Do you want a RAG to cry into? Jason Liu has a great article on how to build a fricken Terrible RAG (Retrieval Augmented System) system! FYI, you use a RAG when you want your LLM to stop hallucinating on magic matrix mush, and to back up their discourse with sources… in most cases, from a wiki you embed into a vector database. Trust me, this is way cheaper than finetuning your models to stop hallucinating. Check it out! 👏People: Jason Liu 📃Paper:
Marz Goldman’s Post
More Relevant Posts
-
So you’ve heard of prompt engineering, have you? Maybe giggled a bit at how trivial those 'AI influencer’s' prompt cheatsheets were... I know I have! Cos no matter what you tell the ‘monster’, it's going to behave badly in your B-B and commercial products, which is why we don’t trust it to chat with our critical customers, right? (Google’s management didn’t agree with me so they released Search Overview 🥲) Thank God for people like Aman Bhargava, Cameron Witkowski, Manav Shah and Matt Thomson. They’re taking a control theory approach to understanding and mapping the beast. I’ve attached a video where they break their strategy down, and it’s super entertaining, so take a watch! 🎥https://lnkd.in/gzMigX5B 📚https://lnkd.in/gYrxkMn4
To view or add a comment, sign in
-
An LLM as a compiler? Dave Hoover has a great introductory article on how he has a feeling that the next revolution in programming languages is fast approaching... something Cucumber-esc in our journey towards LLM OS (my cheeky addition). In his mind, the human programmer still exists as a separate entity in the software production chain, writing constraints applications are then generated from. Personally, I’m more inclined to believe that the human programmer would be ultimately replaced by models that generate applications optimised for an individual's behaviour and objectives... LLMs might be the wrong model type for this... its more like those Large Action Models that Rabbit Co. popularised. 📚https://lnkd.in/ghtAqGb8
LLM as Compiler
medium.com
To view or add a comment, sign in
-
So you think LLMs are relegated to just your conversation chatbot? Think again! I am convinced LLMs are the future of your computer operating system. LLM_OS:: What better way to bridge the gap between the ones and zeros on your PC architecture and the user-centric UX we all so desire!! Human Readable code is a dead end IMO. It takes up too many resources. We build our hardware architectures around making code human-readable. We spend ages coding up and designing UI and UX. Iterating on our mistakes is so fricken slow. In the future, LLMs will 'dream up’ UI and UX on the fly, optimised on hardware built for purpose. The top LLM_OS will be the one that serves up information in bite-sized packages the best for that particular user's purpose... think YouTube algorithm on steroids, for every single thing you see on your phone, or in your AR glasses. Mark my words. I had this thought for LLM_OS ages ago when reading Andrej Karpathy Medium's article on software 2.0. I started SumBuz.com to Sum the world's Buz for this future. Let’s jump on this train people and make this happenl!!! 📚Andrej’s article: https://lnkd.in/gYsDwq6S 📚Anshuman Mishra: https://lnkd.in/gjQCGHBu 📚João Galego paper on LLM OS: https://lnkd.in/gS-2c-w7
🧪 The Rise of the LLM OS: From AIOS to MemGPT and beyond
community.aws
To view or add a comment, sign in
-
I was 'this' days old when I discovered Brave Browser Apps. It all started when for the umpteenth time, I accidentally closed a critical tab in my browser with LibreChat running in it. In case you're wondering, LibreChat is an open-source GUI that allows me to run models from all my LLM providers... OpenAI, Anthropic, Google etc. via their cheap-as-chips APIs... all in one place. I can even get my LLMs to talk to each other which is just chaos! Anyway, we're not here to talk about LibreChat, we're here to talk about the fact that Brave allows you to click this little button in their browser and the web page you're on gets packaged up as a cute little application, that I can pin in my mac taskbar... like WTF, am I the last person to know that this is possible?!!!
To view or add a comment, sign in
-
Tremble at the MIGHT of my Obsidian Vault! - It looks and works great! - It has a local LLM (LLama3) that talks to my notes (Oh yeah, it can use chatgpt4 as well, if you have an OpenAI Key) - It has Google Search and Wikipedia for quick search - It has the Global News embedded in my notes - It has a bunch of templates to organise my thoughts - It has dedicated project spaces... I could go on forever! And it’s all thanks to: 👏 SoRobby for his Obsidian Started Vault https://lnkd.in/gRgwGXxs 👏 Long Huynh for his Obsidian Chatbot GUI that runs my Local Model LLama3 https://lnkd.in/gD8vG66A 👏 Kinmury for his theme Ukiyo, which is my go-to... with some style changes of course. 👏 Emile for his internet search capability within Obsidian https://lnkd.in/g9FZsdmP
To view or add a comment, sign in
-
Web3 is just another iteration of the cyclic argument for decentralised 🔄 centralised governance that’s been going on since time immemorial. Web3 promises privacy, just like Bitcoin, but what people don’t get is that every transaction a person makes is public on the blockchain, and if your digital ID is linked to your physical ID, every single person in the world will have an immutable record of that pair of sexy panties you bought for your wife ‘privately' 15 years ago. More work is needed to make Web3 less trojan horsey
We all highly value our privacy, safety, and security, especially when using the internet, right? With every click and keystroke, we're leaving behind a digital trail that could be exploited. That's why it's essential to protect your business. I believe that the concept and innovations of Web3 will benefit us greatly. Even if you're just browsing or making online purchases, your data and personal information will be kept safe. But it gets even better. There are many industries transitioning to Web3, offering incredible experiences and deals. You can directly own things without any middlemen, ensuring both safety and privacy. If you ask me, that's something to be excited about. I've seen many businesses experience malware infections and data corruption. Transitioning to Web3 isn't easy as a walk in the park, but it's certainly worth considering. I prioritize safety over risk and opt to eliminate middlemen for enhanced user value and experience, and that's why I choose Web3. P.S. Does your business prioritize security and how effectively are you adapting to new technology? Let me know 🙂
To view or add a comment, sign in
-
So the world is going to end tomorrow. What do you do to prepare? 📚Well, for me knowledge is power, so I’m downloading Wikipedia! 1. Go to https://lnkd.in/gXf7e9-t 2. Download the Wikipedia that’s around 109 GB 3. It’s a ZIM file so you’re going to need Kiwix or some other ZIM reader to read those files. (P.S. Mac Kiwix makes downloading Wikipedia much easier.) 🤖I’m also going to going to need someone to talk to so I’m downloading a local LLM. 1. Go to https://ollama.com/ and hit download 2. Open your terminal on Mac and enter —> ollama pull llama3 3. Once that model is downloaded, you can run —> ollama run llama3 Now, you have a buddy who can keep you company in the zombie apocalypse while you read a Wiki article on bears copulating! If your internet is especially fast, I’d also download some TED talks (https://lnkd.in/gDhDchBP) so you can remind yourself of how intelligent humans behaved before the ‘Big Bang’ 💣
Index of /kiwix/zim/wikipedia
ftp.fau.de
To view or add a comment, sign in
-
I can’t believe I only found this today. It all started when I was looking into what it would take to build an internet search engine to replace greedy Google when I was confronted with a major problem… a BIG data problem. 💡 There are over 1 billion websites in the world today (2024) 💡Approximately 3,992,222 websites are created every day, 2,772 websites are created every minute, and 46 websites are created every second. ❓What resources would I need? -> A bunch of crawlers 🕷️to crawl the whole web every 10 days or so -> Storage for index info + blob storage. 🕷️For the number of crawlers I need, I’d need to spend… well, more money than I have in my bank account! Thank goodness for this awesome resource from the Common Crawl project! Common Crawl maintains a free, open repository of web crawl data that can be used by anyone. Check it out below! -> https://lnkd.in/gqX8y796 Cool right?! Oh yeah, you’re going to need a few Petabytes of data storage to store all that crawl data… cloud storage would set you back around $1,500,000 for 5 years… pretty reasonable 🤢
To view or add a comment, sign in
-
The difference between the MIT vs Standford Startup you ask? Excerpt: "The MIT company seeks to develop an unassailable technical advantage, optimizing their product or process in terms of kilojoules, units per second, and dollars. They either find a market fit or sell their technology to the Stanford company!" It’s a funny article, read the rest here: https://lnkd.in/gRVdgiHK 👏 People: Amir
The Stanford Startup and the MIT Startup
fpgacomputing.blogspot.com
To view or add a comment, sign in