Models & Releases Research Products & Apps Business & Funding

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

← Today

Earlier stories

The full Modelwire feed, ordered by publish time.

Research

Live LTL Progress Tracking: Towards Task-Based Exploration

Researchers propose a tracking vector framework for monitoring agent progress on complex, multi-stage tasks specified in linear temporal logic. The method labels task states as true, false, or open at each step, enabling new performance metrics and exploration strategies for non-Markovian reinforcement learning objectives.

arXiv cs.LG·Apr 18

Tools & Code Research

TensorHub: Rethinking AI Model Hub with Tensor-Centric Compression

TensorHub introduces tensor-level deduplication and compression to shrink model storage across repositories without sacrificing performance. The system identifies redundancy across models automatically, addressing a growing pain point as model sizes balloon.

arXiv cs.LG·Apr 18

Research

Tree of Concepts: Interpretable Continual Learners in Non-Stationary Clinical Domains

Researchers propose Tree of Concepts, a continual learning framework that maintains interpretability while adapting to distribution shifts in clinical settings. The approach uses a shallow decision tree interface with a concept bottleneck model, keeping explanations stable as the system learns from new data without forgetting.

arXiv cs.LG·Apr 18

Illustration for: EvoComp: Learning Visual Token Compression for Multimodal Large Language Models via Semantic-Guided Evolutionary Labeling

Research Models & Releases

EvoComp: Learning Visual Token Compression for Multimodal Large Language Models via Semantic-Guided Evolutionary Labeling

Researchers propose EvoComp, a token compression technique that cuts visual token counts in multimodal LLMs while maintaining accuracy, using an evolutionary labeling strategy to train a lightweight transformer compressor that jointly considers image and text context.

arXiv cs.LG·Apr 18

Policy & Regulation Business & Funding

Just ten minutes of using AI as an answer machine can measurably erode problem-solving skills, new study finds

Researchers from the US and UK found that 10–15 minutes of relying on AI to answer questions measurably reduces problem-solving ability and persistence on subsequent unaided tasks, suggesting cognitive dependency risks from casual AI use.

The Decoder·Apr 18

Tools & Code Research

Claude system prompts as a git timeline

Simon Willison converted Anthropic's published Claude system prompts into a git repository with timestamped commits, enabling visual tracking of how the prompts have evolved across model versions and families.

Simon Willison·Apr 18

Products & Apps Tools & Code

Schematik Is ‘Cursor for Hardware.’ Anthropic Wants In

Schematik, a tool designed to streamline hardware development through AI-assisted code generation, has attracted interest from Anthropic. The platform positions itself as an analog to Cursor, extending LLM capabilities to physical device programming.

WIRED — AI·Apr 18

Opinion & Analysis Business & Funding

Anthropic CEO Amodei declares "there is no end to the rainbow" for AI scaling

Anthropic CEO Dario Amodei argues AI scaling has no foreseeable limits and calls on the industry to address job displacement by ensuring economic gains outweigh disruption rather than downplaying the risks.

The Decoder·Apr 18

Tools & Code Opinion & Analysis

Adding a new content type to my blog-to-newsletter tool

Simon Willison demonstrates a practical agentic engineering pattern using a short LLM prompt to automate adding new content types to his blog-to-newsletter tool, showcasing how minimal prompting can accomplish substantial work in a single API call.

Simon Willison·Apr 18

Business & Funding Products & Apps

Sam Altman’s project World looks to scale its human verification empire. First stop: Tinder.

Sam Altman's World, an identity-verification startup using biometric Orbs, is expanding partnerships to scale its platform, with Tinder reportedly among its first commercial targets.

TechCrunch — AI·Apr 17

Business & Funding Products & Apps

Andreessen Horowitz, Thrive Capital to Co-Lead Cursor Investment at $50 Billion Valuation

Andreessen Horowitz and Thrive Capital are co-leading a $2 billion funding round in coding assistant Cursor at a $50 billion pre-money valuation, with Nvidia also participating. The investment marks a significant milestone for the AI-powered developer tool, which both VCs backed in 2024.

The Information — AI·Apr 17

Policy & Regulation

Lawyers for Musk Propose Dropping Fraud Claims Against OpenAI

Musk's legal team signaled they may abandon fraud allegations in their lawsuit against OpenAI over the company's departure from its nonprofit mission. The move would align Musk's case with prior court filings and potentially narrow the scope of claims.

The Information — AI·Apr 17

Business & Funding

OpenAI’s former Sora boss is leaving

Bill Peebles, leader of OpenAI's Sora video generation team, is departing the company following OpenAI's decision to discontinue the Sora project last month. The exit reflects OpenAI's broader strategic shift to eliminate lower-priority initiatives.

The Verge — AI·Apr 17

Business & Funding

Meta Eyes Layoffs of 10% of Company Starting in May

Meta plans to cut approximately 8,000 employees—roughly 10% of its global workforce—beginning May 20, with additional layoffs anticipated later in 2026. The move signals continued workforce restructuring at the social-media giant.

The Information — AI·Apr 17

Products & Apps Business & Funding

Should you stare into Sam Altman’s orb before your next date?

Tinder is partnering with Worldcoin to offer users five free app boosts in exchange for biometric verification via Worldcoin's facial-scanning orbs, expanding the orb network's real-world adoption beyond crypto and financial services.

The Verge — AI·Apr 17

Business & Funding

Kevin Weil and Bill Peebles exit OpenAI as company continues to shed ‘side quests’

Kevin Weil and Bill Peebles have departed OpenAI as the company discontinues Sora and dissolves its science team, marking a strategic shift from consumer-focused experiments toward enterprise AI solutions.

TechCrunch — AI·Apr 17

Business & Funding

Kevin Weil Among Executives Leaving OpenAI

Kevin Weil, former Meta and Twitter executive and member of OpenAI's management team, has announced his departure from the AI firm. His exit marks at least the third senior leader to leave OpenAI in recent days, signaling potential organizational shifts at the company.

The Information — AI·Apr 17

Models & Releases Policy & Regulation

Anthropic’s new cybersecurity model could get it back in the government’s good graces

Anthropic's new cybersecurity-focused model, Claude Mythos Preview, may ease tensions with the Trump administration after weeks of public criticism. The release signals potential rapprochement between the AI company and government officials concerned about national security.

The Verge — AI·Apr 17

Business & Funding

OpenAI Executive Kevin Weil Is Leaving the Company

Kevin Weil, former Instagram VP, is departing OpenAI after leading an AI science application that the company is now consolidating into Codex. The departure marks a shift in OpenAI's organizational structure as it refocuses product priorities.

WIRED — AI·Apr 17

Business & Funding

Sources: Cursor in talks to raise $2B+ at $50B valuation as enterprise growth surges

Cursor, an AI-powered code editor, is in advanced fundraising talks to secure $2B+ at a $50B valuation, with a16z and Thrive Capital expected to lead the round as the company capitalizes on surging enterprise adoption.

TechCrunch — AI·Apr 17

Hardware & Infra Business & Funding

Meta Paid Broadcom $2.3 Billion in 2025

Meta disclosed it paid Broadcom $2.3 billion in 2025 for custom AI chip design assistance, marking a rare public window into how major tech firms fund semiconductor development for AI infrastructure.

The Information — AI·Apr 17

Opinion & Analysis Tools & Code

‘Tokenmaxxing’ is making developers less productive than they think

Developers optimizing for token efficiency may be counterproductively increasing costs and maintenance burden, according to TechCrunch analysis. The practice generates more code that requires extensive refactoring, offsetting perceived productivity gains.

TechCrunch — AI·Apr 17

Research

Geometric regularization of autoencoders via observed stochastic dynamics

Researchers propose a geometric regularization method for autoencoders that better preserves tangent-bundle structure when learning reduced models from high-dimensional dynamical systems. The approach uses ambient covariance to constrain latent geometry and consistency, addressing limitations of existing chart-based and autoencoder methods for metastable systems.

arXiv cs.LG·Apr 17

Research Models & Releases

Evaluating the Progression of Large Language Model Capabilities for Small-Molecule Drug Design

Researchers introduced a chemically-grounded benchmark suite for evaluating LLMs on drug discovery tasks, formulated as RL environments across molecular property prediction and design. Frontier models show growing proficiency but substantial gaps remain, particularly in low-data experimental scenarios.

arXiv cs.LG·Apr 17

Older stories →