AI & Technology Updates

  • LocalGuard: Auditing Local AI Models for Security


    I built a tool to audit local models (Ollama/vLLM) for security and hallucinations using Garak & InspectAILocalGuard is an open-source tool designed to audit local machine learning models, such as Ollama, for security and hallucination issues. It simplifies the process by orchestrating Garak for security testing and Inspect AI for compliance checks, generating a PDF report with clear "Pass/Fail" results. The tool supports Python and can evaluate models like vLLM and cloud providers, offering a cost-effective alternative by defaulting to local models for judgment. This matters because it provides a streamlined and accessible solution for ensuring the safety and reliability of locally run AI models, which is crucial for developers and businesses relying on AI technology.


  • HomeGenie v2.0: Local Agentic AI with Sub-5s Response


    HomeGenie v2.0: 100% Local Agentic AI (Sub-5s response on CPU, No Cloud)HomeGenie 2.0 introduces an advanced "Agentic AI" designed to operate entirely offline, leveraging a local neural core named Lailama to run GGUF models such as Qwen 3 and Llama 3.2. This system goes beyond typical chatbot functions by autonomously processing real-time data from home sensors, weather, and energy inputs to make decisions and trigger appropriate API commands. With an optimized KV Cache and history pruning, it achieves sub-5-second response times on standard CPUs, ensuring efficient performance without relying on cloud services. Built with zuix.js, it features a programmable UI for real-time widget editing, emphasizing privacy and independence from cloud-based solutions. This matters as it provides a robust, privacy-focused AI solution for smart homes, enabling users to maintain control over their data and operations.


  • Plaud’s New AI NotePin S & Desktop Notetaker


    Plaud launches a new AI pin and a desktop meeting notetakerPlaud has unveiled the Plaud NotePin S, a new AI-powered notetaker, and a desktop app designed to enhance digital meeting note-taking. The NotePin S, priced at $179, features a physical button for recording control, a range of accessories for versatile wearing options, and Apple Find My support for easy tracking. It offers 64GB of storage, 20 hours of battery life, and two MEMS mics for clear audio capture, although it has a shorter recording range and battery life compared to the Note Pro. The desktop app, compatible with various meeting platforms, uses AI to structure transcriptions and supports multimodal inputs, aiming to compete with other meeting notetakers like Granola and Fireflies. This matters because it highlights the ongoing innovation in AI-driven tools that enhance productivity and accessibility in both in-person and virtual meetings.


  • Train Models with Evolutionary Strategies


    Propagate: Train thinking models using evolutionary strategies!The paper discussed demonstrates that using only 30 random Gaussian perturbations can effectively approximate a gradient, outperforming GRPO on RLVR tasks without overfitting. This approach significantly speeds up training as it eliminates the need for backward passes. The author tested and confirmed these findings by cleaning up the original codebase and successfully replicating the results. Additionally, they implemented LoRA and pass@k training, with plans for further enhancements, encouraging others to explore evolutionary strategies (ES) for training thinking models. This matters because it offers a more efficient method for training models, potentially advancing machine learning capabilities.


  • 15 Years of Evolving ML Research Notes


    [D] My Machine learning research notes: 15 years of continuous writing and 8.8k GitHub stars!Over 15 years of continuous writing and updates have resulted in a comprehensive set of machine learning research notes that have garnered 8.8k stars on GitHub. These notes cover both theoretical and practical aspects of machine learning, providing a dynamic and evolving resource that adapts to the fast-paced changes in the industry. The author argues that traditional books cannot keep up with the rapid advancements in machine learning, making a continuously updated online resource a more effective way to disseminate knowledge. This matters because it highlights the importance of accessible, up-to-date educational resources in rapidly evolving fields like machine learning.