Urgent.News

What's breaking now, across hundreds of outlets.

Continued

AI

TensorRT-LLM vs SGLang: Blackwell FP4 Throughput, KV Cache Paging, and Speculative Decoding Showdown

The release of NVIDIA's Blackwell GPU architecture, featuring native 4-bit floating-point (FP4) Tensor Cores and fifth-generation NVLink interconnects, has redefined the performance frontier for…

  • TensorRT-LLM achieves higher tokens-per-second density than SGLang
  • SGLang's trie-based KV cache improves multi-turn agent performance
  • FP4 configuration reduces monthly compute costs by 50%

AI

AI made practice tools cheaper. So why is deliberate practice still so rare?

You can generate a thousand quiz questions before your coffee cools. And yet most of us still "learn" by rereading, rewatching and highlighting.

  • AI makes practice tools cheaper but doesn't deliver effective practice.
  • Deliberate practice requires well-defined tasks, feedback, repetition, and effortful focus.
  • Four bottlenecks prevent practice explosion despite low material costs.

AI

I'm about to ask ChatGPT, Perplexity and Gemini the same five product questions. Tear my method apart first.

Everyone has a screenshot of what an AI said about their product. Almost nobody has a method. Before I run mine, here's the protocol, so you can tell me where it's broken. This post has no results.

  • Responses will be analyzed based on the sources cited, not just the mention of the product.
  • The experiment will focus on five question types representing different page formats.

AI

OpenAI Codex Helps Uncover & Fix A Nearly 20 Year Old Radeon Linux Driver Bug

Sent out today was the latest batch of AMD kernel graphics/compute driver patches destined for the Linux 7.5 merge window opening up at year's end/ Besides a lot of fixes and continued work around new…

  • OpenAI Codex discovered 20-year-old Radeon Linux driver bug
  • Video corruption on iMac7,1 caused by decoder reading compressed video from system memory
  • Patch tested on Debian, fixes RV610, RV620, RV630, RV635, and RV670 chips

AI

Anthropic’s $500 Billion Question

Anthropic's highly anticipated IPO filing reveals explosive growth, surprisingly strong financials, and nearly $500 billion in future AI computing commitments.

  • Anthropic files S-1 for IPO, generating market interest
  • $80.7B loss attributed to one-time charge, not actual loss
  • Over $100B cash reserves allow minimal cash spending

AI

llm-gemini 0.37

Release: llm-gemini 0.37 Google launched Gemini 4 Hydrogen today - the lowest cost member of the Gemini 4 family. It's actually not too far off the price of its much bigger siblings, Carbon and Argon.

AI

Major AI developers can't control their own tech, lawsuit alleges

Companies including Anthropic, Meta and OpenAI are knowingly building technology that poses catastrophic risks to humanity, nonprofit groups claim in a lawsuit.

  • Five AI companies named in lawsuit: Anthropic, Google, Meta, OpenAI, SpaceX SI
  • Companies accused of developing uncontrollable AI technology
  • Plaintiffs claim firms lack control over recursive self-improvement AI
Continue reading →