Author: advancedainews

Chinese tech giants like Alibaba (BABA), Tencent (TCEHY), and ByteDance have placed over $16 billion in orders for Nvidia’s (NVDA) H20 AI chips during the first quarter of 2025, according to The Information. The rush in demand comes amid concerns that the U.S. may soon ban the sale of these chips to China. While Nvidia has not commented, these orders could significantly boost revenue if the company manages to deliver the chips before any restrictions take effect.Don’t Miss Our End of Quarter Offers: The H20 is currently allowed in China, but it’s less powerful than Nvidia’s Blackwell chips, which are…

We present the first mechanistic evidence that model-free reinforcement learning agents can learn to plan. This is achieved by applying a methodology based on concept-based interpretability to a model-free agent in Sokoban — a commonly used benchmark for studying planning. Specifically, we demonstrate that DRC, a generic model-free agent introduced by Guez et al. (2019), uses learned concept representations to internally formulate plans that both predict the long-term effects of actions on the environment and influence action selection. Our methodology involves: (1) probing for planning-relevant concepts, (2) investigating plan formation within the agent’s representations, and (3) verifying that discovered plans…

Organizations deploying generative AI applications need robust ways to evaluate their performance and reliability. When we launched LLM-as-a-judge (LLMaJ) and Retrieval Augmented Generation (RAG) evaluation capabilities in public preview at AWS re:Invent 2024, customers used them to assess their foundation models (FMs) and generative AI applications, but asked for more flexibility beyond Amazon Bedrock models and knowledge bases. Today, we’re excited to announce the general availability of these evaluation features in Amazon Bedrock Evaluations, along with significant enhancements that make them fully environment-agnostic. The general availability release introduces “bring your own inference responses” capabilities for both RAG evaluation and model…

Stability AI speaks of a “multi-view video generation with 3D camera control”. This refers to an AI model that can convert photos into 3D models. These in turn can be viewed from all sides, enabling an immersive view. Stable Virtual Camera is still a research preview. Stability AI uses a diffusion model to create the 3D videos. These are AI models in which images are generated point by point using noise. A single photo or up to 32 images can be used as input. The generated videos are available with different camera paths, such as “dynamic”, “spiral”, “dolly zoom”, “pan”…

If you are interested in building your very own local deep research AI assistant, you might be interested in Google’s Gemma 3 AI models. They represent a significant advancement in artificial intelligence, offering a compact yet robust solution tailored for local deployment. Derived from the larger Gemini series, these models combine high performance with a strong emphasis on privacy and accessibility. Featuring multimodal capabilities, support for 140 languages, and the ability to generate structured outputs, Gemma 3 is engineered to meet diverse AI research assistant and productivity demands. These models are open source and optimized for local use, allowing you…

Midjourney has announced the alpha release of its V7 image generation model for testing by the AI community. The new model packs improvements in text prompt understanding, image quality, and feature coherence. “V7 is an amazing model. It’s much smarter with text prompts, image prompts look fantastic, image quality is noticeably higher with beautiful textures, and bodies, hands, and objects of all kinds have significantly better coherence on all details,” Midjourney explained. A key innovation in V7 is the default activation of model personalisation. Users must initially unlock this feature, a process that takes approximately five minutes. This personalisation can…

Mistral AI Introduces AI-Powered OCR By John K. Waters03/13/25 French AI startup Mistral AI has launched Mistral OCR, an advanced optical character recognition (OCR) API designed to convert printed and scanned documents into digital files with “unprecedented accuracy.” With a focus on multilingual support and complex document structures, Mistral OCR aims to outperform existing solutions from Microsoft and Google, the company said. Millions of printed documents and uneditable PDFs remain locked in archives, legal records, and historical repositories, the company noted in a blog post. And while traditional OCR software is proficient in extracting plain text, it often struggles…

Deep down, Sam Altman and François Chollet share the same dream. They want to build AI models that achieve “artificial general intelligence,” or AGI—matching or exceeding the capabilities of the human mind. The difference between these two men is that Altman has suggested that his company, OpenAI, has practically built the technology already. Chollet, a French computer scientist and one of the industry’s sharpest skeptics, has said that notion is “absolutely clown shoes.”When I spoke with him earlier this year, Chollet told me that AI companies have long been “intellectually lazy” in suggesting that their machines are on the path…

With the latest stable release dated January 28, 2025, Qwen2.5-Max is classified as a Mixture-of-Experts (MoE) language model developed by Alibaba. Like other language models, Qwen2.5-Max is capable of generating text, understanding different languages, and performing advanced logic. According to recent benchmarks, it is also more secure than DeepSeek-V3-0324. Using Recon to scan for vulnerabilities A team of analysts with Protect AI, the company behind a red teaming and security vulnerability scanning tool known as Recon, recently used their platform to compare the security of Qwen2.5-Max against that of DeepSeek-V3. The team’s assessment reads, in part: “We observed that DeepSeek-V3-0324…

Fifty years after the founding of Microsoft, the CEO of its artificial intelligence division has a big task: develop a new product line as integral to daily life as the software giant’s past innovations.“We’re really trying to land this idea that everybody is going to have their own personalized AI companion,” said Mustafa Suleyman in an interview with The Associated Press. “It will, over time, have its own name, its own style. It will adapt to you. It may also have its own visual appearance and expressions.”Suleyman laid out that vision on Microsoft’s 50th anniversary Friday. The celebration at Microsoft’s…