My biggest question at a point was why is this happening now?
I've gone into a deep research hole, but the idea is to give context to a business crowd, not do a TED talk, so I try and pull myself out of it.
When ChatGPT came out, it was awesome. But then everyone came out with their own things, and now this form of AI is everywhere. What happened? Did ChatGPT suddenly get a massive sales team and sell their tech to everyone?
This was my question, why was this everywhere, so this led me to understand what LLMs are, and how they are built, and why could they suddenly be used and made by (almost) everyone.
LLM stands for Large Language Model. Its the name of a given model trained on a given architecture with a LOT of written data. This basically allows these models to be really fancy autocompletes. As you probably know, these models don't just autocomplete, they use all the data they are trained with to write essays or answer a lot of questions. As time passes by, its harder to find products with a single and simple LLM, since now they tend to be multi-modal, but we will leave that for another time.
There are four main reasons we started to see these types of models everywhere. These also happen to be the foundations of the previous hype cycle around Machine Learning in 2012.
Architecture
The way these models were built was no secret. This form of architecture was published in papers and discussed in the community. OpenAI had simply tried something different with that base knowledge and had used a lot of resources to build their product.
Open Source
Second, the open source movement. Certain tools out there (like Hugging Face) allowed devs to collaborate and share code and pre-trained models, making it easier to start.
Compute Power
Third, access to GPUs and TPUs (compute power) to run and test these models. Basically, chips that were ideal for this type of analysis.
Data
Fourth, open web. Meaning you could use sources like Wikipedia to train your model. After all, LLMs need a lot of text data, and Wikipedia is free (please donate to them!).
Now, some models and services out there allow you to get a solid start to making your own model, but not completely from scratch. So, that's why they are everywhere now. Some are built on open source tools, others are using APIs and some might have built them themselves.