Models & Releases Research Products & Apps Business & Funding

Modelwire

A curated feed of what matters in AI. Independent, ad-supported, built in Denver, Colorado.

Read

Today
Models & Releases
Research
Business & Funding

About

About Modelwire
Methodology
Our sources
Editor's notes
Contact
Advertise

Legal

Privacy policy
Terms of use
DMCA & takedowns
Corrections

© 2026 Modelwire. All article links go to the original publishers.Summaries generated by Modelwire. We don’t republish full articles.

Earlier stories

The full Modelwire feed, ordered by publish time.

Illustration for: Improving clinical interpretability of linear neuroimaging models through feature whitening

Improving clinical interpretability of linear neuroimaging models through feature whitening

Researchers propose a feature whitening technique to improve how linear neuroimaging models identify brain biomarkers, addressing the problem that correlated brain regions produce uninterpretable weights that obscure region-specific contributions to disease.

arXiv cs.LG·Apr 22

52

Illustration for: ORPHEAS: A Cross-Lingual Greek-English Embedding Model for Retrieval-Augmented Generation

Research Models & Releases

ORPHEAS: A Cross-Lingual Greek-English Embedding Model for Retrieval-Augmented Generation

Researchers introduced ORPHEAS, a specialized embedding model optimized for Greek-English bilingual retrieval-augmented generation. Unlike general multilingual models that spread capacity across many languages, ORPHEAS uses knowledge-graph-based fine-tuning on domain-specific corpora to better capture Greek morphology and terminology.

arXiv cs.CL·Apr 22

42

Illustration for: GRPO-VPS: Enhancing Group Relative Policy Optimization with Verifiable Process Supervision for Effective Reasoning

GRPO-VPS: Enhancing Group Relative Policy Optimization with Verifiable Process Supervision for Effective Reasoning

Researchers propose GRPO-VPS, a refinement to Group Relative Policy Optimization that uses process-level supervision to improve LLM reasoning. The method tracks model confidence in correct answers across generation steps, addressing credit assignment problems that cause inefficient reasoning chains.

arXiv cs.LG·Apr 22

58

Illustration for: Cooperative Profiles Predict Multi-Agent LLM Team Performance in AI for Science Workflows

Research Models & Releases

Cooperative Profiles Predict Multi-Agent LLM Team Performance in AI for Science Workflows

Researchers benchmarked 35 open-weight LLMs using behavioral economics games and found that cooperative profiles from these games reliably predict how well LLM teams perform on real AI-for-Science tasks like collaborative data analysis and modeling.

arXiv cs.CL·Apr 22

62

Illustration for: Now Meta will track what employees do on their computers to train its AI agents

Business & Funding Policy & Regulation

Now Meta will track what employees do on their computers to train its AI agents

Meta is deploying employee monitoring software called Model Capability Initiative on US workers' computers to capture mouse movements, keystrokes, and screenshots for training AI agents. The practice raises immediate questions about consent, data governance, and whether surveillance-derived training data will become standard across the industry.

The Verge — AI·Apr 22

81

Illustration for: OpenAI teams up with Infosys to bring AI tools to more businesses

Business & Funding Products & Apps

OpenAI teams up with Infosys to bring AI tools to more businesses

OpenAI and Infosys are partnering to embed AI capabilities into enterprise software development workflows. The collaboration targets legacy system modernization, DevOps automation, and AI deployment across Infosys's client base.

TechCrunch — AI·Apr 22

58

Illustration for: Self-Guided Plan Extraction for Instruction-Following Tasks with Goal-Conditional Reinforcement Learning

Research Tools & Code

Self-Guided Plan Extraction for Instruction-Following Tasks with Goal-Conditional Reinforcement Learning

SuperIgor pairs a language model's plan generation with reinforcement learning feedback in a closed loop, letting both components improve jointly without predefined subtasks. The framework shows stronger instruction adherence and generalization than baselines on stochastic environments.

arXiv cs.CL·Apr 22

58

Illustration for: Self-Aware Vector Embeddings for Retrieval-Augmented Generation: A Neuroscience-Inspired Framework for Temporal, Confidence-Weighted, and Relational Knowledge

Research Tools & Code

Self-Aware Vector Embeddings for Retrieval-Augmented Generation: A Neuroscience-Inspired Framework for Temporal, Confidence-Weighted, and Relational Knowledge

Researchers propose SmartVector, a framework that embeds temporal awareness, confidence decay, and relational metadata into vector embeddings for RAG systems. The approach addresses a real failure mode where conventional RAG achieves only 58% accuracy on versioned queries by modeling knowledge consolidation after hippocampal-neocortical memory processes.

arXiv cs.CL·Apr 22

62

Illustration for: Trust, Lies, and Long Memories: Emergent Social Dynamics and Reputation in Multi-Round Avalon with LLM Agents

Trust, Lies, and Long Memories: Emergent Social Dynamics and Reputation in Multi-Round Avalon with LLM Agents

Researchers observed LLM agents playing repeated rounds of The Resistance: Avalon develop persistent reputation systems across games, with agents explicitly referencing past behavior and role-conditional trust patterns emerging organically. High-reputation players saw 46% higher team inclusion rates, suggesting agents internalize social memory to inform strategic decisions.

arXiv cs.CL·Apr 22

58

Illustration for: Ask Only When Needed: Proactive Retrieval from Memory and Skills for Experience-Driven Lifelong Agents

Ask Only When Needed: Proactive Retrieval from Memory and Skills for Experience-Driven Lifelong Agents

Researchers introduce ProactAgent, a lifelong learning framework that retrieves relevant past experiences during task execution rather than waiting for task completion. The approach combines policy updates with memory refinement to help agents identify and fill knowledge gaps in real time.

arXiv cs.CL·Apr 22

52

Illustration for: Where Reasoning Breaks: Logic-Aware Path Selection by Controlling Logical Connectives in LLMs Reasoning Chains

Where Reasoning Breaks: Logic-Aware Path Selection by Controlling Logical Connectives in LLMs Reasoning Chains

Researchers pinpoint logical connectives as failure points where LLMs derail multi-step reasoning chains, then propose a framework that intervenes at these high-entropy decision nodes to steer models toward correct logical paths and improve chain stability.

arXiv cs.CL·Apr 22

58

Illustration for: LLM StructCore: Schema-Guided Reasoning Condensation and Deterministic Compilation

LLM StructCore: Schema-Guided Reasoning Condensation and Deterministic Compilation

Researchers submitted a two-stage system for automatically filling clinical case report forms from medical notes, using schema-guided reasoning to produce structured JSON summaries followed by deterministic compilation. The approach tackles extreme data sparsity and high false-positive penalties in healthcare documentation tasks.

arXiv cs.CL·Apr 22

42

Illustration for: Anthropic manager hints that Pro and Max plans are outgrown by today's Claude workloads

Products & Apps Business & Funding

Anthropic manager hints that Pro and Max plans are outgrown by today's Claude workloads

Anthropic's Head of Growth signaled that Claude's Pro and Max tiers no longer fit user behavior after the company briefly yanked Code features from Pro before reversing course. The incident suggests subscription architecture misalignment with actual workload demands.

The Decoder·Apr 22

68

Illustration for: LayerTracer: A Joint Task-Particle and Vulnerable-Layer Analysis framework for Arbitrary Large Language Model Architectures

Research Tools & Code

LayerTracer: A Joint Task-Particle and Vulnerable-Layer Analysis framework for Arbitrary Large Language Model Architectures

Researchers introduce LayerTracer, a framework for analyzing any LLM architecture by tracking where task knowledge forms across layers and identifying structural vulnerabilities. The tool works across Transformers, Mamba, and other designs, offering insights for model optimization and hybrid architecture development.

arXiv cs.CL·Apr 22

58

Illustration for: Toward Cross-Lingual Quality Classifiers for Multilingual Pretraining Data Selection

Toward Cross-Lingual Quality Classifiers for Multilingual Pretraining Data Selection

Researchers show that quality signals in embedding space transfer across languages, enabling high-resource language classifiers to filter training data for low-resource ones. Testing on a 1B model trained on 103B tokens, multilingual pooling outperformed monolingual baselines in both stability and accuracy, suggesting a scalable path for data curation in imbalanced multilingual settings.

arXiv cs.CL·Apr 22

58

Illustration for: Enhancing Research Idea Generation through Combinatorial Innovation and Multi-Agent Iterative Search Strategies

Research Tools & Code

Enhancing Research Idea Generation through Combinatorial Innovation and Multi-Agent Iterative Search Strategies

Researchers propose a multi-agent LLM framework that iteratively generates and refines research ideas by combining knowledge search with combinatorial innovation theory, aiming to reduce repetition and surface novel directions in crowded scientific domains.

arXiv cs.CL·Apr 22

52

Illustration for: Aligning Stuttered-Speech Research with End-User Needs: Scoping Review, Survey, and Guidelines

Aligning Stuttered-Speech Research with End-User Needs: Scoping Review, Survey, and Guidelines

Researchers analyzed gaps between stuttered-speech AI research and real-world needs through a scoping review and survey of 70 stakeholders including people who stutter and speech pathologists. The work reveals current speech recognition systems underperform on atypical speech and proposes a taxonomy to realign research priorities with end-user requirements.

arXiv cs.CL·Apr 22

52

Illustration for: Startups Brag They Spend More Money on AI Than Human Employees

Business & Funding

Startups Brag They Spend More Money on AI Than Human Employees

A cohort of AI startups are redirecting hiring budgets toward compute infrastructure, signaling a shift in how early-stage companies allocate capital in the race for model training capacity. The move reflects both the capital intensity of AI development and startups' bet that compute spending outpaces traditional headcount.

404 Media·Apr 22

65

Illustration for: Effects of Cross-lingual Evidence in Multilingual Medical Question Answering

Effects of Cross-lingual Evidence in Multilingual Medical Question Answering

Researchers tested how cross-lingual retrieval strategies affect multilingual medical QA across six languages, finding that English web data helps high-resource languages most while low-resource languages benefit from hybrid retrieval approaches. The work reveals scaling disparities: larger models dominate English baselines, but external knowledge narrows gaps for underserved languages.

arXiv cs.CL·Apr 22

52

Illustration for: AI is spitting out more potential drugs than ever. This start-up wants to figure out which ones matter.

Business & Funding Products & Apps

AI is spitting out more potential drugs than ever. This start-up wants to figure out which ones matter.

10x Science closed a $4.8M seed round to build AI tools that help pharma researchers prioritize drug candidates by interpreting molecular complexity. The startup addresses a real bottleneck: generative models now produce far more potential compounds than labs can feasibly validate.

TechCrunch — AI·Apr 22

65

Illustration for: Unauthorized users breach Anthropic's restricted Mythos AI model

Models & Releases Policy & Regulation

Unauthorized users breach Anthropic's restricted Mythos AI model

Unauthorized users gained access to Anthropic's restricted Claude Mythos model, marking a security breach at a major AI lab. The incident raises questions about access controls on frontier models during limited deployment phases.

The Decoder·Apr 22

73

Illustration for: CHASM: Unveiling Covert Advertisements on Chinese Social Media

CHASM: Unveiling Covert Advertisements on Chinese Social Media

Researchers released CHASM, a 4,992-instance dataset for evaluating multimodal LLMs on detecting covert ads disguised as organic posts on Chinese social media. The benchmark addresses a gap in current LLM evaluation suites by testing real-world deception detection capabilities.

arXiv cs.CL·Apr 22

52

Illustration for: ChatGPT Images Tool Upgraded With ‘Thinking’ Capability

Products & Apps Models & Releases

ChatGPT Images Tool Upgraded With ‘Thinking’ Capability

OpenAI has integrated its reasoning capability into DALL-E, enabling the image generation model to apply deeper computational thinking to visual tasks. The upgrade represents a convergence of reasoning and vision modalities within the company's product suite.

AI Business·Apr 22

61

Illustration for: Meta tracks US employees' clicks and keystrokes to train AI agents

Policy & Regulation Business & Funding

Meta tracks US employees' clicks and keystrokes to train AI agents

Meta deployed keystroke and mouse-tracking software across US employee computers to harvest behavioral data for AI agent training. The move raises questions about consent, data governance, and whether large-scale employee surveillance is becoming standard practice in AI development.

The Decoder·Apr 22

85

Illustration for: Musk's SpaceX bets $60 billion on Cursor to fix xAI's coding gap

Business & Funding Tools & Code

Musk's SpaceX bets $60 billion on Cursor to fix xAI's coding gap

SpaceX has secured an option to acquire Cursor, an AI coding startup, for $60 billion to bolster xAI's developer tooling capabilities. The deal signals Musk's strategy to compete in the AI coding space by acquiring rather than building internally.

The Decoder·Apr 22

85

Illustration for: Knowledge Capsules: Structured Nonparametric Memory Units for LLMs

Research Tools & Code

Knowledge Capsules: Structured Nonparametric Memory Units for LLMs

Researchers propose Knowledge Capsules, a structured memory system that stores external knowledge separately from LLM weights, enabling faster updates and more reliable multi-hop reasoning than standard retrieval-augmented generation. The approach uses an External Key-Value Injection framework to compile knowledge directly from documents without retraining.

arXiv cs.CL·Apr 22

58

Illustration for: Google Maps is about to get a big dose of AI

Products & Apps

Google Maps is about to get a big dose of AI

Google is embedding generative AI into Maps to enhance its core navigation and discovery features. The rollout marks a significant shift toward AI-powered location intelligence for billions of users.

TechCrunch — AI·Apr 22

69

Illustration for: Exclusive: Google deepens Thinking Machines Lab ties with new multi-billion-dollar deal

Business & Funding Hardware & Infra

Exclusive: Google deepens Thinking Machines Lab ties with new multi-billion-dollar deal

Thinking Machines Lab, led by Mira Murati, secured a multi-billion-dollar Google Cloud infrastructure deal centered on Nvidia's GB300 chips. The partnership signals major capital commitment to competing AI infrastructure stacks beyond hyperscaler in-house silicon.

TechCrunch — AI·Apr 22

87

Illustration for: MOMO: A framework for seamless physical, verbal, and graphical robot skill learning and adaptation

Research Tools & Code

MOMO: A framework for seamless physical, verbal, and graphical robot skill learning and adaptation

Researchers introduced MOMO, a robot skill framework combining kinesthetic guidance, natural language commands, and graphical interfaces to let non-experts reprogram industrial robots for new tasks. The system uses energy-based intention detection and tool-calling LLMs to translate human input across modalities into executable robot behaviors.

arXiv cs.CL·Apr 22

58

Illustration for: Finding Duplicates in 1.1M BDD Steps: cukereuse, a Paraphrase-Robust Static Detector for Cucumber and Gherkin

Tools & Code Research

Finding Duplicates in 1.1M BDD Steps: cukereuse, a Paraphrase-Robust Static Detector for Cucumber and Gherkin

Researchers released cukereuse, a static analysis tool that detects duplicate test steps in Gherkin/Cucumber BDD suites using embeddings and fuzzy matching. Analysis of 1.1M steps across 347 GitHub repos found 80% exact duplication rates, revealing significant maintenance overhead in test automation codebases.

arXiv cs.CL·Apr 22

52

Older stories →