Capturing interesting articles, videos, papers, and projects. Enjoy!
- Overview of Google Antigravity, featuring AI control of the browser, artifacts, and agent orchestration
- Author: Google Antigravity on YouTube
- Experimented with Google Gemini 3 with Nano Banana to create website favicons.
- Images generated with 'Thinking with 3 Pro' followed the prompt better than those generated with 'Fast' mode.
- Experimented with Bing Image Generator to create images using DALLE-3 and MAI-Image-1.
- DALLE-3 images looked more like drawings or digital design, while MAI-Image-1 looked more like Microsoft Clip Art.
- MAI-Image-1 was better at following complex prompts.
- Completed the AI-900: Microsoft Certified: Azure AI Fundamentals certification exam.
- Topics included AI/ML concepts and model types, responsible AI, and Azure AI/ML services.
- Paper Shows that video models can perform some zero-shot learning and reasoning tasks already, and may be on track towards general-purpose visual understanding.
- A collection of trending AI papers curated by Hugging Face and upvoted by the community.
- Hugging Face Daily Papers
- Paper Proposes extending reinforcement learning with verifiable rewards (RLVR) to domains that don't have well-defined boolean reward functions.
- Paper ChipNeMo implements domain adaptation techniques to adapt LLMs to chip design (domain-adaptive tokenization, domain adaptive pretraining, retrieval-augmented generation, etc.).
- Part of a project team training SLMs using ART (Agent Reinforcement Trainer)
- Relevant Links: Kaggle: CUREBench Competition, ToolUniverse: Scientific Model Context Protocol CURE-Bench Starter Kit Qwen-Agent TxAgent-T1 Model Qwen3 Models
- Paper Applies several ML models to predict employee attrition. Interesting insights and presents a good approach that could be leveraged in other domains
- Paper Efficient fine-tuning of large language models using low-rank adaptation (LoRA)
- Paper Using human feedback and reinforcement learning to improve model performance
- Paper Paper introducing GPT (Generative Pre-Training Transformer)
- Paper Paper introducing BERT (Bidirectional Encoder Representations from Transformers)
- Visualization of multi-headed attention and transformer computations
- Author Grant Sanderson on YouTube
- Paper Proposes a new architecture to reduce training and deployment costs of LLMs
- Paper arguing that small language models (SLMs) are more efficient for agentic AI than large language models (LLMs)
- Set up Claude Code to analyze and refactor code
- Useful Links: Claude Code Setup, GitHub CLI,
- Paper advocating for keeping chain of thought (CoT) in LLMs to improve interpretability and safety
- Paper introducing the Transformer architecture, an improvement over RNNs and CNNs
- Paper detailing Microsoft's BrainWave Neural Processing Unit (NPU) architecture
- Benefits of "tensor slices" (matrix mult blocks) in FPGAs to accelerate ML workloads
- PPA tradeoffs of Compute-RAMs in FPGAs
- Presenter: Lizy K. John on YouTube
- Used Python-SDK (FastMCP) to create a Model Context Protocol (MCP) server for hardware design tools
- Model Context Protocol Documentation
- Open source tool to visualize the circuits of Large Language Models
- Author: Anthropic on Anthropic Research
- Neuronpedia, GitHub Repository
- Brief overview of MCP (Model Context Protocol)
- Author: IBM Technology on YouTube
- Google Gemini CLI is a command line interface providing features similar to GitHub Copilot (but via the terminal)
- Tested on removing a warning displayed in unit tests; not successful but the interface shows promise
- GitHub Repository
- High-level overview of Deep Learning concepts
- Instructor: Kumaran Ponnambalam on LinkedIn Learning (1hr 13m course)
- A brief intro to Reinforcement Learning
- Instructor: Khaulat Abdulhakeem on LinkedIn Learning (44m course)
- GitHub Copilot can be assigned tasks and will open Pull Requests (source)
- Pull Request: Add button to toggle display of collected coins
- Walks through using Python to build AI applications (1h 47m course)
- Packages: textblob, fastapi, langchain, streamlit
- Instructor: Priya Mohan on LinkedIn Learning (GitHub)
- Google Jules can create Pull Requests for GitHub Open Source projects (source)
- Pull request: Automated CI/CD for App Store Deployments with GitHub Releases
- Note: Code had several problems and needed to be reworked
- How Google DeepMind's AlphaFold advanced protein folding research
- Author: Veritasium on YouTube
- An overview and visualization of how neural networks work
- Author: 3Blue1Brown on YouTube
- Interview with Demis Hassabis, CEO of Google DeepMind, about the future of AI
- Author: 60 Minutes on YouTube
- Uploaded the schematics, specs, and YouTube videos to NotebookLM to create an antique radio knowledge base
- Used NotebookLM to developed a detailed repair guide for the antique vacuum tube radio
- Used Azure AI Document Intelligence to convert old type-written WWII documents into searchable text
- Uploaded the documents to NotebookLM to create a knowledge base
- Used the podcast feature to generate a podcast about the documents
- Article introducing the Azure Maia 100 AI accelerator chip (source)
- How Google DeepMind used AlphaTensor to discover a new matrix multiplication algorithm
- Author: Quanta Magazine on YouTube
- High level overview of Microsoft's AI supercomputer effort
- Author Microsoft Mechanic on YouTube
- Using Reinforcement Learning to train an AI to play Pokemon
- Author: Peter Whidden on Youtube
- Overview of Amazon's custom chip efforts for AI, including the Trainium and Inferentia chips
- Author: CNBC on YouTube
- Recent history of Nvidia's growth and its role in AI, including the A100 chip
- Author CNBC on YouTube
- Brief overview of Large Language Models and early models
- Instructor: Jonathan Fernandes on LinkedIn Learning (1h 22m course)
- Topics Covered: Large Language Models, Generative AI, Transformers, Attention
- Basic introduction to using GitHub Copilot
- Instructor: Ronnie Sheer on LinkedIn Learning (1h 17m course)
- Topics Covered: GitHub Copilot
- A step-by-step guide to Principal Component Analysis
- Author: StatQuest with Josh Starmer on YouTube
- Explanation of K-means clustering
- Author StatQuest with Josh Starmer on YouTube
- Used Machine Learning to recommend unit tests to run based on code changes
- Instructor: Jeff Prosise at Wintellect (GitHub)
- Topics Covered: k-means Clustering, k-nearest Neighbors, Random Forests, Gradient Boosting, Support Vector Machines, Naive Bayes Classifier, Principal Component Analysis, Convolutional Neural Networks
- Instructor: Wen-mei Hwu at University of Illinois at Urbana-Champaign on
CourseraArchive.org (9 week course) - Topics Covered: Heterogeneous Computing, Parallel Programming, Tiled Matrix Multiplication, CUDA
- Instructor: Andrew Ng at Stanford on
CourseraCoursera (10 week course) - Topics Covered: Supervised Learning, Unsupervised Learning, Linear Regression, Artificial Neural Networks, Backpropagation, Support Vector Machines, Kernel Methods, Clustering, Principal Component Analysis, Anomaly Detection, Recommender Systems