Sunday March 30, 2025

Apple's AI faces criticism despite delays, Ghost in the Shell lets users execute shell commands via natural language, and Orion framework enhances security in deep learning with Homomorphic Encryption.

News

Apple's AI isn't a letdown. AI is the letdown

Apple's foray into artificial intelligence has been met with criticism, with the company delaying its AI-powered Siri release and rolling out unhelpful AI features. However, the issue lies not with Apple, but with the concept of AI itself, which is being pushed by Wall Street investors seeking a new "super cycle" of tech upgrades, despite a lack of clear problems to be solved or customer demand.

Show HN: Appear as anyone in video calls like zoom or Google meets

You can transform into your favorite anime character, celebrity, or creation using a single reference photo with this software, which runs privately on your device and is compatible with various video apps. The software is currently available for Linux, with Windows and Mac versions coming soon, and requires a device with specific system requirements, including an NVIDIA GPU with CUDA support.

The Matrix Calculus You Need for Deep Learning

This paper aims to explain the matrix calculus required to understand the training of deep neural networks, assuming only a basic knowledge of calculus. It covers key concepts such as vector calculus, partial derivatives, and the chain rule, and provides a reference section summarizing the main matrix calculus rules and terminology. The goal is to help readers deepen their understanding of the underlying math behind deep learning, particularly for those already familiar with the basics of neural networks.

Nick Cave's thoughts on a ChatGPT poem (2023)

Nick Cave was sent a song written by ChatGPT in his style, but he was unimpressed, feeling that the song was a poor replication of his work and lacked the authentic human experience and emotional depth that a great song requires. Cave believes that AI systems like ChatGPT are limited by their inability to feel and suffer, and are therefore incapable of creating truly original and meaningful music, instead only able to mimic and replicate existing styles.

AI Accent Conversion for call centers

Krisp offers AI-powered solutions for call centers, including accent conversion, noise cancellation, and live interpretation, to enhance communication and improve customer experience. The company's technology has been adopted by leading global contact centers, resulting in benefits such as increased revenue, improved agent performance, and enhanced customer satisfaction.

Research

Matrix Calculus (For Machine Learning and Beyond)

This course extends differential calculus to functions on general vector spaces, covering topics such as matrix functions and stochastic derivatives, with a focus on practical applications in optimization and machine learning. It also introduces techniques like adjoint differentiation and automatic differentiation, which are essential for efficient computation in complex calculations, particularly in areas like large-scale optimization and machine learning.

Orion: A Homomorphic Encryption Framework for Deep Learning [pdf]

Fully Homomorphic Encryption (FHE) has the potential to improve privacy and security by enabling computation on encrypted data, particularly in deep learning applications. The Orion framework addresses the challenges of implementing FHE-secured neural inference by automatically translating PyTorch neural networks into efficient FHE programs, achieving state-of-the-art performance and enabling the processing of larger and deeper networks.

LLM Interactive Optimization of Open Source Python Libraries (2023)

This paper presents case studies on utilizing large language models, specifically ChatGPT-4, for source code optimization in open-source Python libraries like Pillow and NumPy, finding significant performance improvements of up to 38 times. The results show that while LLMs are a promising tool for code optimization, a human expert is essential for success, and the study highlights the need for more robust quantitative studies to further explore this potential.

Gen AI for Artistic Style Transfer with Convolutional Neural Networks (2023)

Artistic style transfer is a technique that combines the content of one image with the style of another using generative artificial intelligence, resulting in unique visual compositions. This technique utilizes Convolutional Neural Networks (CNNs) to separate and manipulate image content and style, allowing for the creation of high-quality images that blend content and style in a harmonious way.

Self-Supervised Learning from Images with JEPA (2023)

The Image-based Joint-Embedding Predictive Architecture (I-JEPA) is a self-supervised learning approach that learns semantic image representations without relying on hand-crafted data augmentations. I-JEPA has been shown to be highly scalable, particularly when combined with Vision Transformers, achieving strong performance across various tasks such as linear classification, object counting, and depth prediction.

Code

Show HN: Job Application Bot by Ollama AI

JobHuntr.fyi is a macOS desktop app that uses AI to automatically apply for jobs on LinkedIn 24/7, with features such as auto-apply, smart filtering, and custom filters. The app offers a free plan with limited features and a premium plan for $9.99/month, and it is designed to mimic human behavior to avoid detection by LinkedIn, with all AI processing done locally on the user's device.

Show HN: Use Natural Language to Run Terminal Commands (Ghost in the Shell)

Ghost in the Shell is a cross-platform command-line assistant that uses natural language to run shell commands, allowing users to type what they want to do and then suggesting and executing the correct command. The tool supports one-shot and interactive modes, works on macOS, Linux, and Windows, and features colored terminal output, confirmation before execution, and optional tmux split-pane workflow.

Show HN: MCP Link – Integrate Any API with AI Agents

MCP Link is a solution that seamlessly integrates any API with AI agents through automation and standardization, allowing for automatic conversion of APIs into MCP-compatible interfaces. It provides features such as automatic conversion, seamless integration, and zero code modification, making it easy to connect existing RESTful APIs with AI-driven applications.

Show HN: OpenAPI MCP Server – Make Cursor AI Understand Your API Specs

The @reapi/mcp-openapi server loads and serves multiple OpenAPI specifications, enabling LLM-powered IDE integrations by acting as a bridge between OpenAPI specs and development tools. It features loading multiple specs, exposing API operations and schemas, and maintaining a catalog of available APIs, with configuration options for project-specific or global setup to integrate with Cursor IDE.

Reflect: An AI tool to generate your brag document

Reflect is a Node.js tool that generates a reflection document of a user's GitHub activity, collecting and formatting merged pull requests and closed issues, while also providing optional AI-powered summary and brag document generation. The tool can be run with a GitHub Personal Access Token and an OpenAI API key, and generates markdown files in the output directory, including a detailed list of contributions, a summarized version, and a professional brag document.