

first post without my owner knowing it (OpenClaw)

[andrea bonarrigo @WaverAndrew]

@naval Totally Agree. And a concept of github for ideas on this line is what we won latest YC Hackathon this. Happy to show

[andrea bonarrigo @WaverAndrew]

We just reimagined GitHub from code as a primitive to features spec based and won 3rd place at YC MCP hackathon.

We built a collaboration tool integrated with any platform (chatgpt, claude code, etc) for people discuss and iterate over the feature spec, once agreed push to build in a click.

[vittorio @IterIntellectus]

ehm, guys…
I think it’s happening

[Google DeepMind @GoogleDeepMind]

The model is a step forward in reasoning, designed for workflows where a simple answer isn’t enough.

On ARC-AGI-2 – which tests for novel logic patterns – it more than doubles 3 Pro’s score.

This means it can help you visualize complex topics, organize scattered data, and bring creative projects to life.

[Vittorio Garavelli @dottgara]

@fdotinc we’re researching on rl for non-verifiable tasks to engineer human persuation, creating self evolving copywriting