The LLM archive monitors model releases, quantization updates, reasoning capabilities, and real-world deployment implications for local and hybrid AI. We focus on what materially changes selection and operations: context windows, latency, memory footprint, licensing, and evaluation evidence across open and commercial families. This section is designed for teams that need dependable model intelligence, not hype cycles. Pair these updates with the LLM pillar and references to hardware constraints and framework integration.
The Trends Explore page for users to analyze search interest just got a major upgrade. It now uses Gemini to identify and compare relevant trends.
Google is integrating "personal intelligence" into Gemini, allowing the chatbot to connect to Gmail, Photos, Search, and YouTube. The goal is to provide more useful and personalized answers. The feature is optional and available to AI Pro and AI Ultra subscribers, who can choose which data sources to connect. The integration leverages Google's vast amount of personal data to improve the accuracy of responses.
The West Midlands police admitted using hallucinated information from Microsoft Copilot to ban Maccabi Tel Aviv football fans from the UK. Initially denied, the use of AI was confirmed after weeks of controversy surrounding a safety advisory group meeting for the Aston Villa-Maccabi Tel Aviv match, amid heightened tensions following a terrorist attack in Manchester.
Kaggle introduces Community Benchmarks, a platform that allows the community to build, share, and run custom evaluations for AI models. The initiative aims to foster transparency and reproducibility in model evaluation, enabling researchers and developers to compare performance more effectively and identify areas for improvement.
Two years ago, companies like Meta and OpenAI were united against military use of their tools. Now all of that has changed.
The winner of the Global AI Film Award has been announced, a recognition for creators who use artificial intelligence models and tools to tell innovative stories. The initiative celebrates the creative use of AI in cinema.
A new study introduces MedES, a dynamic benchmark for aligning large language models (LLMs) with Chinese medical ethics. The system uses an automated evaluator to provide structured ethical feedback, improving model performance in complex clinical scenarios. Results show significant improvements over baseline models, paving the way for similar deployments in other legal and cultural contexts.
A new study introduces State-Centric Retrieval, a unified paradigm for Retrieval-Augmented Generation (RAG) that uses "states" to connect embedding models and rerankers. The approach, based on a fine-tuned RWKV model, promises significant improvements in efficiency and speed, reducing computational redundancy and accelerating inference. Experimental results show near-complete performance retention with reduced resource usage.
A new study highlights the challenges of regularization-based continual learning in EEG-based emotion classification. Existing methods show limited performance due to inter- and intra-subject variability, and tend to prioritize mitigating catastrophic forgetting over adapting to new subjects. This limits robust generalization to unseen subjects.
A novel approach to compressing large language models (LLMs) promises to significantly reduce memory requirements and computational resources. The technique, called Hierarchical Sparse Plus Low-Rank (HSS) compression, combines sparsity with low-rank factorization to compress models while maintaining competitive performance. Results show significant memory savings with minimal accuracy loss.
New research addresses the challenge of ensuring that Large Language Models (LLMs) adhere to safety principles without refusing benign requests. The study evaluates the impact of explicitly specifying extensive safety codes versus demonstrating them through illustrative cases, proposing a case-augmented deliberative alignment method (CADA) to enhance the safety and robustness of LLMs.
A new study introduces a hybrid explainable AI (XAI) framework for assessing maternal health risks in resource-constrained settings. The model, validated by clinicians in Bangladesh, combines ante-hoc fuzzy logic with post-hoc SHAP explanations, enhancing trust and clinical adoption. Healthcare access was identified as the primary predictor.
By rolling out ChatGPT Enterprise company-wide, Zenken has boosted sales performance, cut preparation time, and increased proposal success rates. AI-supported workflows are helping a lean team deliver more personalized, effective customer engagement.
US Defense Secretary Pete Hegseth said he plans to integrate Elon Musk's AI tool, Grok, into Pentagon networks later this month. The announcement comes weeks after Grok drew international backlash for generating sexualized images of women and children. Hegseth also rolled out an "AI acceleration strategy" for the Department of Defense.
Anthropic has announced the launch of Anthropic Labs, a new division focused on cutting-edge research and development projects in the field of artificial intelligence. The initiative aims to accelerate innovation and explore new frontiers in the sector.
A consumer watchdog has raised concerns about Google's new Universal Commerce Protocol, arguing it could lead to higher prices for consumers. Google strongly denies these claims, defending the integrity of its system.
OpenAI and Anthropic have recently launched healthcare-focused products. Doctors are interested in adopting AI, but with reservations about using chatbots for patient care. The integration of AI in the medical field opens new perspectives, but requires careful evaluation of risks and benefits.
LangSmith Agent Builder is now generally available, allowing users to create no-code AI agents to automate routine tasks such as research, follow-ups, and updates. Agents can be shared, integrated with other tools, and customized with specific models. Ideal for daily briefings, market research, and project tracking.
LangSmith Agent Builder is now generally available, designed to automate routine tasks. It allows the creation of no-code agents that learn from feedback, aiming to reduce workload and improve operational efficiency. Ideal for briefings, market research, and project management, Agent Builder integrates with existing tools and adapts to team needs, enabling users to share, customize, and extend agent capabilities.
Salesforce has announced Slackbot, a new artificial intelligence-powered agent designed to allow users to complete complex tasks within various enterprise applications directly from Slack. The goal is to simplify workflows and improve productivity by centralizing task execution in a single interface.