At today's Rokid Jungle 2024 partner conference and product launch, Rokid officially unveiled its next-generation AR glasses - Rokid Glasses. These glasses not only support customized myopic and astigmatism lenses based on user needs, but also feature a convenient detachable buckle design, providing more personalized options for users with different vision requirements. The biggest highlight of Rokid Glasses is its integration with Alibaba's Tongyi Qianwen multimodal large model, enabling phone calls, Q&A searches, and object recognition.

Stability AI has once again broken through technical barriers by launching the new Stable Diffusion 3.5 Medium model. This AI art tool aimed at the general public is not only completely free for commercial use but also achieves a perfect balance between high performance and accessibility. With its multimodal diffusion transformer (MMDiT-X) architecture and a streamlined design of 2.5 billion parameters, this model cleverly addresses the hardware threshold for ordinary users, requiring only 9.9GB of VRAM to run smoothly on most consumer-grade graphics cards.

Today in Chengdu, a significant breakthrough in technological innovation has been achieved! According to the official Chengdu WeChat account, the Chengdu Humanoid Robot Innovation Center has unveiled the country's first ultra-lightweight humanoid robot - 'Gongga No. 1'. This robot is not only lightweight and agile but also equipped with powerful intelligence, making it a proud 'Chengdu creation'. 'Gongga No. 1' is one of only three humanoid robots globally to possess the 'strongest brain', with a total weight of just 25 kilograms, well below the industry standard of 60 to 250 kilograms.

French AI startup Les Ministraux has launched two brand new lightweight models, Ministral3B and Ministral8B, specifically designed for edge devices, with parameters of 3 billion and 8 billion, respectively. Both models excelled in instruction-following benchmark tests, with Ministral3B surpassing Llama38B and Mistral7B, while Ministral8B outperformed both models in all aspects except for coding capabilities.

Meta's Chief Technology Officer Andrew Bosworth recently confirmed an important decision to The Verge: the company has canceled the high-end headset Quest Pro2 (codename La Jolla), originally scheduled for release in 2027, and is now developing a new ultra-lightweight headset. This news marks a significant change in Meta's hardware strategy and reveals a new direction for the company's VR technology development. According to Bosworth, the main reason for canceling Quest Pro2 was primarily...

Google recently announced the beta availability of the Gemini Nano lightweight AI model for Android developers. This model is designed for lightweight tasks on-device, suitable for low-performance smartphones and tablets. Developers can integrate this model into their applications using Google's AI Edge SDK. The Gemini Nano model is part of the Google Gemini model family, offering three main advantages: ensuring privacy, operation without internet connectivity, and no subscription required.

Microsoft has released Phi-3.5-vision, a lightweight, multimodal open source AI model designed for processing textual and visual inputs, supporting a context length of 128K. This model is suitable for resource-constrained environments and features capabilities such as image understanding, OCR, chart parsing, and multi-image summarization, showcasing excellent performance and low latency. Comprised of 4.2 billion parameters, it is trained with high-quality data to ensure performance and privacy. It includes three models: lightweight AI, expert mix, and multimodal model, all demonstrating outstanding performance in image and video processing benchmarks.

The Baidu Xirang lightweight Metaverse SaaS platform was officially launched on August 9. This service is designed specifically to accelerate the digital transformation of enterprises. The platform integrates Baidu Xirang's technological expertise in areas such as the Metaverse, artificial intelligence, cloud computing, and mobile applications, providing lightweight scenario subscription capabilities. By freely combining spaces and components, it enables the construction of customized marketing, exhibition media, and cultural education scenarios, helping enterprises reduce costs and improve efficiency.

On July 4, 2024, the Shanghai Artificial Intelligence Laboratory launched the new version of the Shusheng·Puyang series model, InternLM2.5, at the WAIC Scientific Frontier Main Forum. This version has significantly enhanced reasoning capabilities in complex scenarios, supports 1M ultra-long context, and can autonomously perform internet searches and integrate information from hundreds of webpages.

CatVTON is a lightweight AI virtual dressing model designed for fashion enthusiasts, with a total parameter count of 899.06M and only 49.57M trainable parameters needed during training. It requires less than 8G of GPU memory for inference and supports 1024x768 high resolution, making it suitable for personal computers. The model can be quickly deployed on ComfyUI and Gradio applications, supporting inference on VITON-HD and DressCode datasets wit....

In the increasingly fierce competition in the field of artificial intelligence, OpenAI once again leads the trend by introducing GPT-4o mini, a lightweight large language model. This move not only marks a new trend in the miniaturization of AI large models but also embodies OpenAI's strategy of making advanced AI technology "as widely accessible as possible." Upon its launch, GPT-4o mini demonstrated astonishing performance, scoring 82% on the MMLU benchmark test, surpassing its predecessor, GPT

Google has launched Gemma2, the latest version of its open-source lightweight language model, offering parameter sizes of 9 billion (9B) and 27 billion (27B). Compared to its predecessor, the Gemma model, this new version promises enhanced performance and faster inference speed.Product Access: https://top.aibase.com/tool/google-gemma-2 Gemma2 originates from Google's Gemini model and aims to make it easier for researchers and developers to access, significantly improving speed and efficiency. Un

The model is small enough to run on smartphones and fast enough to render frames in real-time based on text prompts. He noted that while the emergence of generative AI image diffusion models is exciting, these models need a significant speed increase to have a major impact on augmented reality. Therefore, Snap's team has been dedicated to accelerating machine learning models.

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.

The 15-inch M3 MacBook Air has appeared on the Geekbench 6 benchmarking site with a single-core score of 3157 and a multi-core score of 12020, showing a significant improvement over the M2. The M3 features an 8-core CPU and a 10-core GPU, with a CPU clock speed of 4.05GHz, and offers memory options of 8GB, 16GB, and 24GB, along with a maximum storage capacity of 2TB. It comes pre-installed with macOS Sonoma and is available in multiple color options.

LG has released a new model of the LG gram series laptops with an ultra-lightweight design. The new laptop features AI functionalities available through the LG gram Link application, powered by Intel Core Ultra processors and NVIDIA GeForce RTX 3050 graphics cards. The LG gram Link application enables file and photo sharing, as well as screen mirroring. The LG Pro series comes equipped with a dual cooling system, high-brightness IPS panel, and four times ultra-high definition resolution.

Google has released the PaLI-3 visual language model, which is lightweight yet achieves SOTA performance. This model employs a contrastive pre-training method and deeply explores the potential of VIT, demonstrating outstanding performance in multilingual modality retrieval. PaLI-3 perfectly integrates natural language understanding and image recognition, becoming an important force in AI innovation. The model's SigLIP-based contrastive pre-training method opens up a new era for multilingual cross-modal retrieval. Although it has not been fully open-sourced, multilingual and English SigLIP models have been released, providing opportunities for researchers to experiment.

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.