Search AI Products and News

A research team from Fudan University has developed the 'Concept Erasure Technique' (RECE), which can efficiently and accurately modify inappropriate ideas in AI models, significantly reducing the likelihood of generating inappropriate images while maintaining the AI's other creative capabilities. This technology uses a 'closed solution' to locate and modify specific parts, ensuring the AI behaves more appropriately during creation and stays on track. Experiments show that the RECE technology can effectively enhance the morality and quality of AI creations, providing new directions for the future development of AI, making future AI assistants smarter and more perceptive.

IncarnaMind is an AI model specifically designed for interactive chatting with multiple documents and precise information querying. It employs an innovative sliding window chunking technique that dynamically adjusts information retrieval methods based on document content and user needs, supporting multi-hop queries while processing several documents, breaking the traditional limitation of querying one document at a time. By balancing comprehensive contextual information with fine details, IncarnaMind enables more efficient and accurate information retrieval. It is compatible with various large language models, including the GPT series.

The groundbreaking Transformer acceleration technology FlashAttention-3 has been launched! This is not just an upgrade; it heralds a direct increase in the inference speed of our Large Language Models (LLMs) and a direct decrease in costs! Let's talk about FlashAttention-3 first, which is a significant improvement over its predecessors: Significant increase in GPU utilization: Training and running large language models with FlashAttention-3 doubles the speed, up to 1.5 to 2 times faster, and thi

A recent study suggests that through special training, language models can partially achieve more efficient multi-step reasoning capabilities. This ability is similar to the "System 2 reasoning" described by psychologist Daniel Kahneman, which is a slow and conscious way of processing information. Researchers at Meta have developed a new method that "refines" the computationally intensive multi-step reasoning process into parameters of a language model. The results show that in some cases, model