Browse AI news across models, agents, media, industry, and compute policy.
CMU researchers leverage large models to automatically optimize vision-language prompts without parameter access. This AI breakthrough is part of the forward-looking topics series (batch 3), published outside the main timeline of Jan 2025–May 2026.
Tencent has released its largest open-source Mixture of Experts (MoE) model with 389 billion parameters, available free for commercial use and demonstrating superior performance compared to Llama 3.1.
Training-free knowledge editing for large models enables more efficient absorption of new data. This is an AI forward-looking topic (batch 3, outside the main timeline of Jan 2025–May 2026).
Exploring how a female student from Chengdu influenced the development of artificial intelligence, highlighting her contributions beyond mainstream timelines.
Three large models challenge o1. A real-world test demonstrates how collaboration among over 360 models can replace prompt engineering.
A Tsinghua University team has established a '3D Scaling Law,' pushing the boundaries of AI-generated 3D content. This research addresses forward-looking topics outside the main timeline.
Boosting inference efficiency by over 200% and matching vLLM in usability, this domestic framework raises questions about its origins. (AI Forward Agenda, batch3; outside the main timeline of Jan 2025–May 2026).
TuSelf-driving addresses all rumors, stating its shift to AIGC gaming is essential for survival. (AI Forward Agenda, batch 3; published outside the main timeline of Jan 2025–May 2026).
Tencent launches GameGen-O, an AI video model capable of generating game footage reminiscent of Black Myth: Wukong. This initiative falls outside the main timeline of AI forward-looking topics (batch 3, scheduled for Jan 2025–May 2026).
A new AI coding tool is trending, enabling users to build apps in just two minutes using their smartphones.