Mistral releases Codestral 25.01, an updated open-source coding framework with double the speed, optimized for low-latency operations and excelling in Python coding tests. Available via IDE plugins and APIs, it supports large data handling and outperforms competitors like Codellama and DeepSeek Coder. Mistral's specialized model gains rapid attention, reflecting growing demand for expert-focused programming tools.

Adobe launches Firefly Bulk Create tool enabling batch editing of tens of thousands of images with focus on resizing and background changes. Supports PNG, JPEG, future PSD support planned. Uses generative AI for seamless background adjustments. Paid service likely for premium members.

Illustrious, a Stable Diffusion XL-based text-to-image model, excels with Danbooru tag system for precise image generation. Trained on anime images, it efficiently understands visual concepts without complex captions. The model supports high-resolution outputs and is optimized for generating high-quality anime-style images. Key versions include v0.1, v1.0, and v2.0, each enhancing performance and resolution. Installation involves placing the checkpoint in the appropriate folder, avoiding natural

AI is advancing to perform mid-level software engineer tasks, reshaping workflows and productivity. Human-machine collaboration will redefine roles, focusing engineers on strategic, creative tasks while AI handles repetitive processes. This shift may reduce demand for routine roles but creates opportunities in supervision, entrepreneurship, and specialized areas, fostering innovation and lowering barriers to entry. Ensuring quality and reliability of AI-generated code remains crucial, emphasizin

Qwen Chat is a free web-based AI assistant offering code generation, model comparison, and contextual document integration. It supports open-source and proprietary models, enhancing productivity with features like artifact generation and customizable interfaces. Planned updates include web search, image generation, and voice mode, aiming to improve user experience across various skill levels. Built on an open Web UI framework, it prioritizes accessibility and versatility for developers and resea

Musk gains support from government agencies in legal battle against OpenAI. FTC and DOJ emphasize principles suggesting potential anticompetitive behavior by OpenAI and Microsoft. Musk, co-founder of OpenAI, sues to halt company's IPO, accusing it and Microsoft of stifling AI competition. OpenAI denies claims, but regulators are investigating partnerships and potential violations. Musk alleges OpenAI coerced investors and shared board members with Microsoft, despite denials. FTC and DOJ find mer

US Department of Justice initiates antitrust case against Apple and Google over a $20 billion search engine deal. The case scrutinizes revenue-sharing agreements impacting market competition and consumer choice. Apple defends the pact as crucial for its ecosystem, while Google emphasizes the importance of such deals for service quality. The outcome may reshape regulatory approaches and competitive dynamics in the tech sector, potentially enhancing innovation and consumer benefits.

Zuckerberg discusses AI reshaping software engineering by 2025, with AI handling mid-level tasks. Initially costly, AI will evolve to integrate into standard processes, enhancing efficiency. Human engineers will focus on creative and complex problem-solving, complementing AI's capabilities. Unique human traits like creativity and intuition will remain irreplaceable.

YY Live launched AI companion broadcaster "Lingyun", serving in over 6000 live rooms and attracting over one million daily users. Since its release, user interactions increased by 30%, with some broadcasters seeing a 670% rise in interaction devices and an 80% increase in paying users. YY plans to expand Lingyun's capabilities in 2025, including free conversations and AI-created content. Despite past acquisition challenges, YY Live remains a leader with 41 million monthly active users.

China Cybersecurity Association launches a platform offering 27 datasets totaling 2.7TB of Chinese internet corpus resources for AI development. Developed with guidance from Cyberspace Administration and collaboration with National Internet Emergency Center, the platform provides categorized resources facilitating user-specific downloads. This initiative strengthens AI innovation and industrial growth.

DeepSeek has quietly launched a mobile app on iOS and Android platforms, accessible via specific links. Users should verify developer information to avoid counterfeit versions. The app supports online search and file uploads but currently lacks multimodal capabilities, using OCR for image parsing. This move enhances user convenience and expands DeepSeek's product offerings. Future updates may introduce advanced features.

Cohere launches North, an AI workspace platform outperforming competitors in finance, HR, and IT. It integrates large language models, search, and automation securely, ideal for regulated sectors. Early tests show over 80% faster task completion. North emphasizes data quality and synthetic data over computational scale, enabling non-technical users to build custom AI tools. Partnerships with major enterprises like Royal Bank of Canada demonstrate its potential to redefine enterprise AI implement

OpenAI is recruiting for its first robotics team to develop general-purpose, adaptive robots with human-like intelligence for dynamic environments. The goal is large-scale production exceeding one million units. OpenAI plans to integrate advanced AI models with custom sensors and computing hardware, focusing on versatile robots for various tasks. Job listings suggest potential humanoid designs and mass production efforts. OpenAI also invests in robotics startups and develops custom chips for AI

Adobe and HKUST developed TransPixar, a technology generating RGBA videos with natural transparency for seamless VFX like smoke and explosions. It uses a diffusion transformer model and LoRA-based adaptation ensuring consistent RGB and alpha channels. TransPixar optimizes attention mechanisms to maintain quality despite limited data, enabling dynamic scenes from text prompts and offering applications in VR, AR, and education. Available as open-source, it promises cost-effective VFX creation.

UC Berkeley's NovaSky team unveiled Sky-T1-32B-Preview, an open-source inference model demonstrating competitive performance against early OpenAI o1 models. Training cost less than $450 using synthetic data. Sky-T1 excelled in MATH500 and LiveCodeBench but lagged in GPQA-Diamond. The team aims to develop more efficient models with strong inference capabilities.

UK PM Starmer unveils ambitious AI action plan to boost economy and transform public services with 50 initiatives including AI growth zones and a new supercomputer. The plan aims to generate £47 billion annually and attract significant investment while addressing regulatory challenges.

Plaintiffs allege Meta used pirated content from LibGen to train its Llama AI model with Zuckerberg's knowledge and attempted to cover up infringement by removing attribution. Despite dismissal of initial lawsuit, new evidence suggests internal awareness of copyright violations. Meta denies illegal use but faces ongoing legal challenges and policy criticism.

Elon Musk discussed AI's data depletion issue during a live broadcast, suggesting synthetic data as a solution. Major tech companies like Microsoft and Google are already using synthetic data to enhance AI model training, reducing costs significantly but facing challenges such as potential bias and reduced creativity. By 2024, 60% of AI project data may be synthetically generated, according to Gartner.

Musk's attorney urges California and Delaware to force OpenAI to auction its shares, escalating tensions with CEO Altman. Musk accuses OpenAI of deviating from its mission, while OpenAI claims transitioning to a PBC will maximize value for early supporters. The nonprofit's stake could be worth billions, and an auction could give external investors significant influence. OpenAI aims to become a PBC within two years, valued at $157 billion.

Midjourney V7 in final testing with new architecture and methods. V6 improved over V5.1, offering better results and additional features like image editing and style customization. The team is also exploring video generation while focusing on user feedback and enhancing creative tools. Development includes mood boards integration and two new generation modes for varied output speeds. Challenges lie in making video creation user-friendly and cost-effective.