“We’re in an LLM bubble,” Hugging Face CEO says—but not an AI one

Microsoft has issued a warning regarding its Copilot Actions feature in Windows, which is currently in beta and disabled by default. The company cautioned that this experimental AI agent could potentially infect devices and steal sensitive user data, raising alarms among security researchers about the implications for user safety and privacy.

DeepMind has introduced AlphaProof, an artificial intelligence designed to tackle mathematical proofs. While it shows promise in handling math challenges, it currently requires some assistance to function effectively.

Thieves at the Louvre successfully exploited human psychology to avoid suspicion during their heist. The article discusses how actions perceived as ordinary can go unnoticed, highlighting a critical aspect of both human behavior and artificial intelligence (AI) systems.

The DRG-Lola is a new concept designed to inspire the next generation of electric single-seaters, featuring innovative twin suction turbines and the capability to handle 3-Gs in slow corners. This design aims to push the boundaries of electric racing technology.

The article discusses the impact of deep learning on drug discovery, particularly highlighting the Tox21 Challenge, which marked a significant shift in the field. Since 2015, deep neural networks have outperformed traditional methods, leading to widespread adoption in the pharmaceutical industry. However, the integration of the Tox21 dataset into benchmarks like MoleculeNet has raised concerns about data alterations, which may affect the comparability of bioactivity and toxicity predictions across studies.

The paper titled 'Deep Learning and Machine Learning -- Natural Language Processing: From Theory to Application' examines the integration of natural language processing (NLP) with machine learning and deep learning. It emphasizes the significance of large language models (LLMs) in various sectors, including healthcare and finance. The study discusses advanced data preprocessing techniques and frameworks like Hugging Face, while addressing challenges such as multilingual data handling, bias reduction, and model robustness to ensure effective AI solutions.

Microsoft is addressing significant security and privacy concerns associated with AI agents in Windows 11, which have read/write access to user files. The introduction of these agents raises alarms about potential unauthorized access to sensitive information, prompting Microsoft to implement measures to mitigate these risks. As the company prepares for the rollout of these features, the implications for user data security remain a critical focus.

Google has announced the launch of its Gemini 3 AI model, marking the second major upgrade of the year. This new model is designed to enhance user interactions by better understanding requests and is part of Google's ongoing efforts to improve its AI capabilities. The introduction of Gemini 3 also includes the unveiling of an AI-first integrated development environment (IDE) called Antigravity, aimed at streamlining AI application development.