China’s Manus AI has formed a strategic partnership with Alibaba’s Qwen team to boost its new “general AI agent,” which it claims outperforms OpenAI’s DeepResearch. Unlike chatbots, Manus AI’s agent can execute tasks independently. The deal leverages Qwen’s open-source models, potentially addressing Manus AI’s growing user demand and giving Alibaba an edge against competitors like DeepSeek. Manus AI launched last week and has gone viral on Chinese social media, offering free task completions for select users. The AI agent remains invite-only, with a site beset by heavy traffic and technical issues. (Reuters)
1 Like
Meta has begun testing its first in-house AI training chip as part of an effort to reduce its reliance on suppliers like Nvidia. The new chip, designed specifically for AI tasks, is intended to cut costs for Meta’s expanding AI infrastructure, which is projected to consume up to $65 billion in capital expenditures. If tests succeed, Meta will scale up the chip’s production significantly. The move represents a strategic shift toward internal chip development, initially targeting recommendation systems, and eventually broader generative AI applications by 2026. Previously, Meta relied heavily on Nvidia GPUs for its massive AI workloads. (Reuters)
1 Like
Sam Altman announced that OpenAI has developed a new AI model specifically designed for creative writing, noting it’s the “first time I’ve been genuinely impressed” by AI’s creative capabilities. (Twitter)
2 Likes
Google released Gemma 3, a versatile AI model optimized for efficiency, capable of running on anything from high-end GPUs to smartphones. Built with a massive 128K context window, Gemma 3 comes in four sizes—1B, 3B, 12B, and 128K context windows—targeted at developers needing scalable AI solutions. Gemma 3 outperforms earlier Google models in tasks like math, coding, and complex instructions, running efficiently even on single-GPU setups. While users can freely download it from platforms like Kaggle and Hugging Face, Google’s licensing restricts commercial use. Gemma 3 also features an integrated image-safety tool called ShieldGemma 2 to filter harmful content. (ArsTechnica)
1 Like
LibreNews: FOSS infrastructure is under attack by AI companies
According to this news article, LLM scraper bots are attacking open-source infrastructure sites like SourceHut, GitLab, GNOME, Fedora, Diaspora, and ReadTheDocs. Diaspora says 24.6% of their web traffic over the last 60 days is coming from GPTBot, with Amazonbot and Claudebot taking another 14.9% and 4.3% respectively.
GPTBot is how OpenAI identifies its training crawler. They claim they respect robots.txt. Either they are lying, or third parties are falsely identifying themselves as GPTBot. Either way, it’s angering a lot of open-source developers and polarizing them against AI.