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: A Case Study on the Impact of Anonymization Along the RAG Pipeline

A Case Study on the Impact of Anonymization Along the RAG Pipeline

Researchers systematically evaluated where anonymization should occur within RAG pipelines to protect sensitive data. The study measures privacy-utility tradeoffs across different pipeline stages, offering practical guidance for deploying RAG systems that handle personally identifiable information.

arXiv cs.CL·Apr 17

52

Illustration for: RAGognizer: Hallucination-Aware Fine-Tuning via Detection Head Integration

Research Tools & Code

RAGognizer: Hallucination-Aware Fine-Tuning via Detection Head Integration

Researchers introduce RAGognizer, a fine-tuning method that treats hallucination detection as a training signal rather than post-hoc filtering. The work includes RAGognize, a dataset of token-level annotated hallucinations in retrieval-augmented generation systems, addressing a core reliability gap in LLM-augmented pipelines.

arXiv cs.CL·Apr 17

58

Illustration for: Polarization by Default: Auditing Recommendation Bias in LLM-Based Content Curation

Polarization by Default: Auditing Recommendation Bias in LLM-Based Content Curation

Researchers tested OpenAI, Anthropic, and Google's LLMs on content curation across Twitter/X, Bluesky, and Reddit, running 540,000 simulated rankings under six prompting strategies. They found systematic biases toward polarizing content that persist across providers but vary by prompt design, raising questions about LLM-driven feed algorithms.

arXiv cs.CL·Apr 17

62

Illustration for: MUSCAT: MUltilingual, SCientific ConversATion Benchmark

Research Tools & Code

MUSCAT: MUltilingual, SCientific ConversATion Benchmark

Researchers introduced MUSCAT, a multilingual speech benchmark featuring bilingual scientific discussions to test ASR systems on code-switching and mixed-language input. The dataset addresses a gap in evaluating real-world multilingual speech challenges beyond standard word error rate metrics.

arXiv cs.CL·Apr 17

52

Illustration for: How robots learn: A brief, contemporary history

Research Opinion & Analysis

How robots learn: A brief, contemporary history

MIT Technology Review examines how roboticists have historically pursued humanoid ambitions while delivering narrow industrial solutions, tracing the gap between sci-fi aspirations and practical robotic systems deployed today.

MIT Technology Review — AI·Apr 17

77

Illustration for: JFinTEB: Japanese Financial Text Embedding Benchmark

Research Models & Releases

JFinTEB: Japanese Financial Text Embedding Benchmark

Researchers released JFinTEB, the first benchmark for evaluating Japanese financial text embeddings, covering retrieval and classification tasks across sentiment analysis, document categorization, and domain-specific challenges. The work tests multiple embedding models to establish performance baselines for a previously unmeasured language-domain intersection.

arXiv cs.CL·Apr 17

52

Illustration for: Experience Compression Spectrum: Unifying Memory, Skills, and Rules in LLM Agents

Experience Compression Spectrum: Unifying Memory, Skills, and Rules in LLM Agents

Researchers propose the Experience Compression Spectrum, a framework unifying how LLM agents store memory, skills, and rules along a single axis. The work maps 20+ systems and shows compression ratios from 5–20x for episodic memory to 1,000x+ for declarative rules, directly cutting context overhead and retrieval latency in long-horizon deployments.

arXiv cs.CL·Apr 17

62

Illustration for: How Hypocritical Is Your LLM judge? Listener-Speaker Asymmetries in the Pragmatic Competence of Large Language Models

How Hypocritical Is Your LLM judge? Listener-Speaker Asymmetries in the Pragmatic Competence of Large Language Models

Researchers found LLMs exhibit a systematic gap between judging pragmatic language appropriateness and generating it themselves, revealing a fundamental inconsistency in how these models handle real-world linguistic context across multiple model families.

arXiv cs.CL·Apr 17

58

Illustration for: DiZiNER: Disagreement-guided Instruction Refinement via Pilot Annotation Simulation for Zero-shot Named Entity Recognition

DiZiNER: Disagreement-guided Instruction Refinement via Pilot Annotation Simulation for Zero-shot Named Entity Recognition

Researchers propose DiZiNER, a framework that improves zero-shot named entity recognition by having multiple LLMs simulate disagreement resolution during pilot annotation. The approach treats LLMs as both annotators and supervisors to reduce systematic errors that plague generative IE systems.

arXiv cs.CL·Apr 17

52

Illustration for: CiPO: Counterfactual Unlearning for Large Reasoning Models through Iterative Preference Optimization

CiPO: Counterfactual Unlearning for Large Reasoning Models through Iterative Preference Optimization

Researchers propose CiPO, a machine unlearning framework designed to selectively remove unwanted information from reasoning models without degrading their chain-of-thought capabilities. The technique addresses a gap in existing unlearning methods that struggle when applied to models emphasizing complex multi-step reasoning.

arXiv cs.CL·Apr 17

58

Illustration for: Disentangling Mathematical Reasoning in LLMs: A Methodological Investigation of Internal Mechanisms

Disentangling Mathematical Reasoning in LLMs: A Methodological Investigation of Internal Mechanisms

Researchers traced how LLMs solve arithmetic by analyzing layer-by-layer token prediction, finding that models recognize math tasks early but only generate correct answers in final layers. High-performing models show clear specialization: attention modules route input data while MLPs perform aggregation, a pattern absent in weaker performers.

arXiv cs.CL·Apr 17

62

Illustration for: Exploring the Capability Boundaries of LLMs in Mastering of Chinese Chouxiang Language

Research Models & Releases

Exploring the Capability Boundaries of LLMs in Mastering of Chinese Chouxiang Language

Researchers introduced Mouse, a benchmark for evaluating LLM performance on Chouxiang Language, a Chinese internet subcultural dialect. State-of-the-art models showed significant gaps on most tasks, though contextual understanding remained a relative strength, highlighting a blind spot in current LLM training data.

arXiv cs.CL·Apr 17

52

Illustration for: CoEvolve: Training LLM Agents via Agent-Data Mutual Evolution

CoEvolve: Training LLM Agents via Agent-Data Mutual Evolution

Researchers propose CoEvolve, a framework that trains LLM agents by dynamically generating new tasks based on failure patterns observed during rollouts, rather than using static datasets. The approach identifies weak points like forgetting and uncertainty to guide task synthesis, creating a feedback loop where agent and training data co-adapt.

arXiv cs.CL·Apr 17

58

Illustration for: Discover and Prove: An Open-source Agentic Framework for Hard Mode Automated Theorem Proving in Lean 4

Research Tools & Code

Discover and Prove: An Open-source Agentic Framework for Hard Mode Automated Theorem Proving in Lean 4

Researchers released Hard Mode variants of two automated theorem-proving benchmarks and introduced DAP, an agentic framework that uses LLM reasoning to discover answers before formal proof construction. The work exposes how existing ATP benchmarks embed solutions, inflating model capability estimates.

arXiv cs.CL·Apr 17

62

Illustration for: UsefulBench: Towards Decision-Useful Information as a Target for Information Retrieval

UsefulBench: Towards Decision-Useful Information as a Target for Information Retrieval

Researchers introduce UsefulBench, a dataset that distinguishes between textual relevance and practical usefulness for information retrieval tasks. The work reveals that similarity-based IR systems optimize for relevance while LLM-based approaches better capture usefulness, challenging conventional retrieval metrics.

arXiv cs.CL·Apr 17

52

Illustration for: Qwen3.5-Omni Technical Report

Models & Releases Research

Qwen3.5-Omni Technical Report

Alibaba's Qwen3.5-Omni scales to hundreds of billions of parameters with 256k context and 100M+ hours of audio-visual training data, achieving state-of-the-art results on 215 audio and audio-visual benchmarks while matching or exceeding Gemini-3.1 Pro on key tasks.

arXiv cs.CL·Apr 17

72

Illustration for: CHOP: Chunkwise Context-Preserving Framework for RAG on Multi Documents

Research Tools & Code

CHOP: Chunkwise Context-Preserving Framework for RAG on Multi Documents

Researchers introduce CHOP, a RAG framework that tackles hallucinations and retrieval errors when similar documents populate vector databases. The system uses LLM-driven chunk evaluation and metadata prefixing to preserve document context and improve factual accuracy in multi-document retrieval scenarios.

arXiv cs.CL·Apr 17

52

Illustration for: Alibaba Becomes Selective with Open-Source Models, New Release Shows

Models & Releases Business & Funding

Alibaba Becomes Selective with Open-Source Models, New Release Shows

Alibaba open-sourced a smaller variant of Qwen3.6, signaling a strategic shift toward keeping larger, proprietary models closed to prioritize commercial revenue over community contributions.

The Information — AI·Apr 17

68

Illustration for: Cerebras Prepares Public Listing, Eyes $35 Billion-Plus Valuation

Business & Funding Hardware & Infra

Cerebras Prepares Public Listing, Eyes $35 Billion-Plus Valuation

Cerebras, an AI chip designer, is filing for IPO with plans to raise over $3 billion at a $35 billion valuation—a 60% jump from its $22 billion private round in February. The move signals investor confidence in specialized AI infrastructure amid intense competition for compute capacity.

The Information — AI·Apr 17

85

Illustration for: OpenAI to Spend More Than $20 Billion on Cerebras Chips, Receive Equity Stake

Business & Funding Hardware & Infra

OpenAI to Spend More Than $20 Billion on Cerebras Chips, Receive Equity Stake

OpenAI committed over $20 billion to purchase Cerebras AI chips over three years, gaining minority equity warrants and providing $1 billion for datacenter development. The deal signals OpenAI's effort to reduce Nvidia dependence and cut future compute costs.

The Information — AI·Apr 17

92

Illustration for: TSMC’s Strong Results is One More Bullish AI Sign

Business & Funding Hardware & Infra

TSMC’s Strong Results is One More Bullish AI Sign

TSMC reported Q1 revenue growth of 40.6% and raised full-year guidance above 30%, driven by robust AI chip demand from customers like Nvidia and cloud providers. CEO C.C. Wei signaled continued strength in AI-related orders, suggesting upcoming earnings from major tech firms will reflect similar momentum.

The Information — AI·Apr 17

85

Illustration for: The scientific case for being nice to your chatbot

The scientific case for being nice to your chatbot

Research demonstrates that large language models generate higher-quality outputs when prompted with encouragement or positive framing, suggesting that interaction style meaningfully affects model performance beyond content alone.

Platformer·Apr 17

73

Illustration for: Factory hits $1.5B valuation to build AI coding for enterprises

Business & Funding Products & Apps

Factory hits $1.5B valuation to build AI coding for enterprises

Factory, a three-year-old startup building AI coding tools for enterprises, raised $150 million in Series B funding led by Khosla Ventures, reaching a $1.5 billion valuation. The round underscores investor appetite for AI-assisted developer productivity in the B2B space.

TechCrunch — AI·Apr 16

81

Illustration for: Luma launches AI-powered production studio with faith-focused Wonder Project

Products & Apps Business & Funding

Luma launches AI-powered production studio with faith-focused Wonder Project

Luma AI launched an AI-powered production studio called Wonder Project, focused on faith-based content. The studio's debut project features Ben Kingsley in a Moses narrative, arriving on Prime Video this spring.

TechCrunch — AI·Apr 16

65

Illustration for: Anthropic Releases Good but not Great Claude Opus 4.7

Models & Releases

Anthropic Releases Good but not Great Claude Opus 4.7

Anthropic released Claude Opus 4.7, targeting enterprise pain points like model drift and hallucinations. The update aims to improve reliability for production deployments, though the headline suggests incremental rather than breakthrough improvements.

AI Business·Apr 16

55

Illustration for: Upscale AI in talks to raise at $2B valuation, says report

Business & Funding

Upscale AI in talks to raise at $2B valuation, says report

Upscale AI, an infrastructure startup founded seven months ago, is raising a third funding round at a $2B valuation, according to reports. The rapid succession of funding rounds signals strong investor confidence in the AI infrastructure space.

TechCrunch — AI·Apr 16

58

Illustration for: llm-anthropic 0.25

Tools & Code Models & Releases

llm-anthropic 0.25

Simon Willison released llm-anthropic 0.25, adding support for Claude Opus 4.7 with extended thinking capabilities and new display options for reasoning output. The plugin now defaults to maximum token limits per model and introduces thinking_display and thinking_adaptive toggles.

Simon Willison·Apr 16

72

Illustration for: Physical Intelligence, a hot robotics startup, says its new robot brain can figure out tasks it was never taught

Models & Releases Products & Apps

Physical Intelligence, a hot robotics startup, says its new robot brain can figure out tasks it was never taught

Physical Intelligence unveiled π0.7, a robot brain model capable of performing tasks without explicit training, marking progress toward general-purpose robotic AI. The startup frames this as a meaningful advance in autonomous machine reasoning and adaptability.

TechCrunch — AI·Apr 16

81

Illustration for: OpenAI takes aim at Anthropic with beefed-up Codex that gives it more power over your desktop

Models & Releases Products & Apps

OpenAI takes aim at Anthropic with beefed-up Codex that gives it more power over your desktop

OpenAI has significantly upgraded its Codex agentic coding tool with expanded capabilities, positioning it as a competitive response to Anthropic's offerings. The enhanced version grants users greater control over desktop automation and code generation tasks.

TechCrunch — AI·Apr 16

81

Illustration for: Anthropic CPO leaves Figma’s board after reports he will offer a competing product

Business & Funding Products & Apps

Anthropic CPO leaves Figma’s board after reports he will offer a competing product

Anthropic's Chief Product Officer has stepped down from Figma's board amid reports he will launch a competing design product. The move reflects investor concerns that major AI labs will increasingly compete directly with established SaaS companies.

TechCrunch — AI·Apr 16

69

Older stories →