large model technology

MediaTek Research, a renowned research organization under MediaTek, has publicly announced that they have successfully developed and launched a new open-source Large Language Model (LLM), MR Breeze-7B.This model, with its powerful bilingual processing capabilities and data insights characteristics, is leading the artificial...

Recently, Ant Group announced the launch of its new 2 billion parameter multimodal remote sensing base model - SkySense, which has attracted wide attention in the industry. It is worth mentioning that SkySense's related papers have been included in CVPR 2024, the top international AI conference, and have achieved first place in all 17 tests...

Apple is pressing ahead with the integration of artificial intelligence (AI) elements in its products, with a revolutionary update coming to the Spotlight search feature, Bloomberg exclusively reports. The quick search tool, which is built into macOS and iOS, is expected to be augmented by AI technology in the future to enable more...

Google is making a wider range of Gemini big language model interfaces available to developers through its Vertex AI platform. According to TechCrunch, Gemini 1.0 Pro is now officially available after a public preview, while the higher-level Gemini 1.0 Ultra, though currently only available through a whitelist, has already raised industry concerns about its performance...

Recently, tech giant Google announced the launch of its new AI grand model - Gemini 1.5, a major breakthrough that once again pushes the boundaries of AI technology to new heights. As Google's latest masterpiece in the field of AI, Gemini 1.5 not only realizes an amazing improvement in performance, but also shows the functions and application scenarios...

Google recently released a music generation system called MusicRL, which significantly improves the quality of generated music by combining human feedback with reinforcement learning to make it more compatible with human tastes. This breakthrough technology is based on the pre-trained MusicLM model, which was originally capable of generating textual descriptions based on...

Google has recently released a new update for its chatbot Bard, which will support text-generated images and extended double-checking functionality. The new image generation feature added to Bard will be available free of charge in supported regions around the world, powered by the Google Imagen 2 model, but will require the use of English commands. Users will be able to follow the...

Researchers at Amazon warn of the need to be wary of data traps during the training process of large language models, Techradar reports. They point out that there is currently a large amount of content on the web that is generated by machine translation, and that this low-quality content can be a problem for the training process. The researchers found that a large number of web...

The Allen Institute for Artificial Intelligence (AI2) recently announced that it will open-source to the public its newly developed text-generating AI models, as well as the data used to train these models. This initiative aims to advance the field of artificial intelligence and promote communication and collaboration between academia and industry. It is reported that AI2 is open-sourcing this text generation...

Recently, Vivo released a list of the top 10 product technology innovations for the "2023 Technology Innovation", which signifies Vivo's strong strength in technology development and innovation. The technological innovations released cover a wide range of areas, including chip technology, imaging technology, battery life system, etc., to bring global consumers better quality...

Shanghai AI Lab has released a new generation of Shusheng-visual grand model, InternVL-6B. This model has a visual coder parameter count of 6 billion, and adopts the progressive alignment technique of contrast-generation fusion, which realizes the fine-grained alignment of visual grand model with linguistic grand model on Internet-level data. In addition to the on ...

A research team from Huawei and the University of Hong Kong recently released a new image generation model called CompAgent. The model is mainly used to solve the problem of combining text to image generation, which brings a new breakthrough in the field of artificial intelligence. The core idea of CompAgent is to adopt a divide-and-conquer approach to complex...

Meta recently published a paper describing Mosaic-SDF, a new 3D format customized for AI models and designed to accelerate the development of 3D generative models. Mosaic-SDF employs a small volume mesh with different centers and scales to approximate arbitrary symbolic distance functions. This design allows...

Tencent has recently released its video generation model VideoCrafter2, which has been dramatically improved in terms of lighting and shadow effects, etc. VideoCrafter2 can quickly generate high-quality videos of a few seconds based on text descriptions provided by the user. Compared to the previous version, the new model has a...

Google recently released a new large model, Alpha Geometry, which is specifically targeted at the field of mathematical geometry, and whose mathematical geometry capabilities have reached the level of human Olympic gold medalists. Notably, the model is trained on synthetic data rather than existing datasets, an innovative approach...

Recently, Baichuan Intelligence released a new character model, Baichuan-NPC, aiming to bring richer and more realistic character experience to the game industry. The model optimizes "character knowledge" and "dialogue capability", enabling the model to better understand the contextual semantics of dialogue, and carry out conversations and actions more in line with the character's personality....

With the rapid development of technology, artificial intelligence has permeated every aspect of our lives. Recently, Vivo announced that its new flagship phone S18 Pro will be officially on sale on January 13, which is equipped with the latest big model of artificial intelligence, bringing more convenience and surprise to our daily life. As ...

On January 2, MiracleVision (Chimera Intelligence), Meitu's self-developed AI visual grand model, formally passed the Interim Measures for the Administration of Generative Artificial Intelligence Services for the record and was opened to the public. This innovative technology has been continuously iterated since its introduction, and has now been upgraded to version 4.0, which is not only widely used in Meitu's...

A team of researchers from Tsinghua University and Harvard University recently jointly released LangSplat, the latest large language modeling system.According to the Arxiv page, this model, based on 3DGS's 3D language field approach and introducing SAM and CLIP technologies, performs well on open vocabulary 3D object localization and semantic segmentation tasks, not only outperforming...

Racer, in conjunction with Harbin Institute of Technology, recently open-sourced the KwaiAgents model, which achieves the effect of spanning the 7B/13B model and outperforms GPT-3.5.This open-source project injects new vitality into the entire community and provides a wealth of resources and references for researchers. The KaiAgents model consists of three parts...

Recently, a news about Google co-founder Sergey Brin's personal involvement in the development of Google's latest big model, Gemini, has attracted widespread attention. According to the report, Brin not only invested a lot of time and energy in the development process, and even personally wrote code for Gemini at critical moments, working weekly...

In the field of artificial intelligence, large models have become the cornerstone of many applications. However, with the wide range of applications, the illusion problem of large models has become more and more prominent. Recently, a research team from Tongji University and Fudan University has released a new method called "retrieval-enhanced generation (RAG)", which aims to solve this challenge...

In October of this year, Apple and a team of researchers from Columbia University jointly released a multimodal large language model called Ferret. Compared to traditional models, Ferret is unique in its ability to accurately recognize and describe the content of an image, as well as accurately identify and locate various elements in the image. Ferret...

In today's era of big models, many companies and technicians are keen to train their own models. In response, Baichuan Intelligence founder and CEO Wang Xiaochuan said in a speech on December 21 that both using models and creating models are very important. Wang Xiaochuan believes that there are many models in the current industry, and many enterprises and...

Recently, researchers from the Max Planck Institute for Intelligent Systems, ETH Zurich, and Technische Universität Darmstadt released an AI hairstyle generator, a tool that generates 3D hairstyle models from text, opening up unlimited possibilities in the field of fashion design. The name of this AI hairstyle generator...

The Danish University of Science and Technology recently announced their latest research - a big AI model called Life2vec that predicts when humans will die. This discovery has attracted widespread attention as this technology has the potential to change our understanding of life and death. The Life2vec model is based on a large amount of data trained...

Researchers at Adobe, the University of Massachusetts, Google, and the University of Toronto recently released a neural architecture called VecFusion that utilizes AI macromodels to design fonts. This innovative research opens up new possibilities in the field of font design. VecFusion is a cascading diffusion model that consists of a...

Baidu released a number of AI-native cloud products at the CloudSmart Conference 2023 - Smart Computing Conference, including the AI heterogeneous computing platform "Bage 3.0", Smart Computing network platform and self-developed cloud-native database GaiaDB 4.0, etc. These new products are fully optimized and upgraded in terms of AI compute, network and storage to support AI applications. These new products have been fully optimized and upgraded in terms of AI computing, network and storage for AI application...

Recently, Tsinghua University and Huawei's research team jointly released a new technology that can improve the limitation of big model text input through semantic compression technology. The introduction of this technology provides a broader scope for the application of big models in the field of text processing. The technique draws inspiration from source coding in information theory...

Recently, a team of researchers from several universities teamed up with AI technology company Aizip to say that it is now possible to make large language models self-replicating to some extent. This breakthrough will open up new possibilities in the field of AI and is expected to boost the development of small AI tools. According to Aizip CEO Yan Su...

Google DeepMind recently used large-scale language models to crack a long-standing problem in pure mathematics. In a paper published in the journal Nature, the researchers say it's the first time a large-scale language model has been used to discover a solution to a scientific puzzle, producing verifiable...

Recently, Google announced the launch of its newest Vincennes large model, Imagen 2, to provide users with higher-quality and more realistic image generation services. Imagen 2 was developed using technology from Google DeepMind and was previewed at the tech giant's I/O conference in May. Compared to the first generation of Imagen, Image...

Recently, NyunAI and Transmute AI Lab published a joint research paper on the Arxiv page, revealing a new approach to large model compression. This method is based on the parameterization of reduced-order modeling, which provides an effective solution for large model compression. The core of the method lies in the feature space into...

Microsoft Research recently released its new AI language model Phi-2, a small language model that demonstrates excellent reasoning and language understanding. With only 2.7 billion parameters, the Phi-2 model is small, but its performance in complex benchmarks rivals some larger models and even surpasses...