Scott Baggett’s Post

We decided to spend the weekend seeing how quickly we could build a clone of Perplexity, the AI search engine, on Salt. Then we open sourced it. Last week we released the public beta of Salt AI. The driving idea behind Salt, which lets users build, test, and deploy powerful workflows on an elastic GPU cloud, has always been to accelerate AI development. By that, we mean providing a development environment and open ecosystem that makes it easier and faster to go from idea, to proof-of-concept, to scalable, production-ready app. Our most active users, of course, are the members of our own team, who are always exploring the next exciting use case for the Salt platform. In that pursuit, we decided to spend the weekend seeing how quickly we could build a clone of Perplexity, the AI search engine. Like many others, we’re excited about Perplexity challenging Google in the search space by delivering a fantastic LLM-powered search experience. Perplexity recently raised $250M on an extremely impressive $3B valuation. We’re big fans of their product and agree that the best way to search in 2024 is through a conversational interface. But it’s not just Perplexity — a revolution is underway when it comes to the rapid development of AI-powered applications that can fetch enormous valuations and deliver huge value to their customers. The faster that innovators can ideate, prototype, iterate, and collaborate, the more we will see highly specialized and targeted AI solutions – with applications in healthcare, finance, logistics, commerce, and beyond – come from anywhere. We believe these solutions will be built on Salt. So we accepted our own challenge and set out to build our Perplexity clone over the weekend (spoiler alert…we did). Then we open sourced it to see what the community does next. Check out the video below to see how we did it. https://lnkd.in/gqgMqn5h

Using the Chrome browser is about to get a little easier and a little more efficient thanks to the addition of three new #AI-assisted features set to roll out soon. https://lnkd.in/exqDFzHY

People often say true learning happens through implementation, and I couldn't agree more. While I've explored Gemini on Android, web, and even the LLM interface On-Device, recently, I delved into the exciting world of Chrome extension development. The result? My very first Chrome extension, designed to offer valuable insights into Google Sheet data, is nearing launch on the Chrome Web Store! Here's the best part: it'll be completely free and ad-free, making it accessible to everyone. #chromeextension #learningbydoing #googleapps #ai #freetools #gemini

Back with part 2, Yujian Tang covers how to build RAG apps with Milvus, #mixtral hosted through OctoAI, and LangChain. Follow along with the journey: https://lnkd.in/eQ_VrciZ

Using the Chrome browser is about to get a little easier and a little more efficient thanks to the addition of three new AI-assisted features set to roll out soon. https://lnkd.in/gGT_ZhcH

Fixkey is #1 on Product Hunt! 🚀 This is my third launch, but the first time leading, let's hope we can keep it up to become product of the day 🥇 I'm blaming it on the incredibly talented team, it's been a pleasure working with Raphael Volpert and Robert L.. AI will be integrated into every app and every OS in the future. While we wait for that, Fixkey is trying a new approach: A native macOS app that changes the way you type! Just one shortcut and Fixkey automatically selects the current sentence and fixes spelling and grammar. In every app. No looking back, just type as fast as you can and GPT cleans up after you. https://lnkd.in/eTKZ3iKq

Just explored an amazing GitHub repo: LangChain Experiments. It's a treasure trove for anyone diving into the world of Large Language Models (LLMs) and their practical applications. LangChain is revolutionizing how we use AI, making complex LLMs like OpenAI GPT more accessible and applicable in real-world scenarios. This repository is a perfect blend of theory and practice, and it's super inspiring for AI enthusiasts and professionals. Highly recommend checking it out for some insightful AI experiments and use cases! Check Out his YouTube Channel https://lnkd.in/gYH_EdvR Dave Ebbelaar #AI #MachineLearning #LangChain #OpenAI #GPT4 #Innovation

🤔 Pondering the AI Tools That Shaped one-minute-changelog.com As I mentioned in my last post, I wanted to share some thoughts on the AI tools I've been using while building my side project. 📐"Look before you leap": It's always valuable to assess the initial cost of LLM usage. An instrument that shows the number of tokens per request can be very helpful! One of the great ones is the OpenAI Tokenizer (https://lnkd.in/e7p2qpeu). It can save you from a few "oops" moments. It's especially relevant for previous models with high pricing (gpt-4 and Claude 3 Opus). 💸 "A penny saved is a penny earned": If your analysis shows that you need to cut costs, don't hurry up to choose a worse model. Probably some preprocessing can save you money. In my case, I found great library cheerio.js.org to work with a raw HTMLs. It's like a Swiss Army knife for HTML parsing! I'm honestly amazed at how easy it extracts the content I need. It helped me keep LLM costs down by cleaning up the data first. 📈 "Bite the bullet": Once you're good with estimated costs and results, you start to deploy your model. I found a cool instrument called lunary.ai that helps with monitoring deployed LLM apps. I'm still experimenting with it, but the start was promising. #AITools #TokenCalculator #Cheerio #Lunaryai #LLMChallenges

If you find yourself using Bing Copilot often (as I do) and you want more than the side-panel view (plus perhaps you want it more readily accessible) you can turn it into an app with Edge. WAIT! It's easy! Step 1 - Go to https://bing.com/chat. Step 2 - Press ALT+F (or click the "Settings and more" button at the top left; it looks like 3 dots) Step 3 - Select "Apps" Step 4 - Select "Install this site as an app" Step 5 - Give it a name (or accept the default) and select Install. Now you can pin it to your taskbar and it's always available to you even if your browser isn't open.

I'm excited to share a powerful Generative AI application that allows chatting and extracting information from multiple PDF files, powered by Gemini Pro. This tool is incredibly useful for researchers and professionals to easily gather information from PDFs. It's amazing how generative AI has simplified building such apps with minimal code. 🔗 Check out the code repository: https://lnkd.in/d9wgu7ME 🌐 Live app: https://lnkd.in/dXWZ4kjQ TechnicalDetails: This app works by converting text from PDFs into embeddings using Google Gemini Pro, storing these embeddings in a FAISS index. When a user inputs a question, the app performs a similarity search on this index to find relevant text chunks, then uses a conversational AI model to generate detailed answers based on the context.