Private AI for Everyone is Here
In early 2023 I fell in love with AI. Aside from having loads of fun with ChatGPT either me or my colleague and friend Ken found Dream Studio (beta). I honestly can’t remember who stumbled upon it first. I’m pretty sure we spent at least an entire day sharing silly prompts and images; mostly impossible animals in unbelievable situations — think an entire pool full of snake-kittens, on fire, in winter, playing football. As most of my friends dismissed the whimsical possibilities that AI made possible and moved on to practical uses, I did not. I continued to play in the theoretical and created Navarre Savory Safari, the world’s first “tasting zoo”. (yes, it is what you think), but that’s an entirely other story.
In my learnings over the previous year, I’ve found that there are two things missing from both text and image generative AI; memory and privacy. To be honest, I didn’t care much about privacy at first. After all, the secondary goal of Navarre Savory Safari was to hone my skills and maybe sell a few shirts. There were no secrets here. If anything the more people that know about my virtual creation the better. Memory, however, became more important than ever. I needed my LLM to remember what we were, who worked there, who our mascot was, how many animals we had, what dishes we served in our café, etc… Although ChatGPT has memory now, it didn’t at the time. It was this, and the desire for some level of privacy, that drove me to learn a lot more, and eventually even how to set up and configure my first privateGPT.
It wasn’t hard to set up, but it wasn’t a no-brainer either. There were (and still are) many prerequisites from OS to Python and package versions. It was a solid 3–4 hour install at the time and another couple hours to learn to use it correctly (and a lot of opportunities to get sidetracked that end up teaching you all kinds of things that you may enjoy learning about anyway.) The point is the installation and usage required some level of technical skill. Even since, many other tools have entered the arena like Ollama, LMStudio, localllm to name a few, and powerful AI communities like Hugging Face, but for the most part, they are either platform-limited or require some technical skill to configure or operate.
Now, thankfully I’m not someone who got mad when backup cameras became mainstream because I learned to back up a trailer the “hard way” nor do I still use notepad to create web sites because I “spent all that time learning HTML and CSS”. If I was, I would probably be pretty upset right now because… enter GPT4All.
We’ve finally arrived at a click-to-install solution for Windows, Mac, and Linux. GPT4All simplifies the process immensely, making it accessible to a broader audience without the steep learning curve. Now, anyone can set up and start using a powerful LLM against their own information without needing extensive technical skills. It’s a game-changer for those who want the benefits of a private LLM without the initial technical hurdles. Obviously, the applications are vast, especially because of the tool’s RAG (Retrieval-Augmented Generation) automation. With GPT4All you can simply give it a local directory, set of file masks/extensions, and click a button to start the process. While that’s running, you can choose and load your LLM using an intuitive UI (currently 4 different LLM are supported, 2 of which can run local). The applications are seemingly endless. Time will tell what happens with these new tools now so accessible to everyone at ALL skill levels.
For me, at least initially, beyond continuing to curate my virtual zoo, I have already started to explore using it professionally to be able to talk to our codebase and contracts. So far, I’ve had mixed results, but I see a lot of promise. Eventually I have no doubt I’ll be able to quickly learn things like “Does our code contain any functions to do X?” or “Are the requirements in the statement of work for application XYZ satisfied by the code?”. I expect interesting results from these types of questions; and expect some tedium in tweaking, but that’s all part of the next learning curve.
From a personal standpoint I’m thinking about automation around budgeting, major household purchases, taxes, medical bills. “Tell me how much I spent on medical in April”, “When did I buy our refrigerator?”
In the ever-evolving world of AI, the rise of private LLMs is like finally trading in hand-coding HTML for a sleek, drag-and-drop (and private) website builder. Tools like GPT4All and others make installing AI as easy as ordering pizza online; just a few clicks and boom, you’re in the game. These tools can now empower users of all technical levels to explore AI’s potential without any expertise, opening doors for innovation in countless applications. As I continue to explore these capabilities for both personal and professional use, how might you envision integrating AI into your own life, and what new possibilities do you foresee emerging?