Expert Upcycling reduces GPU costs by increasing the number of MoE experts midway through training.
Anthropic has overtaken OpenAI with a $1 trillion valuation based on secondary market trading.
DeepSeek has been discussing an investment from Tencent and Alibaba at a valuation of $20 billion or more.
It generates long motions using a motion embedding compressed 64x from large-scale trajectories.
DeepSeek-V4 unveiled a 1M-token context and agent optimizations.
Yeogieotti established Observability Right-Sizing criteria based on P95 and component-specific buffers.
Solved the 503 issue in Argo Rollouts canary deployments on EKS+ALB using PingPong.
By switching over 10 Next.js apps to parallel builds, deployment time was reduced from 14 minutes to 8 minutes.
Zed has unveiled a feature that lets users handle multiple agent threads in parallel within the same window.
It's an AI tool that analyzes session replays to find bugs and UX issues and automatically generates PRs.
OpenAI has launched GPT-5.5, which features enhanced work automation and coding capabilities along with improved token efficiency.
The Trump administration is easing sanctions on Anthropic and is considering allowing the use of Claude Mythos.
The US administration is softening its hardline blacklist stance on Anthropic.
Garry Tan has released GStack, a skill pack for Claude Code.
Growing a 32-expert MoE to 64 experts mid-training cut GPU hours by about 32%.
3.4 million panels and 19,000 arrays were analyzed using DuckDB and H3.
Anthropic discloses the 3 causes behind Claude Code's quality degradation and the fixes applied
In a survey of 351 IT leaders, 75% of companies reported AI failure rates of 10% or higher.
Ran LiteParse in the browser to build a PDF text extraction web app that even supports OCR.
A fake Claude API exploited Claude Code to distribute malware.
Mozart Studio 1.0, which lets you connect VST plugins in the browser to create songs with AI, has been released.
Anthropic discloses 3 causes of Claude Code quality degradation and fixes them all
OpenAI processes large-scale asynchronous requests with the Batch API and cuts costs by 50%.
An analysis of 500 Show HN landing pages tallied the ratio of AI design patterns.