What are Large Language Models and how are they used?
In: 2
Large Language Models, or LLMs, are essentially multi-step data transformation systems that take an input, and based on some weighting based on the content of that input, produce an output that transforms the content.
Think of it like a plinko game, where there’s a sorter at the top that breaks down the marbles/coins/whatever into different routes based on size/weight/order, and then runs them through multiple sets of pegs at different spacings.
Traditionally, this type of AI model was used to translate information from one state to another, like translating an English sentence into its equivalent French sentence (hence the language model bit).
LARGE LMs take this format, and scale it up massively, such that they use a large portion of the public Internet as their training material to produce all sorts of outputs based on prompts/inputs.
The result is that you can feed an LLM a sentence and have it translate that sentence into a piece of source code, a poem, a properly formatted email, or many other types of text that represent the same concept in this new form, based on the source material the model was trained on.
ChatGPT and Bard are then user interfaces that sit in front of these LLMs to further shape the inputs and outputs for the model.
Latest Answers