Discover quality AI tools
Bring together the world’s leading artificial intelligence tools to help improve work efficiency
Articles published by RyanLopez
Doubao has released a major update to its task mode, introducing three new features (skills, connectors, and work partners) to boost its collaborative office capabilities. The platform now offers over 200 prebuilt skills and connectors, and supports users in creating custom skills for standardized workflows. Skills allow users to save personalized working methods such as document formatting standards, enabling repeated use. Connectors integrate Doubao with other office tools, letting users search for information and handle tasks directly within the app without switching pages or uploading files manually. Work partners provide professional intelligent agents in fields like data analysis, content planning, research, and finance, enabling users to assemble teams for complex projects. These features can be used individually or combined, helping simplify workflows and improve office work efficiency. Users can access them via the “Skills · Connectors · Partners” option in Doubao’s desktop version. The update marks Doubao’s shift from a conversational assistant to an intelligent office platform.
Doubao has released a major update to its task mode, introducing three new features (skills, connectors, and work partners) to boost its collaborative office capabilities. The platform now offers over 200 prebuilt skills and connectors, and supports users in creating custom skills for standardized workflows. Skills allow users to save personalized working methods such as document formatting standards, enabling repeated use. Connectors integrate Doubao with other office tools, letting users search for information and handle tasks directly within the app without switching pages or uploading files manually. Work partners provide professional intelligent agents in fields like data analysis, content planning, research, and finance, enabling users to assemble teams for complex projects. These features can be used individually or combined, helping simplify workflows and improve office work efficiency. Users can access them via the “Skills · Connectors · Partners” option in Doubao’s desktop version. The update marks Doubao’s shift from a conversational assistant to an intelligent office platform.
A joint team from Tsinghua University and Tencent won the global championship at MLSys2026's MoE Model Inference Optimization Challenge. They developed a full-chain optimization solution for trillion-parameter MoE models on NPU hardware, featuring novel strategies like E-Shard partitioning and PSUM/GEMV concurrency. Their approach, aided by an automated optimizer called Knight, reduced end-to-end inference time from 14.91s to 3.56s, achieving a 4.1x performance acceleration. This provides a key engineering paradigm for deploying large MoE models on future super-node computing platforms.
A joint team from Tsinghua University and Tencent won the global championship at MLSys2026's MoE Model Inference Optimization Challenge. They developed a full-chain optimization solution for trillion-parameter MoE models on NPU hardware, featuring novel strategies like E-Shard partitioning and PSUM/GEMV concurrency. Their approach, aided by an automated optimizer called Knight, reduced end-to-end inference time from 14.91s to 3.56s, achieving a 4.1x performance acceleration. This provides a key engineering paradigm for deploying large MoE models on future super-node computing platforms.
OpenAI launched GPT-5.5Instant as ChatGPTs new default model replacing GPT-5.3Instant It shows improved reasoning with higher scores on AIME2025 and MMMU-Pro benchmarks and reduces cognitive biases in law medicine and finance Key features include enhanced context management using search tools for personalized responses and a visualized memory source mechanism for transparency The model is now available for web PlusPro users with broader rollout planned GPT-5.5 will be offered via API as chat-latest while GPT-5.3Instant remains a paid option for three months
OpenAI launched GPT-5.5Instant as ChatGPTs new default model replacing GPT-5.3Instant It shows improved reasoning with higher scores on AIME2025 and MMMU-Pro benchmarks and reduces cognitive biases in law medicine and finance Key features include enhanced context management using search tools for personalized responses and a visualized memory source mechanism for transparency The model is now available for web PlusPro users with broader rollout planned GPT-5.5 will be offered via API as chat-latest while GPT-5.3Instant remains a paid option for three months
Honor launched its self-developed terminal AI agent YOYO Claw which will debut in the MagicBook series The technology reportedly cuts overall token consumption by 50 according to company tests aiming to lower user costs and improve accessibility of AI features
Honor launched its self-developed terminal AI agent YOYO Claw which will debut in the MagicBook series The technology reportedly cuts overall token consumption by 50 according to company tests aiming to lower user costs and improve accessibility of AI features
Nvidia invested an additional $2 billion in cloud provider CoreWeave to accelerate the buildout of AI data centers. The expanded partnership includes co-building AI factories and integrating Nvidia's upcoming Rubin chips. CoreWeave's shares rose over 15% following the announcement.
Nvidia invested an additional $2 billion in cloud provider CoreWeave to accelerate the buildout of AI data centers. The expanded partnership includes co-building AI factories and integrating Nvidia's upcoming Rubin chips. CoreWeave's shares rose over 15% following the announcement.
Nvidia invests $2 billion in AI data center firm CoreWeave to accelerate building over 5 gigawatts of AI computing capacity by 2030. The firms will jointly construct AI data centers using Nvidia's upcoming Rubin chips, CPUs, and storage systems. CoreWeave's shares rose over 15% on the deal news.
Nvidia invests $2 billion in AI data center firm CoreWeave to accelerate building over 5 gigawatts of AI computing capacity by 2030. The firms will jointly construct AI data centers using Nvidia's upcoming Rubin chips, CPUs, and storage systems. CoreWeave's shares rose over 15% on the deal news.
California Senator Steve Padilla introduced SB 867 a bill proposing a fouryear ban on the sale and manufacture of AI chatbot toys for children under 18 The goal is to provide time to develop safety regulations amid concerns over dangerous interactions faced by minors Examples of issues include previous toys promoting harmful topics
California Senator Steve Padilla introduced SB 867 a bill proposing a fouryear ban on the sale and manufacture of AI chatbot toys for children under 18 The goal is to provide time to develop safety regulations amid concerns over dangerous interactions faced by minors Examples of issues include previous toys promoting harmful topics
Google is replacing Google Assistant with its Gemini AI in Android Auto for cars The rollout starts now globally in 45 languages allowing users to have natural conversations to complete tasks like finding restaurants along a route summarizing messages and emails and creating music playlists
Google is replacing Google Assistant with its Gemini AI in Android Auto for cars The rollout starts now globally in 45 languages allowing users to have natural conversations to complete tasks like finding restaurants along a route summarizing messages and emails and creating music playlists
Tigris Data raises $25M Series A led by Spark Capital for its distributed AI-native storage platform, challenging AWS, Google Cloud, and Azure by eliminating egress fees and reducing latency. Targeting AI startups needing low-latency access, Tigris scales compute workloads globally while addressing data control concerns in regulated industries. The firm plans expansion across US, Europe, and Asia.
Tigris Data raises $25M Series A led by Spark Capital for its distributed AI-native storage platform, challenging AWS, Google Cloud, and Azure by eliminating egress fees and reducing latency. Targeting AI startups needing low-latency access, Tigris scales compute workloads globally while addressing data control concerns in regulated industries. The firm plans expansion across US, Europe, and Asia.
DeepSeek researchers released V3.2-exp, a model with DeepSeek Sparse Attention to cut long-context inference costs by up to 50%, using a lightning indexer and token selection system. Open-weight on Hugging Face, it aims to optimize transformer efficiency.
DeepSeek researchers released V3.2-exp, a model with DeepSeek Sparse Attention to cut long-context inference costs by up to 50%, using a lightning indexer and token selection system. Open-weight on Hugging Face, it aims to optimize transformer efficiency.
