Attention Residuals Revisited kexue.fm/archives/11664

theguardian.com/us-news/ng-int…

Thank you The Washington Post for citing our research. Folks, here are the tankers some of you asked about. wapo.st/3OrEyq0 #OOTT #IranWar #Tankers

I upgraded my Claude token counter tool to compare different models and Opus 4.7 does appear to use 1.46x times the tokens for text and up to 3x the tokens for images - it's priced the same as Opus 4.6 on a per-token basis so this is actually a pretty big price bump

My bold prediction: within the next two years, there will be cases where GPUs cannot be deployed because of CPU shortages. According to industry checks, in some Rubin Ultra configurations, the GPU-to-CPU deployment ratio has already exceeded one GPU to two CPUs.

Interesting shift. These highly subsidized subs are out there to get your code to improve their models. If you use AI for things useful to you, but not code, you are not valuable to them. reddit.com/r/SillyTavernA…

At OFC 2026 last month,  Cisco's chief architect Rakesh Chopra presented on scale-across networking architectures and key deployment trends driving strong demand for traditional DCI equipment. "Traditional" DCI connects CPUs across the frontend network while scale-across connects

R.I.P & thanks Sora

Chrome sucks ? support.google.com/chrome/answer/…

SK hynix recorded KRW 10 trillion ($6.76 billion) in gain on valuation of investment assets in the first quarter alone, driven by its stake in Kioxia.

🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length. 🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models. 🔹 DeepSeek-V4-Flash: 284B total / 13B active params.

this is so amazing, CSA is a new attention arch close to deepseek NSA imo, but sequential instead of in parallel. NSA had this compression of KV, here it's the same, and then they do DSA, and sliding window to keep good local context (also in NSA)

Huge AI models are overwhelming servers and leaving high-powered chips idle. Startup Majestic Labs AI, founded by Google and Meta veterans, says it has the solution wsj.com/tech/chip-star… via The Wall Street Journal

Windows and macOS, routes are accepted by default. The operating system will prioritize routes with the longest prefix match, or in other words the most specific of all configured routes. tailscale.com/docs/reference…

>> AI Demand Drives Inventory Shortages: Apple Raises Mac mini Starting Price by $200 • Apple has raised the starting price of the Mac mini from $599 to $799, reflecting inventory shortages driven by AI demand and tighter processor supply. The price increase was effectively

Andrej Karpathy: From Vibe Coding to Agentic Engineering youtu.be/96jN2OCOfLs?si… via YouTube

原来上游 immortalWrt/OpenWrt 不合并 高通IPQ NSS 的原因不仅仅是高通闭源 forum.openwrt.org/t/qualcommax-n…

Current memory-stock summary: Korean memory stocks surge → U.S. memory stocks rally in response → U.S. memory stocks rally, so Korean memory stocks surge again → then U.S. memory stocks rally again.

人家缺的是 eSSD (SLC pro.logitec.co.jp/column/a202202…