TileLang tutorial simplifies GPU kernel design for AI developers
A new Python DSL reduces the complexity of writing high-performance GPU kernels, targeting tensor-core GEMM, fused softmax, and FlashAttention with autotuning.
Read moreDesk archive
A new Python DSL reduces the complexity of writing high-performance GPU kernels, targeting tensor-core GEMM, fused softmax, and FlashAttention with autotuning.
Read more
GitHub explains the distinct value proposition of Copilot, highlighting its integrated developer workflow benefits compared to direct AI model API access, especially with recent billing…
Read more
Anthropic is set to deploy up to 2 gigawatts of AMD Instinct MI450 GPUs for its Claude AI models through a deal with AMD, valued…
Read more
Google Cloud's new Always-On Memory Agent, built on Gemini 3.1 Flash-Lite, offers a novel approach to LLM memory management by continuously consolidating information without relying…
Read more
Google's highly anticipated Gemini 3.5 Pro model, originally slated for a June rollout, has been delayed, with Bloomberg reporting that coding performance issues are a…
Read more
Kimi has launched K3, a multimodal open-weight model with 2.8 trillion parameters and a million-token context window. Benchmarks suggest K3 rivals models like Claude Fable…
Read more
Anthropic has integrated a web browser directly into Claude Code, allowing the AI to read, click, and type on external websites to assist developers with…
Read more
OpenAI has released its GPT-5.6 model family, including Sol, Terra, and Luna, which introduce programmatic tool calling and demonstrate significant performance improvements in coding and…
Read more
xAI has unveiled Grok 4.5, a new AI model that, despite trailing competitors like GPT-5.5 and Fable 5 in some benchmarks, offers significantly lower pricing,…
Read more
Google AI Studio now allows developers to directly import projects from GitHub into its Build mode, transforming repositories into editable and deployable applications. This update…
Read more