uBlock Origin is no longer available on the Chrome Store

uBlock Origin is a free, open-source content blocker that is efficient on CPU and memory, blocking a wide range of content including ads and trackers with various filter lists available. The extension is completely free, with no data collection, and offers advanced features such as point-and-click blocking of JavaScript and custom rule creation, all while being maintained by a community of contributors.

Read more

A bear case: My predictions regarding AI progress

The author expects that current avenues of AI capability advancement, such as pretraining and test-time compute/RL, will not be sufficient to achieve Artificial General Intelligence (AGI), and that progress will decouple from intuitive measures of intelligence, facing steep diminishing returns. The author predicts that future models, such as GPT-5 and GPT-6, will bring subtle quality-of-life improvements but no paradigm-shifting advancements, and that LLMs are currently good at solving eisegesis-friendly problems and in-distribution problems, but struggle with more complex tasks.

Read more

Show HN: In-Browser Graph RAG with Kuzu-WASM and WebLLM

The Kuzu database team has developed a WebAssembly (Wasm) version of their database, which allows for fully in-browser applications, and has demonstrated its potential by building a chatbot that answers questions over LinkedIn data using Graph Retrieval-Augmented Generation (Graph RAG). The chatbot uses Kuzu-Wasm and WebLLM, a popular in-browser LLM inference engine, to retrieve relevant data from a graph database and generate natural language answers to user queries.

Read more

Llama.cpp AI Performance with the GeForce RTX 5090 Review

The NVIDIA GeForce RTX 5090's AI performance was tested using Llama.cpp, with the graphics card showing promising results compared to prior RTX 40 and RTX 30 graphics cards. The tests were conducted using the NVIDIA 570.86.10 Linux driver on Ubuntu 24.10 with the Linux 6.11 kernel, and included comparisons with several other high-end GeForce RTX graphics cards.

Read more

Generative AI Hype Peaking

The author has been skeptical of exaggerated claims about Generative AI's potential to revolutionize labor productivity, and believes that investor hype around the technology is nearing its peak. While Generative AI has led to real process innovations, such as improved software development and customer support, its impact may be more limited than predicted, and could potentially lead to a degraded customer experience and reduced job opportunities for less experienced developers.

Read more

Probabilistic Artificial Intelligence

Artificial intelligence refers to the development of systems that can perform tasks requiring human-like intelligence, such as playing games, translating languages, and driving cars, with recent advances in machine learning and deep learning enabling unprecedented capabilities. This manuscript focuses on "Probabilistic Artificial Intelligence", covering probabilistic approaches to machine learning and sequential decision tasks, including topics such as uncertainty, active learning, Bayesian optimization, and reinforcement learning.

ArXiv

A Platform for Dialogue-Based Computational Problem Solving with Generative AI

Computing students are increasingly using generative AI tools for programming assistance, highlighting the need to teach effective interaction with AI models through natural language prompts. A novel platform for prompt programming was developed and tested with over 900 students, showing high engagement, iterative refinement, and critical thinking, and is now being made available as a publicly accessible resource.

ArXiv

Generalized Interpolating Discrete Diffusion

Researchers have developed a new approach called general interpolating discrete diffusion (GIDD) to overcome the limitations of current language models, which cannot revise already generated tokens. The GIDD approach achieves state-of-the-art performance and allows for the creation of models that can correct their own mistakes, a significant improvement over traditional autoregressive models.

ArXiv

NoLiMa: Long-Context Evaluation Beyond Literal Matching

Recent large language models (LLMs) struggle to retrieve relevant information from long contexts, with performance degrading significantly as context length increases, despite their claimed support for long contexts. The NoLiMa benchmark, which requires models to infer latent associations to locate relevant information, reveals that even top-performing models like GPT-4o experience significant declines in performance when literal matches are absent and context length exceeds 32K tokens.

ArXiv

Someone Shouts, "|01000⟩ " Who Is Excited?

The correct ordering of amplitudes in a wavefunction is crucial, and understanding how to interpret operator matrices is essential in quantum computation. Proper interpretation of these concepts allows for accurate representation and manipulation of quantum states, which is fundamental to quantum computing and information processing.

ArXiv

Show HN: Tesseract Core – Autodiff-native, self-documenting software components

Tesseract Core is a software framework that provides tools to define, create, and run self-contained, self-documenting components called Tesseracts, which support differentiable programming and can be easily shared and executed. The framework includes a command-line interface, a Python API, and a lightweight runtime environment, and is licensed under the Apache License 2.0, allowing for free use, modification, and distribution.

Repo

Awesome Web Agents: A curated list of AI agents that browse the web

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

Repo

A CLI tool for security testing agent workflows

Agentic Radar is a security scanner designed to analyze and assess agentic systems for security and operational insights, providing a comprehensive HTML report that includes workflow visualization, tool identification, and vulnerability mapping. The tool is intended for developers, researchers, and security professionals to simplify the process of understanding complex agentic system workflows and identifying potential vulnerabilities.

Repo

Show HN: V0.6.0 Update: Dive – An Open Source MCP Agent Desktop

Dive is an open-source desktop application that integrates with various large language models (LLMs) and provides features such as universal LLM support, cross-platform compatibility, and advanced API management. The application offers a range of tools and capabilities, including model context protocol, MCP server integration, and multi-language support, with new features and updates being added regularly, including recent additions such as SSE transport protocol and auto-update mechanism.

Repo

DNS Validator Written with AI

The DNS Validator is a Ruby tool that checks DNS records in a BIND format file against actual DNS lookups to ensure correct configuration and propagation, supporting various record types and offering features like parallel lookups and customizable validation options. It can be installed by cloning the repository and running the script with Ruby, and offers various command line options for customization, such as verbose output, case-sensitive matching, and cache control.

Repo