A recent study from Columbia University's Tow Center for Digital Journalism sounds the alarm on the accuracy of AI search services. The research reveals significant accuracy issues in ChatGPT's citation of news sources, with even publishers directly cooperating with OpenAI not spared. The research team conducted a comprehensive test on 200 news citations from 20 different publishers, resulting in shocking findings: ChatGPT had errors or partially incorrect source information in as many as 153 cases.

With the rapid development of artificial intelligence technology, AI-generated videos are becoming increasingly realistic, posing serious security risks to society. Recently, a $25 million fraud case utilized a lifelike video created by AI. To address this challenge, a research team at Columbia Engineering has developed an innovative tool called DIVID (DIffusion-generated VIdeo Detector) designed to detect AI-generated videos.

Researchers from Apple and Columbia University have developed the Ferret multimodal language model, aimed at achieving advanced image understanding and description. The Ferret model possesses strong global understanding capabilities, able to handle free text and referenced regions, outperforming traditional models. The researchers created the GRIT dataset, which includes 1.1 million samples, to guide the model in reference and localization tasks. Ferret-Bench evaluations show that Ferret outperforms the best MLLM models by an average of 20.4% while reducing object-related errors.

The multi-modal large model 'Ferret' developed by Apple's AI/ML team in collaboration with Columbia University has successfully tackled Google's CAPTCHAs, being capable of recognizing traffic lights and improving the accuracy of large models in 'see-say-answer' tasks. Ferret's innovation lies in its ability to integrate spatial understanding of references and locations while comprehending semantics and targets, distinguishing it from traditional multi-modal models. By employing a hybrid regional representation method that combines discrete coordinates and continuous features, the model has demonstrated exceptional performance in multi-task evaluations, particularly in reference and visual grounding.

The AI research assistant developed by researchers directly replaces you in writing papers, allowing humans to only focus on graduating.