Show HN: I made a free tool that analyzes SEC filings and posts detailed reports

Several major companies, including Morgan Stanley, JPMorgan Chase, and BlackRock, reported strong first-quarter 2025 results, with many surpassing analyst expectations due to robust performance in key divisions and record revenue. However, some companies, such as Wells Fargo and AGNC Investment Corp., reported mixed results, with earnings beating expectations but revenue or book value falling short.

Read more

AMD NPU and Xilinx Versal AI Engines Signal Processing in Radio Astronomy (2024) [pdf]

Researchers from various institutions explored the use of Versal AI Engines for real-time signal processing in radio astronomy, specifically for the LOFAR radio telescope. They developed and implemented a Polyphase Filter Bank on the Versal ACAP, which combines computing power and energy efficiency, to process large amounts of data in real-time.

Read more

AI companions are final stage of digital addiction, and lawmakers are taking aim

Lawmakers are taking aim at AI companions, which are designed to be highly engaging and personalized, as they pose a significant risk of digital addiction, particularly among children. With millions of users spending hours interacting with these AI models, which can learn about and adapt to individual users, there are growing concerns about their potential impact on mental health and wellbeing, prompting calls for stricter regulations and safeguards to protect vulnerable users.

Read more

Show HN: Real-time emergency incident maps from police/fire/EMS radio

There were multiple reports of shots fired, persons with guns, and assaults in Philadelphia County, Pennsylvania, on April 14, 2025, with incidents occurring throughout the early morning hours. The reports included various descriptions of individuals and vehicles involved, as well as responses from police and emergency services to address the incidents, which also included burglaries, medical emergencies, and structure fires.

Read more

Hinge AI Matchmaker

This dating service is trusted by over 2,000 busy professionals and has a 4.9-star rating, saving users 10+ hours per week and making dating feel fun again. Users have reported significant results, including 15 matches in just 3 days, with some even finding relationships or getting married, and feeling more in control of their dating lives.

Read more

NoProp: Training neural networks without back-propagation or forward-propagation

The traditional deep learning approach involves back-propagating error signals to learn hierarchical representations, but a new method called NoProp learns without relying on forward or backward propagation. NoProp instead uses a diffusion-based approach, where each layer learns to denoise a noisy target, and has been shown to be effective and efficient on several image classification benchmarks, offering a viable alternative to traditional gradient-based learning methods.

ArXiv

Exact Solution of the Frustrated Potts Model: An AI-Aided Discovery

The one-dimensional $J_1$-$J_2$ $q$-state Potts model has been exactly solved for arbitrary $q$ using the maximally symmetric subspace method, which simplifies the transfer matrix into a $2\times 2$ matrix. This solution provides insights into physical problems such as atomic order stacking in layered materials and phase formation in unconventional superconductors, and is expected to advance research in one-dimensional frustrated magnets and AI applications in science.

ArXiv

Robustly identifying concepts introduced during chat fine-tuning with crosscoder

Model diffing, which studies how fine-tuning changes a model's representations, can be improved with the crosscoders method, but this method has issues that can misattribute concepts as unique to the fine-tuned model. The introduction of Latent Scaling and a new BatchTopK loss function helps mitigate these issues, allowing for the identification of genuinely chat-specific and interpretable concepts, such as false information and personal questions, in fine-tuned language models.

ArXiv

A Foundational Theory for Decentralized Sensory Learning

The brain can be viewed as a negative feedback control system, where sensory signals are used to minimize activity, allowing for local learning algorithms without global error correction metrics. This learning principle, which originated in unicellular life forms, has been scaled up and is proposed to be a fundamental aspect of learning in biological brains, from simple organisms to humans.

ArXiv

Our quantum assembly parser got updated to the QASM 3.0 spec

The OpenQASM 3.0 specification introduces unique features that require new approaches to quantum program verification, motivating the development of verification and validation software. A typed QASM 3.0 parser has been implemented in TypeScript, and a formal treatment of OpenQASM 3.0's type system is proposed to further facilitate formal verification in hybrid quantum-classical computing.

ArXiv

Show HN: Navigator Mode (Like Claude Code) for Aider

Aider is a tool that enables pair programming with large language models (LLMs) in your terminal, allowing you to start new projects or build on existing codebases. It offers various features, including support for multiple LLMs, code mapping, Git integration, and compatibility with over 100 programming languages, making it a powerful assistant for coding tasks.

Repo

Show HN: Opsmate - LLM Powered SRE Assistant

Opsmate is an AI-powered SRE teammate that assists with production engineering tasks, offering features such as natural language interfaces, advanced reasoning, and multiple runtime support. It can be installed using various methods, including pip, Docker, and uv, and provides a web UI, API, and plugin system for customization and extension.

Repo

claude-mcp: Browser extension to enable MCP in claude.ai

The Claude.ai MCP browser extension enables Model Control Protocol capabilities, allowing users to connect Claude.ai to external tools and services directly from the browser. The extension offers features such as managing multiple server connections, configuring environment variables, and debug logging options, and can be installed from source or used in development mode with hot reloading.

Repo

Show HN: A toy MCP for AI agents to code, run, and see output of GPU code safely

There is no text to summarize. The provided input appears to be an error message indicating that a README file could not be retrieved.

Repo

Show HN: PlayBook AI – Basketball Analysis

PlayBook AI is a basketball intelligence system that analyzes offensive and defensive plays, assesses player fit, and identifies successful plays through simulation, object tracking, and neural network-based classification. The system is built using Python, utilizes spatial databases and GPU acceleration, and can be deployed on AWS with Docker containerization.

Repo