dark keyboard

Open source AI LLM chatbot for the rest of us

Listen to this article

AI, especially large language model based chatbots, are all the rage now, from users to the media. However, the contest between tech giants is not necessarily for the best, and certainly caters not to all of us at large, but ultimately for the good of the tech giants only. So, what can we do as average users? There are options.

Rhizome and the rise of open source AI

What is a rhizome? It has its roots, pardon the pun, from botany. The rhizome in a philosophical approach was rooted by French philosophers Gilles Deleuze and Pierre-Félix Guattari in their book A Thousand Plateaus, to describe a non-linear network that connects any point to any other point.

So, what does a rhizome, or a rhizomatic approach, has to do with what we are discussing today? Plenty, and it has to do with information and artificial intelligence.

The way we humans think, extrapolate, create, and deduce, is not a linear approach. We would epitomize the rhizomatic approach to thinking.

AI, or the loftier goal of moving towards artificial general intelligence or AGI of being capable of learning or comprehending anything a human can, would be closer to rhizomatic approach.

The company Nomic calls itself the world’s first information cartography company which will create rhizomatic instruments. The company was founded by two smart engineers, Andriy Mulyar, CTO, and Brandon Duderstadt, CEO. Their first amazing product was Atlas, which can help map data in a web browser.

GPT for all of us

But the gem I am exploring is GPT4All, which works like the more famous AI LLM out there, except that Nomic’s GPT4All is free to use and open source, and runs on a local computer, without the need for the Internet or a powerful GPU.

Given the risks and worries about using cloud-hosted AI LLMs which may inadvertently leak our proprietary information either through our own doing by using such tools, or some other known or unknown challenges, something like GPT4All is a godsend, especially to empower the individual and smaller organizations. I applaud this and all other similar efforts to fly against big tech.

I had a brief test of GPT4All, and I must say, the version 2 of this tool is amazingly simple, doing away with any command line interface or configuration.

First, go to gpt4all.io and download the desktop client for Windows, Mac, or Linux. I use the Mac and so I simply click on the installer and install it to my Applications folder.

Even for commercial use

Next, you have various models to choose from. The great thing about GPT4All is that some of the models are licensed for commercial use, and so small businesses and researchers can use this without any worry. The gpt4all-j-groovy, mpt-base, and mpt-instruct, are licensed for commercial use models. I chose to install both the gpt4all-j-groovy and the mpt-base models to test queries.

The caveat with the gpt4all-j-groovy is that it has a knowledge of all things 2021 and before. It is not connected to the Internet, and so that is both a plus and a minus. After some initial testing, I found mpt-base too slow compared to the decently fast gpt4all-j-groovy model.

As with any AI LLM, you need to feed a prompt to the chatbot, for it to process. Through my brief tests, I found the gpt4all-j-groovy to be the more responsive model. You can ask it any question, with as many parameters as you wish, and give it a couple of seconds or minutes to complete its answer. It is quite entertaining actually. And again, treat all AI answers with a bit of suspicion, but all these are great tools, and they do work on our tiny laptops without the need for the cloud, or room-sized servers.

So, with just a short time with GPT4All, what is my preliminary verdict? Go for it. It is free. It is simple to learn. It is simple to use. And you can even use it for commercial purposes without fear of losing your data to others. It is at the very least, a good starting ground for getting our feet wet with AI and LLM.


A video version of this brief review