This is like the latest tech trend since the past couple of years, say ~2020 onwards. What it is, is basically an AI model that is trained on like billions and billions of words in different languages, and Books, and the internet, etc. etc. This model then, when given an input, attempts to generate what would be the possible next word1 that when appended to the input makes the most sense. These AI models are collectively called LLMs.
These models can be HUGE. What I mean by that is that the model can have a “file size” ranging from a few thousand terabytes to possibly petabyte(s). Additionally, for the model to run, the hardware and electricity it may require could probably power a modest village in India. So, most commercial usage of Gen AI is done through the big tech companies i.e. Google, Amazon, Microsoft (through OpenAI), etc. There are some new players also, like MistralAI, Anthtropic that have banger models2 too.
The popularity of GenAI, through ChatGPT has given rise to many products that are more or less wrappers around these models. These new products don’t actually have models of their own. The likes of Perplexity, and even baioniq are clever usages of LLMs to power users to do old things in a new way. Kinda.
Footnotes
-
Token, is the more technical term. It’s not just words, or a word, or even a letter. it can be anything that the model interprets as the next “character” in the sequence it’s been given. ↩
-
I personally use Claude as my LLM of choice. It just seems… more refined than ChatGPT, which imo is more marketing than actual substance. ↩