This week’s AI news featured major model releases from Google (Nano Banana 2, Gemini 3.1 Pro), OpenAI (GPT-5.4 Pro), and Perplexity, along with Microsoft’s new Copilot Tasks agent, while the sector saw drama over AI ethics, user backlash, and company controversies. In robotics, advances like Stanford’s FSME and new memory systems are enabling more capable humanoid robots, with companies like BMW and Faraday Future beginning real-world deployments in factories.
This week’s AI news saw major updates from Google, OpenAI, Microsoft, and Perplexity, along with significant developments in robotics. Google released Nano Banana 2, an improved image generation model with enhanced detail, advanced world knowledge, better text rendering, translation, 4K upscaling, aspect ratio control, and subject consistency. They also upgraded Notebook LM with cinematic video overviews, including motion graphics, available to Ultra plan users. Additionally, Google launched Gemini 3.1 Pro, a flagship multimodal model capable of handling video, audio, images, and text with strong reasoning, a million-token context window, and robust function calling—making it ideal for complex professional and developer tasks.
OpenAI responded with the release of GPT-5.4 Pro, now considered the most advanced model for high-stakes reasoning, scientific, and technical work. While Gemini 3.1 Pro excels at multimodal tasks, GPT-5.4 Pro dominates in areas like advanced mathematics and scientific problem-solving, and it has improved conversational reliability over previous versions. Microsoft introduced Copilot Tasks, an agentic AI that automates to-do lists by planning and executing tasks across apps and services, marking a shift from conversational chatbots to AI systems that actively carry out work. This tool is currently in limited preview and aims to bring agentic AI to mainstream users.
Perplexity launched Perplexity Computer, a powerful digital worker that orchestrates 19 different AI models—including Claude Opus, Gemini, Nano Banana, and GPT-5.2—to autonomously research, code, design, and manage projects. It operates securely in the cloud, remembers past work, and integrates with hundreds of services, targeting non-technical power users at a premium price. Meanwhile, Microsoft faced backlash over aggressive AI integration into its products, with the derogatory term “MicroSlop” gaining traction online, especially after the company banned the word on its Discord server.
The AI sector also experienced significant drama. Anthropic refused to allow its models to be used for mass surveillance or autonomous weapons, leading to public criticism from the U.S. government and threats of federal bans. OpenAI faced its own controversies, including the “Quit GPT” movement, which saw millions of users leave ChatGPT following revelations about political donations, government contracts, and declining model quality. OpenAI also fired an employee for insider trading on AI prediction markets. Additionally, the Quen team disbanded due to internal restructuring, highlighting tensions between technical vision and corporate management.
In robotics, Stanford’s FSME memory system enabled robots to learn physical principles in real time, bridging the gap between abstract knowledge and hands-on experience. Physical Intelligence, a leading robotics startup, introduced MEM (Multi-Skill Embodied Memory), allowing robots to combine short-term visual memory with long-term narrative memory for better task adaptation. Faraday Future entered the humanoid robot market, though their products appear to be rebranded Chinese models. Notably, BMW deployed its first humanoid robot in a European plant, signaling the growing integration of advanced robotics into real-world industrial settings.