Nik Bear Brown’s Post

Creating a beautiful terminal application, GUI, or CLI is good, but capturing a high-fidelity demo of it is even better. You may have used asciinema or screen in the past, but I'm always surprised how few folks know about VHS. VHS is an open-source tool that allows you to write terminal GIFs as code for seamless integration testing and stunning demos. With VHS, you can create pixel-perfect recordings of your CLI tools in action, making it easier than ever to showcase your work. 🎬💻 One of the standout features of VHS is its intuitive and expressive syntax. You define your recordings using a simple yet powerful .tape file format, specifying commands, typing actions, and custom settings like font size, window dimensions, and themes. It's like having a virtual director for your terminal demos! 🎨🎛️ You can also record your actions in the terminal and have your .tape file generated. Removing typos or awkwardness is as simple as editing your .tape file and regenerating your gif. VHS allows you to output in multiple formats, including .mp4 and .gif, even simultaneously. This means you can wire VHS up to your CI/CD process to get excellent recordings or create build artifacts to review as sanity checks that everything is working properly. VHS also offers a wide range of commands to enhance your recordings. You can simulate typing with the Type command, navigate with arrow keys, use special keys like Enter and Tab, and even incorporate dramatic pauses with the Sleep command. VHS provides a built-in SSH server, allowing you to access VHS remotely and leverage the host's commands and applications. You can create demos on your machine and seamlessly share them with others. 🌐🔑 Integration with continuous integration pipelines is a breeze with VHS. You can keep your GIFs up-to-date using the official VHS GitHub Action, ensuring that your demos always reflect the latest changes in your codebase. 🔄✅ VHS is your tool if you want to take your terminal demos to the next level. It's open-source, extensively documented, and supported by a vibrant community. Try it and experience the magic of creating stunning terminal recordings with ease! ✨💻 And be sure to follow me and subscribe to my newsletter for more tips and open-source tool spotlights like this!

If you've seen my video (previous post) on using CustomGPT to identify design patterns in a given code snippet, here are my insights on its implementation: Link : https://lnkd.in/da47czC9

it is important to know the entire product design and development cycle.

Delving into the frontier of web development with an intriguing paper: "How Far Are We From Automating Front-End Engineering?" 🚀🤖 This research breaks ground in Generative AI, introducing a Design2Code task and benchmarking 484 real-world webpages. 🎨💡 GPT-4V steals the spotlight, showcasing its prowess in generating visually stunning and content-rich webpages. 🌟 The future is here—explore the possibilities! #FrontEndInnovation #AIRevolution #TechBreakthrough 🚀🌐

Want to clone your favorite landing page design? Claude's new Artifacts feature lets you turn screenshots into code in < 5 minutes. Here's a step-by-step tutorial: --- And if you like carousels like this: Follow me Ben Tossell for tips and tutorials on how to use AI to work smarter, not harder.

Check out this simple guide on how you can setup fully-managed RAG pipeline in minutes using Laminar AI (lmnr.ai) -> https://lnkd.in/eZbsaWAw Laminar AI is an infrastructure-first approach to building LLM pipelines. You define LLM pipelines as graphs in seconds, and get extremely fast Rust infrastructure, observability and evaluations out of the box.

At some point the context window will become limitless. In fact, MY PREDICTION IS BY THE END OF THE YEAR, CONTEXT WINDOWS WILL BECOME PRACTICALLY LIMITLESS BECAUSE OF THE USE OF AI AGENTS. Right now I can load a million tokens worth of environmental data equating to about an hour of video, 11 hours of audio, 30,000 lines of code and 750,000 words into Gemini. And as awesome as this is . . . the AI architecture is metamorphosing from the transformer to a transform-agent architecture. The difference? Night and day. Imagine loading up effectively trillions of tokens worth of environmental data with the help of a new architecture that employs active reasoning through an iterative agent workflow? If you thought it already was daytime . . . get ready to put on your sunglasses. It's going to get bright around here! :) For information about EnviroAI and it's use of artificial intelligence to better protect the planet, see www.enviro.ai.

🗺️ Easy guide to learning different Prompt Engineering techniques, Embedding Models (Gemini, OpenAI, Hugging Face) and Advanced RAG in Generative AI. 📚 Great for anyone wanting to learn GenAI with simple code. 🔍 Find the flow and code on GitHub and Google Colab. Prompt Engineering - https://lnkd.in/enwiPUxs How to choose right Embedding Model - https://lnkd.in/ehAhAeZY Implement Advance RAG (Download the PDF—includes a link to the code ) - https://lnkd.in/eVaXAW8Z #genai #generativeai #promptengineering #rag #embeddingmodels #llm #learning #coding

"The input embedding in LLM models maps the token id in the vocabulary to the corresponding token embedding and has a dimension of (vocab_size, embedding_dim). Conversely, the output fully-connected layer maps the embedding dimension back to the logits prediction across the vocabulary, with weight size of (vocab_size, embedding_dim). By sharing the embedding, we reuse the input embedding weights as the output fully connected layer weights, resulting in a more efficient and compact model architecture." cute: https://lnkd.in/gMDcXb8r