Google DeepMind has added “shutdown resistance” and “harmful manipulation” to its safety framework in response to advanced AI models gaining autonomy and influence.
DeepMind Tackles AI Safety with New Risk Categories Read More »
Meta proposes a handbook system for LLMs that cuts reasoning tokens by up to 46%, improving efficiency and accuracy.
Reusing Reasoning Steps: A New Approach for LLMs Read More »
Google’s Gemini 2.5 Flash Image offers powerful image generation and editing features, making it easier for developers to create creative visual applications.
Nano Banana – Google’s New Image Model Read More »
Karpathy questions the reliability of reinforcement learning in training LLMs, suggesting interactive environments for learning through action instead.
Rethinking How AI Learns: Beyond Reinforcement Strategies Read More »
Microsoft’s new AI models, Mai-Voice-1 and Mai-1-Preview, aim to reduce reliance on OpenAI while enhancing efficiency and lowering costs for users.
Microsoft Introduces New AI Models to Shift the Landscape Read More »
Farang raises €1.5 million to advance its innovative AI models, focusing on specialized applications in programming and medicine while ensuring data privacy.
Farang Secures €1.5M to Transform AI with Brain-like Models Read More »