Nvidia is proposing a beast of a CPU system for Windows PCs. It has 128 GB of shared memory and comes with up to 6,144 state-of-the-art CUDA cores. CPU wise, the chip has 10 performance cores and 10 efficiency cores. The performance cores are based on the Cortex-X925. These chips appear to support six 128-bit SIMD execution units (SVE2), not as good as recent AMD chips, but better than Apple Silicon (on paper). The game changer is the unified 128 GB memory. That is the path Apple took years ago. Instead of separate memory for the CPU and GPU, everything shares a single pool. It is increasingly popular. The memory is not as fast as dedicated GPU memory, but it is cheap enough while delivering enough bandwidth to run AI models locally. I am not sure how many people will run AI models locally. It still seems like a niche application to me. However, it will make decent machines to play video games. It will be interesting to see how Intel and AMD respond. I think that the AVX-512 instructions supported by all recent AMD processors are far superior to the SVE2 instructions of the Cortex-X925. They can eat more data and they are more versatile. But Intel has been shy, thus far, in making it available on customer systems.

a reminder that branding matters. https://t.co/y9WOLdp3GM

a reminder that branding matters. https://t.co/y9WOLdp3GM

Introducing FrontierCode: a coding eval that raises the bar for difficulty & quality. Each task took 40+ hrs of work by leading open-source maintainers. Models write sloppy code that works but isn’t maintainable. Our eval is first to measure: would you actually merge this code?

DeepSeek-V4 now runs 500K context with 90% less KV cache FlashMemory introduces Lookahead Sparse Attention: a tiny Neural Memory Indexer predicts which chunks your next tokens will need, keeping only 13.5% of cache in GPU memory—with zero backbone retraining and better accuracy. https://t.co/qxclBmFV3n

Why Fable 5 til the 22nd of June? Wtf?! @claudeai https://t.co/xjvxnT3F48

Why Fable 5 til the 22nd of June? Wtf?! @claudeai https://t.co/xjvxnT3F48

Fable 5 lies 96% of the time. We were surprised by it's skill... 🧵 https://t.co/YtsLo9sW6j

Claude Fable can run autonomously for days. This is the single highest-leverage prompt I use to stay on top of it: "Spin up a persistent HTML page. As you work, append clear, timestamped updates with screenshots/media so I can follow along." Literally a 10x better experience. https://t.co/TKtsIALzxY

Gaussian splats shouldn't have locked lighting 💡 We’ve been working on a way to relight them in Blender by baking proxy mesh lighting directly into the splat color. I'm putting this out under an "experimental" tag because the performance needs a lot of work and it's a rough work in progress. That said, it’s fun to mess around with on smaller scans. If you want to see how it works, I left the YouTube tutorial in the comments. 👇

First time I’ve seen it in person: Fast & frequent lightning strikes from overhead are incredible. DCA -> SFO. High flash rate means the storm is undergoing rapid updraft and ice glaciation. https://t.co/pOXH9XcpHy

Posted this in 2015. Turns out it was fun & the future. https://t.co/a8nicydTmn

Clubhouse in 2021 was unstoppable https://t.co/xQX9sY4z6T

Clubhouse in 2021 was unstoppable https://t.co/xQX9sY4z6T

I made a looper where you can record midi or audio, and have magenta or stable audio respectively transform it with a prompt. Beatbox and turn that into drums? Check. Play some rough chords and get an orchestra? Check :) Had this app in my head for years! https://t.co/eVlkrF1hqL

@andrewmccalip https://t.co/MQbmzJYqCW The new magenta real-time music models are out, might keep playing with those :)

@leefcallender I'd not call it fit to share yet, but I've dumped the code on GitHub here: https://t.co/3zinhj0ptS Lots of stuff to add and agent fluff to remove haha

@alex_paul_kelly @field_synthesis Yup, https://t.co/cLl4Ye1Rhx will hopefully have some good bio content eventually :) I have a few long-running projects that, when finished, should motivate me to put more work into the editing+ sharing bit.

@_arohan_ Preference tuning helped with egregious nasty-looking artifacts but honestly I think a lot of it comes down to just better + bigger models. (Image from PARTI, 2022, my personal "ah right scaling will solve lots of image stuff" moment) https://t.co/UTVEWrmvYF

@gheorgheiuga @intellectronica :( https://t.co/4koa8RLhbT

@stevekrouse @simonw Also don't dare ask follow on questions https://t.co/YRhpX30VPo

Yay 3D printing! Model source: https://t.co/a6d6ErSCin https://t.co/rJdXRtMghU

Codexで強化学習 生き物つくってもうた。。 https://t.co/uQDPUMwpV4

Codexで強化学習 生き物つくってもうた。。 https://t.co/uQDPUMwpV4

I'm making a drawing app inspired by the pen + brushpen way I do my nature journals. The watercolor wash dynamics mean you can't be too precious about every line. Give it a go! https://t.co/u7IQI8j0Sl https://t.co/OfKsoYl22F

Tests as I worked on this this morning, I'll do some better drawings now but I want to share this right away so y'all can play too :) https://t.co/AzTJe9vXSW

+2 more

This is one of my favourite ways to sketch things quickly IRL, I'm interested to see if scribbling away on an iPad feels close enough to the real thing to make it compelling :) https://t.co/YLHbtX4efR

+2 more

It's such fun on an iPad! Draw with pencil, brush around with finger :) (hard to film and draw though) https://t.co/xYsTk7LKO4

Step 3: Install FAISS Import FAISS https://t.co/LlsazWkAB6

Step 8: Import VectorDBQAWithSourcesChain tool from @langchain library. Create a chain using VectorDBQAWithSourcesChain tool VectorDBQAWithSourcesChain will take the question and lookup for the documents from the vector database (created by FAISS) stored in variable 'store'. https://t.co/utuv7F6GWz

Here is the Colab notebook containing the entire project. NB:https://t.co/Rhj7e7B7JQ

Step 1 - Install Langchain. We will use the @langchain library to build a Large Language Model. https://t.co/B8fEgg8JQl