InfoQ Homepage Large language models Content on InfoQ
-
Google DeepMind Introduces ATLAS Scaling Laws for Multilingual Language Models
Google DeepMind researchers have introduced ATLAS, a set of scaling laws for multilingual language models that formalize how model size, training data volume, and language mixtures interact as the number of supported languages increases.
-
Google Introduces TranslateGemma Open Models for Multilingual Translation
Google has released TranslateGemma, a set of open translation models based on the Gemma 3 architecture, offering 4B, 12B, and 27B parameter variants designed to support machine translation across 55 languages and to run on platforms ranging from mobile and edge devices to consumer hardware and cloud accelerators.
-
Google Releases Gemma 3 270M Variant Optimized for Function Calling on Mobile and Edge Devices
FunctionGemma is a new, lightweight version of the Gemma 3 270M model, fine-tuned to translate natural language into structured function and API calls, enabling AI agents to "do more than just talk" and act.
-
DoorDash Applies AI to Safety across Chat and Calls, Cutting Incidents by 50%
DoorDash deploys SafeChat, an AI-driven safety system for moderating chat, images, and voice calls between Dashers and customers. Using a layered text moderation architecture, machine learning models, and human review, SafeChat detects unsafe content in real time, enabling immediate actions and reducing low- and medium-severity safety incidents by roughly 50 percent.
-
How CyberArk Protects AI Agents with Instruction Detectors and History-Aware Validation
To prevent agents from obeying malicious instructions hidden in external data, all text entering an agent's context must be treated as untrusted, says Niv Rabin, principal software architect at AI-security firm CyberArk. His team developed an approach based on instruction detection and history-aware validation to protect against both malicious input data and context-history poisoning.
-
MIT's Recursive Language Models Improve Performance on Long-Context Tasks
Researchers at MIT's CSAIL published a design for Recursive Language Models (RLM), a technique for improving LLM performance on long-context tasks. RLMs use a programming environment to recursively decompose and process inputs, and can handle prompts up to 100x longer than base LLMs.
-
Hugging Face Releases FineTranslations, a Trillion-Token Multilingual Parallel Text Dataset
Hugging Face has released FineTranslations, a large-scale multilingual dataset containing more than 1 trillion tokens of parallel text across English and 500+ languages. The dataset was created by translating non-English content from the FineWeb2 corpus into English using Gemma3 27B, with the full data generation pipeline designed to be reproducible and publicly documented.
-
Mistral Releases OCR 3 with Improved Accuracy on Handwritten and Structured Documents
Mistral has released Mistral OCR 3, the latest version of its optical character recognition model, focused on higher accuracy across a wide range of document types, including handwritten notes, forms, low-quality scans, and complex tables.
-
AI-Powered Code Editor Cursor Introduces Dynamic Context Discovery to Improve Token-Efficiency
Cursor introduced a new approach to minimize the context size of requests sent to large language models. Called dynamic context discovery, this method moves away from including large amounts of static context upfront, allowing the agent to dynamically retrieve only the information it needs. This reduces token usage and limits the inclusion of potentially confusing or irrelevant details.
-
Vercel Open-Sources Bash Tool for Context Retrieval Using Local Filesystems
Vercel has open-sourced bash-tool that provides a Bash execution engine for AI agents, enabling them to run filesystem-based commands to retrieve context for model prompts.
-
Google Releases Gemma Scope 2 to Deepen Understanding of LLM Behavior
Gemma Scope 2 is a suite of tools designed to interpret the behavior of Gemini 3 models, enabling researchers to analyze emergent model behaviors, audit and debug AI agents, and devise mitigation strategies against security issues like jailbreaks, hallucinations and sycophancy.
-
NVIDIA Releases Open Models, Datasets, and Tools across AI, Robotics, and Autonomous Driving
NVIDIA has released a set of open models, datasets, and development tools covering language, agentic systems, robotics, autonomous driving, and biomedical research. The update expands several existing NVIDIA model families and makes accompanying training data and reference implementations available through GitHub, Hugging Face, and NVIDIA’s developer platforms.
-
Meta Applies Mutation Testing with LLM to Improve Compliance Coverage
Meta applies large language models to mutation testing through its Automated Compliance Hardening system, generating targeted mutants and tests to improve compliance coverage, reduce overhead, and detect privacy and safety risks. The approach supports scalable, LLM-driven test generation and continuous compliance across Meta’s platforms.
-
Intel DeepMath Introduces a Smart Architecture to Make LLMs Better at Math
Intel has announced DeepMath, a lightweight agent built on Qwen3-Thinking that specializes in solving mathematical problems. To address common limitations of LLMs in math reasoning, DeepMath generates small Python scripts that support and enhance its problem-solving process.
-
Google’s Eight Essential Multi-Agent Design Patterns
Google recently published a guide outlining eight essential design patterns for multi-agent systems, ranging from sequential pipelines to human-in-the-loop architecture. The guide provides concrete explanations of each pattern along with sample code for Google's Agent Development Kit.