Search AI Products and News

AI company Anthropic recently updated its "responsible scaling" policy, outlining which models require additional safety protections. This move aims to mitigate potential risks before new technologies are released. According to Anthropic's blog, if stress testing reveals an AI model could assist a "resource-constrained state actor" in developing chemical and biological weapons, Anthropic will delay the technology's public launch.

At a recent forum held in Davos, Deloitte released a report revealing the difficulties companies face in translating generative artificial intelligence (Gen AI) experiments into practical applications. Although companies have achieved positive initial results in generative AI projects, there are still many obstacles to scaling deployment. Source of the image note: Image generated by AI, image licensed by service provider Midjourney. According to the report, many companies find that the transformation process from the lab to production environments is not easy, especially.

A research team from New York University, MIT, and Google has recently proposed an innovative framework aimed at addressing the bottleneck of inference time scaling in diffusion models. This groundbreaking research transcends the traditional method of simply increasing the denoising steps, opening up new avenues for improving generative model performance. The framework primarily unfolds from two dimensions: one is leveraging feedback from validators, and the second is implementing algorithms to discover better noise candidates. The research team builds on a pre-trained SiT-XL model with a resolution of 256×256, maintaining 250 fixed denoising steps.

In today's rapid development of Large Language Models (LLM), the training and inference costs of models have increasingly become a focus of research and application. Recently, the Tencent Hunyuan team released an important study that delves into the 'Scaling Laws' of low-bit floating point quantization training, which refers to the principles governing the scale of floating point quantization training. The core of this research lies in exploring how to significantly reduce computational and storage costs without sacrificing performance by lowering the precision of the model.

Meta scientist Thomas Scialom reveals the secrets behind the development of Llama 3.1, whose 405B parameter scale was designed to compete with GPT-4. By increasing the number of training tokens rather than changing the architecture, Llama 3.1 achieves an optimized balance between model scale and the total amount of training data, resulting in significant leaps in knowledge depth and breadth. In terms of data selection, Scialom prefers synthetic data over publicly available internet text. The evaluation and improvement of Llama 3.1 utilize reward models and diverse benchmarks.

July 22, 2024, ChinaZ.com: Xiaomi Corporation recently announced that its artificial intelligence assistant, XiaoAI, has added a new feature called "AI Image Editing," which will be released with version 6.126 of XiaoAI. Users can perform various editing operations on images, including background transformation, style transformation, pedestrian elimination, intelligent expansion, and image Q&A.Image Background Transformation: Users can send an image to XiaoAI and choose to replace the background

OpenAI and DeepMind have differing views and approaches in the research of Scaling Laws. Scaling Laws can predict changes in loss when varying model parameters, data amounts, and computation resources. This competition will drive the development of artificial intelligence and shape the future of human-machine coexistence. Pre-training large language models involves trade-offs between model size, data quantity, and training costs. Scaling Laws can assist in optimizing design decisions. DeepMind argues that model size and data quantity should scale equally.

Recently, AI upscaling has gained significant attention on various social media platforms, with the topic's popularity continuing to rise. Tools like Jianying, Meitu Xiu Xiu, Adobe Photoshop, and Stability AI have been the focus. The AI upscaling features have limitations in imagination and detail depiction. Jianying's intelligent upscaling results are relatively standard, while Meitu Xiu Xiu performs better with landscape photos.