Presented my poster on EWRL today. Check out the benchmark I made here: https://t.co/zweTQSBUe0.
I went for a comic poster design this time, seems people like it.
We created an @factoriogame like GPU accelerated RL simulation environment called Factoriax. We released it today, link below :).
Come say hi to us at #EWRL2026!
New preprint!🚀
We introduce NeuroTrails: ensemble-level performance out of a single model.
1⃣Split into multiple heads
2⃣Sparsify the paths
3⃣Let the topology evolve on-the-fly
This creates powerful diversity with distinct neural trails. Works on ResNets & LLaMA models!
Nate Soares and I are publishing a traditional book: _If Anyone Builds It, Everyone Dies: Why Superhuman AI Would Kill Us All_. Coming in Sep 2025.
You should probably read it! Given that, we'd like you to preorder it! Nowish!
I really enjoyed this interview, @TEDChris & @sama! I’d love to hear you answer these additional questions:
1) Why do you expect superintelligence to remain an empowering tool under human control?
2) If we can build amazing tool AI that complements us but can’t replace or control us, then what’s the value-add of also building superintelligence that makes us all economically obsolete?
Pols 2017: "#slaughterbots are sci-fi"
Pols 2025: "#slaughterbots are here, but will never be used by terrorists as WMDs”
The best defense against bioweapons isn’t bioweapons & the best defense against slaughterbots isn’t slaughterbots. I wish I’d been wrong about the need for a treaty, but it’s not too late, and momentum is building:
What if Studio Ghibli directed Lord of the Rings?
I spent $250 in Kling credits and 9 hours re-editing the Fellowship trailer to bring that vision to life—and I’ll show you exactly how I did it 👇🏼
Absolutely lovely talk by @RemiCadene at the Gemma Dev Day on the LeRobot project. Really cool to see the possibilities, especially on low cost hardware. If you are at all interested in Robotics, check it out! https://t.co/x6JBqiCw5m
At the Gemma3 release keynote. Multi modal support, long context lengths and high performance (27B comparable to DeepSeek-R1) for its size according to Chatbot Arena. Focus on multi-lingual support. Let's see where this goes.