Articles

The article is a compilation of several news items and an editorial on the state of LLMs. The editorial by Andrew Ng argues that while LLMs are more general than previous AI, they still lack human-level generality and require laborious data-centric AI approaches for specific tasks. It highlights the need for continued hard work to build more intelligent models. Following this, the article introduces Runway's GWM-1 family of general world models for real-time, coherent video generation across different applications like simulated worlds, robotics, and conversational avatars. It then covers OpenAI's launch of GPT-5.2 models (Pro, Thinking, Instant) to address competitive pressures, showcasing their improved performance in reasoning benchmarks and significantly lower computational costs. Finally, it details SEMI, a sample-efficient method using a projector and LoRA adapters to adapt LLMs to new data types and specialized domains with minimal examples, greatly reducing the data burden for multimodal integration.