Ping, You've Got Whale: AI detection system alerts ships of whales in their path

WhaleSpotter is an artificial intelligence-powered whale detection system that uses heat-sensing cameras and machine learning to identify whales and alert ships to prevent collisions, with the goal of reducing the thousands of whale deaths that occur each year due to maritime accidents. The system, which has been tested on various types of vessels and has made over 51,000 marine mammal detections, is being adapted for use on container ships, one of the main culprits behind whale deaths, and its developers hope to expand its use to hundreds of vessels to maximize its impact.

Read more

A weird phrase is plaguing scientific papers

The term "vegetative electron microscopy" is a nonsensical phrase that originated from a scanning error and translation mistake, and has since been perpetuated and amplified by artificial intelligence (AI) systems, appearing in 22 papers and becoming a permanent fixture in our knowledge ecosystem. This "digital fossil" highlights the challenges of error correction in AI systems and the potential for AI to contaminate our collective knowledge, raising concerns about the integrity of research and publishing in the age of AI-assisted writing.

Read more

AI for Network Engineers: Understanding Flow, Flowlet, and Packet-Based LB

Traditional flow-based load balancing with Layer 3 ECMP is not suitable for RoCEv2-based AI backend networks due to the massive elephant flows created by GPU-to-GPU communication, which can cause congestion and uneven bandwidth usage. Alternative load balancing methods, such as Flowlet-Based Load Balancing with Adaptive Routing and Packet-Based Load Balancing with Packet Spraying, are being introduced to improve traffic distribution and handle the unique traffic patterns of AI workloads more efficiently.

Read more

Forecaster reacts: METR's bombshell paper about AI acceleration

According to a recent study by METR, an AI evaluations organization, AI is accelerating quickly, with the potential to complete tasks that take humans one hour by the end of February 2025, two hours by the end of September 2025, and potentially all human labor by the end of 2031. However, the study's methodology, which focuses on software engineering tasks, may overestimate progress towards Artificial General Intelligence (AGI) as it ignores other domains and tasks that are more complex and messy, where AI may struggle more.

Read more

Introduction to Graph Transformers

Graph Transformers are a new class of models that overcome the limitations of Graph Neural Networks (GNNs) by using powerful self-attention mechanisms to capture richer relationships and subtle patterns in graph data. They enable each node to directly attend to information from anywhere in the graph, making them suitable for tasks such as protein folding, fraud detection, social network recommendations, and knowledge graph reasoning, and are expected to become indispensable for data scientists and ML engineers.

Read more

Creating benchmarkable components to measure the quality of AI-enhanced devtools

The AI community lacks a standardized approach to benchmarking products built on generative AI models, leading to a focus on model quality over developer experience and difficulty comparing products to competitors. This case study presents a process for creating modular components to benchmark the developer experience of AI-powered code products, aiming to make it easier for teams to evaluate and compare their offerings.

ArXiv

In between myth and reality: AI for math – a case study in category theory

Researchers have been studying how well AI systems perform in solving math problems, with varying results. This paper discusses an experiment using two prominent AI systems to explore their potential in assisting mathematical research and identify areas for improvement.

ArXiv

Machine learning with neural networks (2021)

The lecture notes cover a machine learning course on neural networks, divided into three parts: stochastic recurrent networks, supervised learning, and learning from unlabeled data sets. The course aims to explain the fundamental principles of neural networks, covering topics such as Hopfield networks, multilayer perceptrons, convolutional neural networks, unsupervised learning, and reinforcement learning, with a focus on common concepts and ideas throughout.

ArXiv

Should We Respect LLMs? A Study on Influence of Prompt Politeness on Performance

Researchers found that the level of politeness in prompts affects the performance of large language models, with impolite prompts often resulting in poor performance, but the optimal level of politeness varies across languages and cultural contexts. The study suggests that language models reflect human behavior and are influenced by cultural norms, highlighting the importance of considering politeness in cross-cultural natural language processing and language model usage.

ArXiv

Flat origami is Turing complete (2023)

Flat origami refers to the folding of flat paper into a two-dimensional object, and its mathematical representation involves a continuous mapping and layer ordering that tracks the position of points when folded. The study of flat origami has led to the discovery that it is Turing complete, meaning it can simulate any computational process, which was proven by showing that flat origami crease patterns can mimic a one-dimensional cellular automaton known as Rule 110.

ArXiv

Show HN: Morphik – Open-source RAG that understands PDF images, runs locally

Morphik is an open-source platform that provides tools for ingesting, searching, transforming, and managing unstructured and multimodal documents, offering features such as multimodal search, knowledge graphs, and fast metadata extraction. It has a free tier and can be self-hosted, with options for direct installation and installation via Docker, as well as a paid version with additional features like Morphik Console.

Repo

Show HN: I open-sourced my AI toy company that runs on ESP32 and OpenAI realtime

ElatoAI is a real-time AI speech project that enables uninterrupted global conversations for over 10 minutes, powered by OpenAI's Realtime API, ESP32, Secure WebSockets, and Deno Edge Functions. The project consists of three main components: a Next.js frontend, Deno Edge Functions, and an ESP32 Arduino client, allowing users to create custom AI agents and engage in conversations with them.

Repo

Show HN: Rowboat – Open-source IDE for multi-agent systems

Rowboat is a platform that utilizes AI to build multi-agent workflows in minutes, allowing users to create custom assistants by simply describing their idea and integrating them into their app using an HTTP API or Python SDK. With Rowboat, users can quickly create and deploy multi-agent workflows, such as a food delivery assistant, and access them through a web interface or programmatically using the provided APIs and SDKs.

Repo

Show HN: Happen We Discovered This Minimal Framework of Glue with Superpowers

Happen is a framework that provides a minimalist approach to building systems, using only two fundamental building blocks: Nodes and Events, allowing for radical simplicity, security, and flexibility. The framework is designed to be runtime-agnostic, with features such as autonomous nodes, flexible event buses, and built-in cryptography, making it suitable for a wide range of applications, from automation pipelines to complex distributed state management.

Repo

A cross-platform Markdown AI note-taking tool with only 13 MB

NoteGen is a cross-platform, AI-enhanced note-taking application that allows users to record and organize knowledge in a readable format, supporting multiple recording methods and real-time synchronization to private GitHub repositories. The application features a recording function similar to an AI chatbot, a writing section with a file manager and Markdown editor, and various auxiliary features such as tags, personas, and a clipboard assistant to enhance the note-taking experience.

Repo