Protect the Wayback Machine at all costs 🙏🏻

We release today the next step for distributed training:
--> Streaming DiLoCo with Overlapping Communication.

TL;DR: train data-parallel across the world with low-bandwidth for the same performance: 400x less bits exchanged & huge latency tolerance

"Realistic" cyberpunk setting: Half your chrome doesn't work because you forgot to charge it, the companion app was discontinued, or the manufacturer's backend server is down.

And better don't travel too far, many implants are region-locked.

I love how Denmark isn’t taking any shit from Donald Trump. 🇩🇰 🤣

The man who drove a car into a Christmas market, killing five people & injuring 200, has been identified as a far-right AfD sympathizer. Once again, the far-rights hateful rhetoric translates into real-world violence. This is what far-right hate breeds: destruction, tragedy, and innocent lives lost

This week we released ModernBERT, the first encoder to reach SOTA on most common benchmarks across language understanding, retrieval, and code, while running twice as fast as DeBERTaV3 on short context and three times faster than NomicBERT & GTE on long context.

Six months ago someone put a for-loop around GPT-4o and got 50% on the ARC-AGI test set and 72% on a held-out training set redwoodresearch.substack.com/p/getting-50... Just sample 8000 times with beam search.

o3 is probably a more principled search technique...

🌍 Guessing where an image was taken is a hard, and often ambiguous problem. Introducing diffusion-based geolocation—we predict global locations by refining random guesses into trajectories across the Earth's surface!

🗺️ Paper, code, and demo: nicolas-dufour.github.io/plonk