Author: advancedainews

Fine-tuning a pre-trained Text-to-Image (T2I) model on a tailored portrait dataset is the mainstream method for text-driven customization of portrait attributes. Due to Semantic Pollution during fine-tuning, existing methods struggle to maintain the original model’s behavior and achieve incremental learning while customizing target attributes. To address this issue, we propose SPF-Portrait, a pioneering work to purely understand customized semantics while eliminating semantic pollution in text-driven portrait customization. In our SPF-Portrait, we propose a dual-path pipeline that introduces the original model as a reference for the conventional fine-tuning path. Through contrastive learning, we ensure adaptation to target attributes and purposefully align…

Today, we’re excited to announce the availability of Llama 4 Scout and Maverick models in Amazon SageMaker JumpStart and coming soon in Amazon Bedrock. Llama 4 represents Meta’s most advanced multimodal models to date, featuring a mixture of experts (MoE) architecture and context window support up to 10 million tokens. With native multimodality and early fusion technology, Meta states that these new models demonstrate unprecedented performance across text and vision tasks while maintaining efficient compute requirements. With a dramatic increase on supported context length from 128K in Llama 3, Llama 4 is now suitable for multi-document summarization, parsing extensive user activity for…

Robert Legato – the respected VFX innovator who won Oscars for “Titanic,” “Hugo” and “The Jungle Book” – has joined Stability AI at chief pipeline architect. In doing so, he reteams with James Cameron, an influential board member for the AI firm. Legato is credited with creating the virtual cinematography pipeline for Cameron’s “Avatar,” which went on to surpass the director’s “Titanic” as the highest grossing film of all time with $2.8 billion at the worldwide box office. Legato’s pioneering work has also included “Apollo 13,” “The Aviator,” and the virtual production of “The Lion King.” He’s also held roles…

I don’t care who you are, where you’re fromWhat you did, As long as you love me! — Backstreet Boys Wow! They really love AI in China! Across regions, across pockets!  So, Tencent, from China, has announced its latest Hunyuan-T1—the first Mamba-powered ultra-large model! Well, well, well! Seems the Chinese are in love with AI! First, came DeepSeek, then Baidu ERNIE 4.5, and now, Tencent with Hunyuan-T1, with Google Gemma, in-between, along with OpenAI’s O-series models. That’s really a lot of development done within a space of very little time! Didn’t I remark earlier that AI models will come up thick…

In today’s industry news roundup: French AI specialist to identify new use cases and develop tailor-made models with shipping and logistics firm; Juniper Research expects telco AI investments to hit $22bn per year by 2029; African telco giant MTN is developing a streaming platform with Synamedia; and much more! Mistral AI, the Paris-based generative AI developer, has agreed a €100m multi-year contract with French shipping and logistics company CMA CGM to identify new AI use cases within the business and develop tailor-made models and agents. Sifted reports that as part of the five-year deal, Mistral AI will embed a dedicated…

Over the weekend, Meta dropped two new Llama 4 models: a smaller model named Scout, and Maverick, a mid-size model that the company claims can beat GPT-4o and Gemini 2.0 Flash “across a broad range of widely reported benchmarks.”Maverick quickly secured the number-two spot on LMArena, the AI benchmark site where humans compare outputs from different systems and vote on the best one. In Meta’s press release, the company highlighted Maverick’s ELO score of 1417, which placed it above OpenAI’s 4o and just under Gemini 2.5 Pro. (A higher ELO score means the model wins more often in the arena…

Image: Envato/DC_Studio Researchers from AI company DeepSeek and Tsinghua University have introduced a new technique to enhance “reasoning” in large language models (LLMs). Reasoning capabilities have emerged as a critical benchmark in the race to build top-performing generative AI systems. China and the U.S. are actively competing to develop the most powerful and practical models. According to a Stanford University report in April, China’s LLMs are rapidly closing the gap with their U.S. counterparts. In 2024, China produced 15 notable AI models compared to 40 in the U.S., but it leads in patents and academic publications. What is DeepSeek’s new…

The past week was a whirlwind of AI-related news, with tech giants Microsoft Corp. MSFT, Alibaba Group Holding BABA, and OpenAI making significant strides in the field. From AI roasting tech leaders to the unveiling of new AI models, the week was filled with exciting developments. Let’s dive into the top stories. Microsoft’s AI Copilot Roasts Tech Leaders In a humorous twist during Microsoft’s 50th-anniversary celebrations, the company’s AI assistant, Copilot, took the opportunity to roast tech leaders Bill Gates, Satya Nadella, and Steve Ballmer. The event was part of a special interview designed to celebrate Microsoft’s legacy. Read the full article here. Alibaba…

SUZHOU, China, March 31, 2025 /PRNewswire/ — On 26 March, Suzhou Pudu Co-Intelligence Technology Company, a joint venture between Xi’an Jiaotong-Liverpool University (XJTLU) and Baidu Group, was launched as China’s first AI-focused joint venture co-founded by Baidu and a university. AI+education: Redefining learning Through AI-driven innovation, Pudu Co-Intelligence seeks to transform the whole education value chain, empower industrial evolution and cultivate localised service ecosystems. The company will soon launch the Pudu Co-Intelligence AI Forum, bringing together Baidu’s chief scientists, XJTLU’s AI researchers, policymakers, and industry leaders to share their thinking on cutting-edge technologies, AI research, interpretation of relevant policies, and personal experience…

This Verizon showcase suggests that after years of limited tangible, scalable implementations, private 5G is finally moving from pilot to production As broadcasters face mounting pressure to manage dozens of camera feeds, navigate spotty connectivity and capture every critical moment during live events, a new solution from Verizon Business and NVIDIA is turning heads at NAB Show 2025. The telco debuted a portable private 5G network framework, built in close collaboration with NVIDIA, that it said leveraged AI and high-performance connectivity to reimagine live broadcast workflows. At the heart of this mobile, environmentally controlled setup is NVIDIA accelerated computing, including…