Microsoft's Progress in Training Mini Language Models is Remarkable

Microsoft Research Reveals: Overuse of AI May Diminish Human Critical Thinking Skills

A recent study from Microsoft and Carnegie Mellon University indicates that excessive reliance on artificial intelligence (AI) tools may be eroding people's critical thinking abilities. The research team surveyed 319 knowledge workers and collected 936 real-world examples of generative AI usage across various industries, including IT, design, administration, and finance. The study analyzed six dimensions of critical thinking: knowledge, understanding, application, analysis, synthesis, and evaluation. The findings reveal that people's thinking patterns in problem-solving change significantly after using AI tools.

Are You Really Still Thinking? Microsoft Research Shows Generative AI May Deteriorate Critical Thinking!

Recently, researchers from Microsoft Research and Carnegie Mellon University released a new study revealing potential issues for knowledge workers using generative AI (such as Copilot and ChatGPT). The research team surveyed 319 knowledge workers who use generative AI weekly to explore their application of critical thinking while using these tools. Image source note: Image generated by AI, courtesy of image licensing provider Midjourney. The research findings indicate that those...

AI2 Open Source Training Plan Tülu 3 Breaks the Monopoly of Post-Training Techniques for Large Models

OpenAI Shifts to New Training Techniques: Scaling Models is No Longer the Key Breakthrough

Mobile Large Model Runs 5 Times Faster! Microsoft Research Asia Opens New Technology for Ultra-Fast Experience on CPUs

The T-MAC (Table-Lookup-based MAC) technology aims to address the memory and computational constraints of deploying large language models (LLMs) on edge devices. By quantizing model weights into low-bit representations, T-MAC replaces traditional multiplication operations with look-up tables (LUT), significantly enhancing running efficiency on CPUs. This method greatly reduces the memory required for calculations, enabling billion-parameter LLMs to run efficiently on resource-constrained devices for smart upgrades. Compared to existing implementations, T-

Microsoft Research Introduces AI Framework E5-V: Simplifying Multimodal Learning with Text Pair Unimodal Training to Reduce Costs

Recently, a research team from Microsoft Research and Beihang University has jointly introduced a novel framework called E5-V, aimed at providing a more efficient solution for multi-modal embeddings. With the continuous advancement of artificial intelligence, multi-modal large language models (MLLMs) have become a focal point of research, as they are capable of understanding both textual and visual information simultaneously, thereby better handling complex data relationships. However, effective

Microsoft Researchers' SpreadsheetLLM Project Teaches AI to Understand Spreadsheet Content

Microsoft researchers have recently unveiled an innovative research called SpreadsheetLLM, aimed at addressing the challenges faced by large language models (LLMs) when parsing electronic spreadsheets. According to a paper published on Arxiv on July 12th, SpreadsheetLLM enables LLMs to "read" spreadsheet content through an encoding framework. This research is expected to significantly enhance the efficiency of data management and analysis in spreadsheets, allowing users to ask questions of AI in

Microsoft Researchers Introduce EmotionPrompt Method to Enhance Emotional Intelligence in Multilingual Models

Microsoft researchers have recently released the EmotionPrompt method, which aims to enhance the emotional intelligence of multilingual models. The research findings indicate that emotional stimuli can significantly improve the performance of multilingual models. The study also analyzes the impact of emotional stimuli on model outputs, showing that large language models perform better when processing emotional stimuli.

Microsoft Researchers Unveil LoRAShear Technology: Efficient Structured Pruning for Large Language Models

Microsoft researchers have introduced the LoRAShear technology, achieving efficient pruning and knowledge recovery for large language models. LoRAShear utilizes the LHSPG method to enable progressive structured pruning and dynamic knowledge recovery. This technology improves knowledge retention rates while maintaining high performance through model pruning.

Microsoft Research Enables Large Models to Forget Harry Potter: The Memory Erasure Technique of Llama 2

Microsoft researchers have successfully conducted studies on selective forgetting in large models, allowing models to forget specific information rather than the entire content through fine-tuning and reinforcement learning. The research method has certain limitations and requires further validation.