The Ascend NPU-based SLAI T-Rex framework achieved 34.22% MFU in post-training the trillion-parameter DeepSeek-V4 MoE model.
PaddlePaddle has released HPD-Parsing, a 1B document parsing model based on hierarchical parallel decoding, achieving the top score of 94.91% on OmniDocBench v1.6.
Poolside's Laguna model claims to be its own model when asked in English, but admits to being based on Qwen when asked in Chinese.
GigaToken, a tokenizer that boosts text processing speed to GB/s levels using SIMD and caching techniques, has been released.
feyn, the developer of Chonkie, has released the SQRL model series under Apache 2.0, which generates SQL after directly exploring the DB.
Poolside's Co-CEO Eiso Kant explains the training infrastructure and strategy behind building Laguna S, a 118B MoE model.
pi-computer-use, an MIT open-source extension that lets AI agents directly operate desktop apps without APIs, has been released.
Following the 2024 Network Separation Improvement Roadmap, this guide walks financial companies step by step through safely implementing R&D networks using an AWS multi-account architecture.
NVIDIA has installed and activated a DGX GB300 AI supercomputer at the U.S. Naval Postgraduate School (NPS).
This piece organizes, from a practical perspective, the 7-step process and safety assurance measures for cloud adoption by financial companies based on the Electronic Financial Supervisory Regulation.
LongCat-2 revealed that it trained a 1.6 trillion-parameter model using only 50,000+ Chinese-made AI ASICs.
The Petals project uses a BitTorrent-style approach to let consumer GPUs run huge LLMs like Llama 3.1 405B in a distributed manner.
In a webinar, ElevenLabs revealed 7 key techniques for making AI voice agents sound as natural as humans.
Rosebud AI has integrated the ElevenLabs API to provide music and sound effects by default in every generated game.
FuriosaAI CEO June Paik met Princess Anne at a British Embassy Seoul event to discuss Korea-UK AI technology cooperation.
Grok Build has launched a Workflow feature that orchestrates hundreds of parallel agents.
Vercel Flags has added a real-time evaluation status and metrics view for each flag.
A new Connect tab has been added to the Vercel dashboard, allowing direct access to Sandbox shells and lifecycle management right from the browser.
The `vercel flags versions` command has been added to Vercel CLI, allowing you to check the entire revision history of feature flags.
GitHub Tools has been released as an extension package for the eve agent, allowing 42 tools to be registered with a single file.
OpenAI has launched a Health feature that connects Apple Health and medical records to ChatGPT for users in the US.
The Vercel MCP server has started supporting the ability to deploy code directly and return a shareable URL within an AI assistant chat.
Ant Group's Ling 3.0 Flash model has been added to Vercel AI Gateway and is available for free until August 3.
Anthropic is internally testing a 'Managed Projects' feature in which Claude autonomously performs tasks and manages projects.
AI Briefing