Search AI Products and News

A recent study led by the Department of Cognitive Science at the University of California, San Diego, has achieved a landmark breakthrough in artificial intelligence: OpenAI's latest model, GPT-4.5, has demonstrated superior performance to humans in a standard Turing test using a persona, becoming the current AI system with the most human-like conversational abilities. This achievement not only reshapes our understanding of AI's language capabilities but also unlocks new potential for AI applications in social intelligence. The experiment compared four representative AI systems.

A recent study released by the University of Munich, the Munich Machine Learning Center, and Adobe Research shows that 12 top AI language models, including GPT-4o, Gemini1.5Pro, and Llama-3.3-70B, face significant performance degradation in long-text conceptual reasoning tasks. Despite these models supporting context processing of at least 128,000 tokens, their ability to establish deep logical connections still has fundamental limitations. The research team developed NOLIMA (No Text Matching).

In the application of artificial intelligence, achieving real-time interaction with AI has always been a significant challenge for developers and researchers. Among these challenges, integrating multimodal information (such as text, images, and audio) to form a coherent dialogue system is particularly complex. Although advanced large language models like GPT-4 have made some progress, many AI systems still struggle with achieving smoothness in real-time conversations, context awareness, and multimodal understanding, which limits their effectiveness in practical applications.

Recently, a new study indicated that even advanced AI language models, such as OpenAI's latest o1-preview, struggle with complex planning tasks. This study was conducted by scientists from Fudan University, Carnegie Mellon University, ByteDance, and Ohio State University, testing the performance of AI models on two planning benchmarks: BlocksWorld and TravelPlanner. In the classic planning scenario of BlocksWorld...

In the rapid advancement of artificial intelligence, an international research team is paving the way for the development of AI language models in Europe. They have launched a project called MOSEL (Massive Open-source compliant Speech data for European Languages), creating a comprehensive open-source voice dataset for the 24 official languages of the European Union. This initiative aims to promote the development of open AI language models in Europe, challenging the current dominance of English.

Recently, researchers from the Mila Institute, Google DeepMind, and Microsoft Research conducted an in-depth investigation into the reasoning capabilities of AI language models. They found that small and inexpensive models exhibit significant shortcomings when addressing complex problems. The study utilized a test called 'Combinatorial GSM' to evaluate how these models perform in solving chain-based foundational math problems. Image source note: Image generated by AI, image licensed through Midjourney.

SenseTime Technology, in collaboration with Thailand's DTGO Group and Quinnnova, jointly launched the Thai Language Large Model (DTLM) named "Dongfeng". This is the world's first AI large language model that can operate efficiently in Thai, Chinese, and English. The model combines SenseTime's base model and computational advantages with DTGO's deep understanding of Thai language and culture, aiming to provide localized generative AI experiences, including text comprehension and natural, fluent r

Bilibili has announced the open-source release of the Index-1.9B model, attracting widespread attention. The Index-1.9B model includes multiple versions, including the base model, control group, and dialogue model.

Meta plans to launch Llama3 in July, featuring powerful reasoning abilities and accurate prediction capabilities. Llama3 will better respond to users' controversial questions, enhancing the application value of artificial intelligence technology. The new version employs advanced model architecture and algorithms to improve language understanding and generation performance. Meta demonstrates its ongoing investment in the AI field, driving the development of artificial intelligence technology.

Recent research has found that AI language models can assist in identifying schizophrenia, providing hope for early diagnosis and treatment. The research team combined AI with brain scanning technology to analyze the speech characteristics of schizophrenia patients. The results showed that the language of schizophrenia patients is more difficult for AI models to predict, which helps in assessing the severity of the patients' symptoms. Although the research has some limitations, it offers new possibilities for the early diagnosis and treatment of schizophrenia. Early treatment of schizophrenia holds promise for improving patient outcomes and alleviating suffering and damage.

Scientists at University College London's Institute of Neurology have developed a new tool based on AI language models that can characterize subtle features in the speech of schizophrenia patients. The research aims to understand how automated language analysis can improve the diagnosis and assessment of mental disorders. The study found that the language performance of schizophrenia patients was less predictable by AI models compared to the control group, which may relate to brain learning and cognitive mapping. Scientists plan to widely apply this technology, potentially leading to breakthroughs in the diagnosis of mental illness in the future.

["The Allen Institute for Artificial Intelligence has released the open text dataset Dolma, aimed at promoting transparency and innovation in AI language models.", "Dolma, as the core of AI2's open language model initiative, provides researchers and developers with free data resources.", "The Dolma dataset boasts a scale of 3 billion tokens, operates under the mid-risk ImpACT license, and encourages users to provide contact information and usage feedback."]

Researchers have found political bias in AI language models. ChatGPT and GPT-4 are considered the most left-leaning liberal models, while Meta's LLaMA is regarded as the most right-leaning authoritarian model. The results of this study serve as a reminder to remain vigilant of political bias in the development and use of AI language models and to ensure fairness and diversity in expression.