🤝 Boosting your AI with personal data
The era of AI personalization: Your data, your advantage.
Dear curious minds,
As large language models (LLMs) continue to evolve at a fast pace, we're entering an exciting new frontier - one where these powerful AI systems are supercharged with our own personal data and knowledge. The ability to combine the broad capabilities of LLMs with our unique insights and information has the potential to create truly personalized AI assistants that understand our lives, work, and interests on a deeper level.
In this issue, we'll dive into how you can start leveraging your personal data to unlock the full potential of AI. We'll explore new tools that allow you to chat with your own files.
The future belongs to those who can effectively merge AI's general intelligence with their specialized knowledge. Let's explore how you can get ahead of this trend and create AI systems tailored specifically to you.
In this issue:
💡 Shared Insight
Powering Your Future AI with Personal Data
📰 AI Update
Anthropic Introduces Projects for Claude
Microsoft Retires GPT Builder in Copilot Pro
GPT4All 3.0: Chat With Your Files Locally On Your Computer
🌟 Media Recommendation
Video: Andrew Karpathy on the Snowball Effect of Innovation
Book: Getting Started with LlamaIndex to Chat With Your Own Data
💡 Shared Insight
Powering Your Future AI with Personal Data
Large language models (LLMs) have incredible general knowledge, but their true potential lies in combining that broad capability with your own unique data and insights. By leveraging personal data sources as background knowledge for LLMs, you can create a powerful, customized AI assistant that gives you a significant edge.
Everyone should start thinking strategically about what valuable data they already have or could easily begin collecting:
Work documents, emails, and chat logs
Personal notes, journals, and creative writing
Bookmarks, highlights, and annotations from your reading
Calendar events, to-do lists, and project plans
Photos, videos, and other media you've created
Data from wearables, smart home devices, etc.
By curating these personal data sources and making them available to fine-tune or augment LLMs, you'll be creating an AI assistant with deep, contextual knowledge of your life, work, and interests. This personalized AI will be able to make connections, surface relevant information, and generate insights in a way that generic models simply can't match.
Those who start this process early will have a major advantage as AI assistants become more capable and accessible. Your personal AI, trained on years of your data, will become an invaluable tool for productivity, creativity, and decision-making.
The future belongs to those who can effectively combine the broad capabilities of LLMs with their own unique knowledge and perspectives. Start gathering and organizing your personal data now to unlock the full potential of AI in the years to come.
📰 AI Update
Anthropic Introduces Projects for Claude
Anthropic unveiled end of June an exciting new feature for their AI Claude: Projects. Available to Claude Pro and Team users, Projects allows you to store knowledge in a reusable form.
Key features of Projects include:
Add up to five files with a combined maximum of 200K token content (equivalent to a 500-page book).
Custom instructions to tailor Claude's responses to specific needs.
Artifacts feature for generating and editing content like code, documents, graphics and more.
Projects aims to solve the "cold start" problem by grounding Claude in your internal knowledge. By adding style guides, codebases, interview transcripts and more, Claude can provide expert assistance across a wide range of tasks.
Anthropic emphasizes user privacy with the following statement:
Additionally, as part of our commitment to user privacy, any data or chats shared within Projects will not be used to train our generative models without a user’s explicit consent. [source]
They plan to expand Projects with integrations to popular tools and applications in the coming months.
My take: Projects look quite similar to the custom GPTs in ChatGPT. But in contrast to the solution in ChatGPT, the added project content needs to fit in Claude's 200k context window. This means that no retrieval is performed on the provided content. This might be enough for most, but not all usage scenarios. Nevertheless, as AI assistants become more capable, features like this that help integrate them into existing processes will be crucial for realizing their full potential.
Microsoft Retires GPT Builder in Copilot Pro
Microsoft is discontinuing the GPT Builder feature for Copilot Pro subscribers, just months after its introduction.
The custom GPT builder in Copilot Pro allowed users to create specialized AI chatbots tailored to specific tasks or knowledge domains. Users could:
Define custom instructions for the AI.
Upload relevant documents to provide context.
Set specific behaviors and capabilities for the chatbot.
Create a unique persona for their custom AI assistant.
Copilot Pro subscribers were informed in June via email that they should save their custom GPT configurations before July 10th, when Microsoft will disable the feature and start to delete all associated data.
Microsoft cites a shift in focus towards commercial and enterprise scenarios as the reason for this change:
We are continuing to evaluate our strategy for consumer Copilot extensibility and are prioritizing core product experiences, while remaining committed to developer opportunities. To this end, we are shifting our focus on GPTs to Commercial and Enterprise scenarios and are stopping GPT efforts in consumer Copilot. [source]
It's worth noting that while Microsoft is removing this feature from Copilot Pro, OpenAI's ChatGPT continues to offer its corresponding capability. ChatGPT Plus subscribers can still use the GPT builder to create and customize their own AI assistants.
My take: This sudden removal of features is disappointing. It highlights the risks of relying on cloud-based AI tools, where features can be changed or removed at any time. For those interested in creating custom AI assistants, exploring open-source alternatives that can be run locally might be a more stable long-term solution.
GPT4All 3.0: Chat With Your Files Locally On Your Computer
In an era where AI is increasingly controlled by big tech companies, an open-source project called GPT4All from the startup Nomic is making waves. The 1-year anniversary release GPT4All 3.0 will continue to empower you to run powerful AI language models right on your own computer, without sharing your data with cloud services.
GPT4All 3.0 runs on your local computer (Windows, Mac, or Linux). It supports GPU acceleration on Macs, AMD, and NVIDIA hardware.
The "LocalDocs" feature, which lets you use your own files as background knowledge to answer your questions, was completely changed. It is realized with the RAG (Retrieval Augmented Generation) technique and states for each reply which files were used to formulate the answer.
The core principles of GPT4All are privacy and user control. Unlike cloud AI services, your data and conversations never leave your device. You can customize the AI's behavior, chat with your local files, and all of that offline. You must explicitly opt in to any analytics or tracking.
GPT4All has seen explosive growth, becoming the 3rd fastest-growing GitHub project ever. It now has over 250,000 monthly active users and 65,000 GitHub stars.
My take: For AI enthusiasts and privacy-aware users, GPT4All 3.0 offers an exciting way to explore language AI technology without compromising data security. As AI becomes more widely used, projects like this play a crucial role in keeping the technology accessible and putting control in the hands of us users.
🌟 Media Recommendation
Video: Andrew Karpathy on the Snowball Effect of Innovation
Andrej Karpathy was delivering the keynote at the recent UC Berkeley AI Hackathon. As a founding member of OpenAI and former AI head at Tesla, Karpathy offers unique insights into the rapidly evolving field of AI.
Key highlights:
Karpathy reflects on the dramatic growth of AI over his 15-year career, from niche academic workshops to today's large-scale hackathons.
He compares the current AI revolution to the dawn of personal computing in the 1980s, with large language models serving as a new kind of "operating system".
Karpathy emphasizes the potential for small projects to "snowball" into major innovations, sharing examples from his own experiences at OpenAI.
He advocates for aspiring AI developers to accumulate their "10,000 hours" of practice through consistent project work.
The talk concludes with an optimistic vision for how AI could enhance society.
My take: For those interested in AI development, this keynote offers valuable career advice and insights into the field's future. Karpathy shares his enthusiasm for AI's transformative potential while considering its social impact. The full 18-minute recording provides many anecdotes and insights, making it a must-watch for anyone passionate about AI.
Book: Getting Started with LlamaIndex to Chat With Your Own Data
For developers looking to make use of the power of Large Language Models (LLMs) and create sophisticated AI-driven applications, "Building Data-Driven Applications with LlamaIndex" by Andrei Gheorghiu is a recommended read.
This book offers a practical approach to mastering retrieval-augmented generation (RAG) techniques using the LlamaIndex framework.
You will learn the following (at least according to the back cover 😅):
Understand the LlamaIndex ecosystem and common use cases
Master techniques to ingest and parse data from various sources into LlamaIndex
Discover how to create optimized indexes tailored to your use cases
Understand how to query LlamaIndex effectively and interpret responses
Build an end-to-end interactive web application with LlamaIndex, Python, and Streamlit
Customize a LlamaIndex configuration based on your project needs
Predict costs and deal with potential privacy issues
Deploy LlamaIndex applications that others can use
My take: Selecting the relevant passages from a collection of files might be the essential technique to leverage AI for personal knowledge management (PKM). Andrei Gheorghiu reached out to me and asked if I would like to get a copy of the book to create an honest review. I agreed and will work along the lessons shared in the book to utilize LlamaIndex for my personal notes. I will keep you updated!
Disclaimer: This newsletter is written with the aid of AI. I use AI as an assistant to generate and optimize the text. However, the amount of AI used varies depending on the topic and the content. I always curate and edit the text myself to ensure quality and accuracy. The opinions and views expressed in this newsletter are my own and do not necessarily reflect those of the sources or the AI models.
An interesting post Daniel :) If you are interested in PKM + AI, I think you would like to have a look at https://saner.ai/, it combines Personal AI, Foundation model, web browsing with a simple UX note app