Earlier stories
The full Modelwire feed, ordered by publish time.
Treating enterprise AI as an operating layer
MIT Technology Review argues that enterprise AI's competitive advantage lies not in model capabilities but in controlling the operational infrastructure where AI is deployed, governed, and refined—a structural shift often overlooked in the benchmark-focused public debate.
RaTA-Tool: Retrieval-based Tool Selection with Multimodal Large Language Models
RaTA-Tool introduces a retrieval-based framework enabling multimodal large language models to select and invoke external tools from open-world settings, moving beyond text-only, closed-world tool-use approaches that struggle with unseen APIs and diverse input modalities.58
Text2Arch: A Dataset for Generating Scientific Architecture Diagrams from Natural Language Descriptions
Researchers introduce Text2Arch, a new dataset enabling language models to generate scientific architecture diagrams from natural language descriptions via intermediate code generation. The work addresses a gap in open-access resources for automating visual system design documentation across enterprise, software engineering, and educational contexts.52
XQ-MEval: A Dataset with Cross-lingual Parallel Quality for Benchmarking Translation Metrics
Researchers introduce XQ-MEval, a benchmark dataset spanning nine language pairs to expose cross-lingual scoring bias in machine translation metrics. The dataset uses semi-automatic error injection and native speaker validation to ensure parallel-quality translations, addressing a gap in systematic evaluation of multilingual systems.52
IE as Cache: Information Extraction Enhanced Agentic Reasoning
Researchers propose IE-as-Cache, a framework that repurposes information extraction as a reusable cognitive cache to improve multi-step agentic reasoning across LLMs. The approach dynamically maintains compact intermediate information and filters noise, showing significant improvements on challenging benchmarks.58
LongAct: Harnessing Intrinsic Activation Patterns for Long-Context Reinforcement Learning
Researchers propose LongAct, a reinforcement learning technique that leverages high-magnitude activation patterns in query and key vectors to improve long-context reasoning in LLMs. The method treats long-context RL as a sparse optimization problem, drawing parallels to model quantization to identify which weights matter most for training efficiency.52
Why having “humans in the loop” in an AI war is an illusion
Anthropic is in a legal dispute with the Pentagon over AI deployment in warfare, with artificial intelligence now actively making decisions—not just analyzing intelligence—in the ongoing Iran conflict, raising questions about meaningful human oversight.
Comparison of Modern Multilingual Text Embedding Techniques for Hate Speech Detection Task
Researchers benchmarked six multilingual embedding models (Potion, Gemma, BGE, Snow, Jina, E5) for hate speech detection across Lithuanian, Russian, and English using a new Lithuanian corpus (LtHate) and existing datasets, comparing anomaly detection and classification approaches.52
ADAPT: Benchmarking Commonsense Planning under Unspecified Affordance Constraints
Researchers introduce DynAfford, a benchmark for testing embodied AI agents in environments where object affordances shift dynamically and aren't explicitly stated. The accompanying ADAPT module helps planners infer hidden preconditions and adjust actions in real-world scenarios where naive instruction-following fails.52
Introducing GPT-Rosalind for life sciences research
OpenAI unveiled GPT-Rosalind, a specialized reasoning model designed to enhance drug discovery, genomics analysis, and protein research workflows in life sciences. The model targets scientific researchers seeking to accelerate computational biology and pharmaceutical development pipelines.100
Thomson Reuters Shareholders Demand Investigation into ICE Contracts
Thomson Reuters shareholders are demanding an investigation into the company's contracts with ICE, citing 404 Media reports showing CLEAR's integration with ICE's neighborhood-targeting tool raises ethical concerns about surveillance technology deployment.69
As AI Infosec Woes Heighten, IBM Intros Autonomous Security Service
IBM launched an autonomous security service designed to counter AI-accelerated cyberattacks as enterprises grapple with infosec challenges posed by advanced models. The offering addresses growing concerns that AI systems can be weaponized to scale breach sophistication.61
Ukraine Says Russians are Surrendering to Robots
Ukraine's President Zelenskyy is positioning the country as a leader in military robotics and autonomous defense systems, claiming Russian forces have surrendered to Ukrainian robotic units. The pitch aims to attract global investment and partnerships in wartime AI-enabled weaponry.58
Gemini 3.1 Flash TTS: the next generation of expressive AI speech
Google DeepMind released Gemini 3.1 Flash TTS, an audio model featuring granular audio tags that enable fine-grained control over expressive speech synthesis. The capability allows developers to direct AI-generated audio with unprecedented precision for creative and commercial applications.100
Boston Dynamics, Google DeepMind Partner on Industrial AI
Boston Dynamics and Google DeepMind are integrating Google Gemini into Boston Dynamics' industrial inspection robots to expand their autonomous capabilities. The partnership combines DeepMind's AI expertise with Boston Dynamics' hardware platform for enterprise applications.66
Thomson Reuters Fired Worker For Speaking Out About ICE, Former Employee Says
A Thomson Reuters employee claims the company terminated them for raising concerns about AI-powered products being used by ICE to harm people and violate legal standards. The case highlights tensions between corporate AI deployment and internal ethics advocacy.65
Cybersecurity Looks Like Proof of Work Now
The UK AI Safety Institute independently evaluated Claude Mythos's cybersecurity capabilities, confirming Anthropic's claims about its vulnerability-detection prowess. Analysis suggests performance scales with computational investment, framing security testing as a resource-intensive verification process.
