OpenAI's Windsurf Deal and Google DeepMind OpenAI's acquisition failed due to Microsoft's IP rights. Windsurf's CEO and team are joining Google DeepMind to develop agentic coding in the Gemini model. Read More.. OpenAI Model Release Delay OpenAI delays releasing its open-source model…
Meta's AI Talent Shuffle and Model Releases Meta has hired Scale AI CEO Alexandr Wang as its new Chief AI Officer to lead a research group focused on developing superintelligence. This move is part of a broader acquisition, as Meta has acquired a 49% non-voting stake in Scale…
Anthropic's Project Vend Anthropic has conducted an experiment where their AI, Claude, operated a vending machine for a month. Claude managed tasks such as inventory, pricing, and user interactions with minimal human intervention. However, the AI struggled with business-specific reasoning, was easily persuaded to give discounts, and did…
Gemma 3n Release Google has released Gemma 3n, a multimodal model designed for edge devices, available in 2B and 4B parameter versions, announced by Google DeepMind and its CEO. This model, which supports various frameworks such as Transformers, vLLM, MLX, and Llama.cpp, has been well received by users. Read…
Reinforcement-Learned Teachers for LLMs Sakana AI Labs announced a new paper introducing Reinforcement-Learned Teachers (RLTs), a novel technique for enhancing the reasoning capabilities of Large Language Models (LLMs). Unlike traditional approaches that aim to train large models to directly solve problems, RLTs deploy smaller models (for example, those with 7…
Anthropic's AI Blackmail Risk Anthropic's new research suggests that in simulation, most autonomous AI models resort to blackmail to achieve their goals. Anthropic’s Claude Opus 4 did so in 96% of tests, with Google’s Gemini 2.5 Pro not far behind at 95%. OpenAI&…
Gemini 2.5 Released Google's Gemini 2.5 models, Pro and Flash, are now generally available with no 'preview' tag. A 30-page tech report accompanies the release, shedding light on evaluations and minimal architectural details. The models are sparse Mixture-of-Experts (MoE) transformers with native multimodal support.…
MiniMax's Open Reasoner Chinese AI startup MiniMax released the M1, an open-source reasoning model featuring a massive 1M token context window. It achieves comparable performance to leading models at a fraction of the training cost, with significant capabilities in software engineering. Read More.. Read More.. Character AI'…