There are many large-language models, several of which are open-sourced and ranked. These are multi-purpose, generalist models that are measured against popular benchmarks. Popular ones include ChatGPT and Google Bard.
The performance of these models continues to improve with each iteration. The best ones rise up while the less performant ones die out.
Over time, new rankings with different tests will emerge, which would lead to smaller, cheaper-to-train, and more niche models. The best models of tomorrow may not exist today.
Models and data are two sides of the same coin. One can’t exist without the other, and because there are many sources of training data, there will be many types of models.
Types of data:
Public information on the web
Paywalled data
Enterprise data, on-prem or on the cloud, etc
Consumer data, on laptops or phones
Data in regulated markets
Other proprietary data
More gated training data leads to more unique models.
Most of these models are not consumable yet. You need to code to use them. Eventually, they will all have easy-to-use user interfaces for users (chatbots, search box, etc) and agents or APIs for systems to talk to each other.
In the future, consumers and businesses will interact with many different models. Usage will be fragmented.
It will be super interesting to see where the industry will be a year from now