Beijing ZhiYuan Artificial Intelligence Research Institute, in collaboration with Renmin University of China's Gaoqing School of Artificial Intelligence, has launched an innovative AI model framework called MemoRAG. This framework, based on long-term memory, aims to advance Retrieval-Augmented Generation (RAG) technology to handle more complex tasks beyond simple question answering.

Recently, the Electronics and Telecommunications Research Institute (ETRI) in South Korea announced that they have successfully developed a technology named 'Dejaview'. This new technology utilizes real-time surveillance cameras (CCTV) and Artificial Intelligence (AI) to not only detect criminal activities but also predict the likelihood of crimes, marking a shift in the role of surveillance systems from mere monitoring to actively preventing crime. Image Source Note: Image generated by AI, image authorized by service provider Midjourney. The core of 'Dejaview' technology lies in analyzing...

The Institute of Geochemistry of the Chinese Academy of Sciences, in collaboration with Aliyun, launched the world's first "Lunar Science Multi-modal Professional Large Model" at the 2024 Digital Expo. This model is based on Aliyun's Tongyi series and utilizes RAG retrieval-augmented technology, fine-tuned and trained on Aliyun's Baolian exclusive version, specifically aimed at improving the accuracy of distinguishing the ages and shapes of lunar impact craters, currently exceeding 80%. This innovative model combines visual, multi-modal, and natural language processing technologies, and its optimal application is in identifying impact craters on the moon, which is significant for studying lunar geological evolution.

The research team led by Yuan Li at Peking University, in collaboration with researchers from Nanyang Technological University and Tsinghua University, has developed an innovative image retrieval technology that aims to precisely match user needs through sketches, artworks, and even low-resolution images. The core of this technology is the 'Universal Style Retrieval' method, which can handle multiple query styles including combinations of text, sketches, and artworks, significantly enhancing retrieval flexibility and accuracy. The research team constructed the DSR and ImageNet-X datasets to provide training and testing resources for the new method.

The Zhiyuan Research Institute recently released three new vector models that excel in vector retrieval tasks and have set new best standards across multiple evaluation benchmarks. The three models are: BGE-EN-ICL: an English vector model that enhances the model's semantic expression capabilities by introducing task-related query-document examples as few-shot examples.

In the recently concluded International Mathematical Olympiad, Google's DeepMind AI systems, AlphaProof and AlphaGeometry2, showcased remarkable problem-solving abilities, finishing just one point shy of the championship while surpassing human competitors in speed and accuracy. The AI achieved perfect scores on four problems and solved one in an astonishing 19 seconds, adhering to mathematical competition standards with rigorous clarity. This com....

Nvidia's ChatQA2 model, optimized from Llama3, addresses key challenges in AI long-text context understanding and retrieval-augmented generation by extending the context window to 128K tokens and employing a three-phase instruction tuning process. It significantly enhances instruction adherence, RAG performance, and long-text comprehension. In the InfiniteBench evaluation, ChatQA2 demonstrated accuracy comparable to GPT-4-Turbo-2024-0409 and surp....

Recently, CNKI launched the mobile version of its AI Academic Research Assistant, aimed at providing researchers with more convenient academic support. This AI assistant, after receiving widespread acclaim upon its launch on the PC platform, is now available through the CNKI mobile app, catering to users needs for on-the-go access.The main features provided by the AI Academic Research Assistant include:Enhanced Question-Answering Retrieval: Users can ask questions in natural language, and the AI

On July 11th, Kuaikan Comics announced that they are exploring fine-tuning of open-source large models to train a vertical large model for the two-dimensional (2D) domain. Kuaikan Comics utilizes large language models (LLM) and retrieval-augmented generation (RAG) technology to improve the conversion rate of work searches and inject new vitality into the search process. They have adopted a "fine-tuned large model + RAG enhancement" strategy, building internal PGC (Professional Generated Content)

In the world of AI, making machines understand videos is much harder than understanding images. Videos are dynamic, with sound, movement, and a myriad of complex scenes. In the past, AI viewed videos as if they were reading ancient scrolls, often leaving them baffled. But the introduction of VideoPrism might change everything. It's a video encoder developed by Google's research team that can achieve state-of-the-art levels on a variety of video understanding tasks with a single model. Whether it

Recently, two Chinese scholars from the Georgia Institute of Technology and NVIDIA have proposed a new fine-tuning framework called RankRAG. This framework greatly simplifies the complex RAG pipeline, enabling the same Large Language Model (LLM) to perform retrieval, ranking, and generation tasks, resulting in a significant performance improvement.RAG (Retrieval-Augmented Generation) is a commonly used technique in LLM deployment, especially suitable for text generation tasks that require a larg

New York-based Hebbia announced a $130 million B-round financing, with investors including Andreessen Horowitz, Index Ventures, Peter Thiel, and Google's venture capital department. Hebbia is building something quite simple: a localized productivity interface for LLMs, making it easier to extract value from data, regardless of its type or size. The company has already collaborated with some large companies in the financial services industry, including hedge funds and investment banks, and plans

The AI team from Peking University and Kuaishou has successfully tackled the challenge of complex video generation. They have proposed a new framework called VideoTetris, which can easily piece together various details like a puzzle to generate high-difficulty, complex-instruction videos. This framework surpasses commercial models such as Pika and Gen-2 in complex video generation tasks.

Mita AI Search offers a pure search experience free from advertising interruptions, which is rare in today's environment. Mita AI Search can quickly retrieve and consolidate information, significantly simplifying the process of information collection. It provides detailed information sources that help verify the authenticity of information. The user-friendly design includes features like a simple button layout and one-click sharing. Overall, Mita AI Search is a reliable and efficient information retrieval assistant.

Galileo has recently launched a new tool aimed at helping enterprises develop trustworthy artificial intelligence solutions. The new tool, named Retrieval-Enhanced Generation and Agent Analysis Solution, tackles complex RAG systems. The tool incorporates advanced insights and metrics, providing visibility into every stage of the RAG workflow, enabling rapid assessment and error detection. Galileo's tool transforms the AI development process, making AI applications more efficient and reliable, aligning with the future commercialization trends of AI.

Galileo Lab launches new metrics to help improve the accuracy of generative AI. OpenAI GPT-4 has a 23% tendency for hallucination in basic question-answering tasks. The new metrics from Galileo Labs include correctness and contextual consistency. These new tools can help businesses test and mitigate hallucination issues in AI implementations.

Digital forgery attacks have surged, increasing fivefold from 2021 to 2023, reaching 18%. Video fraud is on the rise, accounting for 80% of biometric defense. Global fraudsters are targeting national ID cards, which make up 46.8% of forged documents. 79% of attacks on financial service institutions are initiated using simple, scalable models.

RA-DIT improves the ability of language models to utilize retrieval information and provide relevant results from the retriever through a two-stage fine-tuning process. RA-DIT outperforms existing retrieval-augmented models in knowledge-intensive zero-shot and few-shot testing. The RA-DIT65B model shows improved performance in tasks that require knowledge, demonstrating the effectiveness of fine-tuning. The enhancement of retrieval capabilities brought by RA-DIT has shown good results in knowledge-intensive evaluations, confirming the effectiveness of fine-tuning.

InternLM-123B has shown outstanding performance in recent competitions among top large models, ranking first in 12 metrics and surpassing some performance levels of GPT-4. SenseTime's large model has iterated rapidly, upgrading from InternLM-104B to InternLM-123B in just six months, with strong data processing capabilities. SenseTime possesses a robust SenseCore computational foundation and collaborates with leading research institutions, boasting strong technical expertise.

Benshi Writing and Xueersi have reached a settlement regarding the incident of stolen large model data, with Benshi Writing deciding not to pursue legal action against Xueersi. After thorough investigation and communication, both parties cleared up misunderstandings and will continue their collaboration. This incident marks the first case of stolen large model data, significantly impacting data security issues in the education sector.