Browse AI news across models, agents, media, industry, and compute policy.
A new multimodal segmentation method is proposed to enhance AI's understanding of the physical world. This topic falls outside the main timeline (2025-01–2026-05).
MiHoYo co-founder Cai Haoyu has launched a new AI company, recruiting a key scientist from the Llama 3 team to lead its large language model development.
Shanghai Jiao Tong University research explores whether LLMs can detect contradictions within prompts. This topic falls outside the main AI foresight timeline (Batch 3, Jan 2025–May 2026).
AI frontier topic (batch3, outside main timeline Jan 2025–May 2026): Another test reveals large language models struggle to count the letter 'r' in 'Strawberry'.
A home-built AI cluster using PCs and tablets runs a 400B-parameter model, attracting significant attention on GitHub.
A new algorithm from Meitu and the University of Chinese Academy of Sciences accelerates AI video restoration tenfold, correcting overexposure and color shifts frame by frame.
AI models collectively fail to compare 9.11 and 9.9, highlighting persistent reasoning flaws in large language models.
Claude's coding tool now allows one-click sharing, with early adopters already posting their projects. This is part of the AI Forward Agenda (batch 3).
An analysis of why top-tier Chinese large model developers are opting for CPU-based architectures, exploring the strategic rationale behind this hardware choice.
A new RNN architecture outperforms Transformers by treating each hidden state as an independent model. The first author claims this approach fundamentally changes language modeling.