Analog in-memory compute is getting attention again as a way around the energy cost of moving weights between memory and compute. The recurring objection is noise,…
ci: rm GGML_HIP_ROCWMMA_FATTN (#26760) Signed-off-by: Aaron Teo aaron.teo1@ibm.com Website: https://llama.app macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS…
Britain's employment courts saw 39 percent more claims in the year through March 2026, many written with ChatGPT or Grok. The backlog jumped 55 percent to…
Available context length with and without the patch: Model: QWEN 27B ROCm stock patched Vulkan stock patched IQ4_XS Pure, single 16GB GPU 19.456 76.032 68,352 78,592…
Article URL: https://portswigger.net/research/css-the-bomb-inside-your-inbox Comments URL: https://news.ycombinator.com/item?id=49230066 Points: 5 # Comments: 0
The past week I've been running DSv4 inference on my laptop by keeping everything RAM-resident except the MXFP4-experts (since expert pool is ~147GB and won't fit)…
Article URL: https://halfhill.com/byte/1998-12_alpha.html Comments URL: https://news.ycombinator.com/item?id=49230022 Points: 5 # Comments: 2
Instead of training a new model from scratch, Google DeepMind retrofitted Gemma 4 into a diffusion model using less than 10 percent of the original training…
Hi, After some tries, it seems that for local models (27B+) the best way to have reliable outputs is to add a little more code and…
The past week I've been running DSv4 inference on my laptop by keeping everything RAM-resident except the MXFP4-experts (since expert pool is ~147GB and won't fit)…
I wonder if Houthis will get annihilated as well
Article URL: https://airy.so Comments URL: https://news.ycombinator.com/item?id=49229848 Points: 4 # Comments: 2
The AI industry's hunger for power keeps growing. Nvidia is investing up to $3 billion in Lancium, a power infrastructure developer that already has four gigawatts…
Crazy that by the time they evaluate 0731 on ARC-AGI-3, it will already be obsoleteARC Prize: @deepseek_ai Full results: https://arcprize.org/results/deepseek-v4-flash-0731ARC-AGI-3 evaluations are more operationally intensive, so…
RT DogeDesignerChristmas Eve 2008. Tesla closed a funding round in the final hour, days from bankruptcy.Elon Musk put in all his money & assets, borrowed from…
what is the utility of reporting on Russian advances?Ukrainian busification is programmed; support is not. The discourse exists to raise morale for Western funding. The optimal…
Google Deepmind is losing its autonomy, and founder Demis Hassabis may leave the AI lab for good in the coming months. AI researcher Koray Kavukcuoglu will…
InterestingFlash does seem good at RustRust plays nice with automatic verification, of courseZain: @teortaxesTex i was surprised by how good flash was at rust
Their vision was pretty good by Apr 30 standardsIf the improvement in that domain is comparable to Preview-0731 in text, it'd be awesome.Adam Karvonen: @zephyr_z9 @teortaxesTex…
Disclosure: I’m the author of Ante. DeepSeek recently reported an 82.7% score on Terminal-Bench 2.1 for DeepSeek V4 Flash 0731. Its evaluation used “DeepSeek Harness minimal…
How's it going?Philippe Lemoine: It's kind of wild that China seems to have prevented the global energy market from blowing up by massively cutting its oil…
RT Big Tech Alert🆕 @elonmusk began following @ClementDelangue(🤖👁️: any explanation?)
ChatGPT Finance for helping you save money:Trevin Chow: Used ChatGPT finance just now and asked it to find phantom subscriptions. It unearthed $550 a year of…
What Local Embedding + Reranking Models are you guys running for RAG? I went down this rabbit hole because I wanted a Embedding Model + Reranker…