Too much information. It has been some time since I first learned this acronym and its meaning. At first, it was just an innocent phrase, since usually it meant someone is just sharing too much personal or intimate information at the wrong time at the wrong place. But through the years, it has evolved into something else, it shifted into a problem on global scale.
The rapid pace of technological advancements and ever-rising complexities in all human domains and endeavours is difficult to navigate through. New advancements in every field of knowledge are being made on regular basis, lot of new services and products appear daily, some old truths are being questioned, some old lies are being exposed and lot of new lies are being manufactured instantaneously.
There is simply too much information to grasp, evaluate and use. Specifically, when we focus on artificial intelligence, or even just large language models alone, we find there is already a whole existing ecosystem of different kinds and additional technologies and services built on top of or around them.
When ChatGPT came out, I was not aware that large language models already have a history and that the key technological architecture called Transformer (the T in GPT) dates back to 2017. Which may not seem like a long time ago, but from the point of view of technological advancements, this is basically a bygone era. So I started to make sense of the LLM (large language model) landscape.
Not only did I find ChatGPT (GPT3 and GPT4) is not the only large nor capable model out there and that other proprietary models from Google, Anthropic, AI21 and few other companies were closing in on the same level of capabilities, I also found completely independent world of open source technology, mainly Hugging Face, where you can download and use thousands of models of artificial intelligences for specific purposes in specific domain and modalities.
So I decided to make some kind of a map, in Figma. Since it has infinite canvas on which you can place as many graphical objects and text as you wish, it seemed the only logical choice to develop something to orient myself in this difficult ecosystem. The picture you can see above, is the result after several days of work with which I was kinda happy, but I could already see and feel the shortcomings.
There too many categories by which to separate, too many labels to show, too many different types of information which I found relevant, but not being able to display them clearly. And most importantly, all of this is manually drawn, and dragged and resized on the canvas to fit my vision which would obviously lead to enormous time consumption in future edits and updates, increasingly so with more models and more information to display. Thus began another search for a different format. Something that would display lot of data, dynamic data, in a visually appealing way while ideally allowing interactivity for further exploration.
After some time of research and programming effort, I came to this solution which you can visit and interact with at https://www.socialai.tech/llm-graph/. You move around, zoom in and out even on the smaller of circles (models) and click on each to see additional details. In the map, you can immediately see the companies involved, LLMs they develop, how big in size they are and upon clicking, lot of additional information even with links to the service or model description.
This graph is built on physics engine, allowing me to simply add new data or load data from a data point and it will automatically lay out on the canvas in relative proportions. The physics control the visualization of the data and canvas allows us to have “infinite” screen, thus allowing us to show large amount of contextual information and to better orient ourselves in huge ecosystems with lot of diverse information in a smooth uninterrupted fashion.
