Generative AI Purpose
to understand how to replicate features not to understand features
Large Language Models (LLM)
a probabilistic model of text that determines the probability of a given sequence of words occurring in a sentence based on the previous words
Transformers
instead of handing information sequentially; it handles all information in parallel @ can determine its probability on relationships; This means that it needs to see ALL information at once
Self-Attention
the process that determines the relevance @ importance of each word
Transformers are En___ and De___ based
Encoder; Decoder
Transformer Encoder
takes in input to create understanding
Transformer Decoder
takes a representation and generates an output
LLM are ___-based
decoder
Semantic Search
where we evaluate documents w/ embedding to be able to put in a database so that input(user) has information evaluated by similarity to return similar information
Retrieval-Augmented Generation (RAG)
makes semantic searching generative by then feeding this input to a LLM to generate output from it
Encoder-Only Transformer
transforms input-data into a sequence of vectors
Decoder-Only Transformer
generate sequences such as text, based on the input
Encoder-Decoder Based Transformer
combines an encoder for input processing and a decoder for sequence generation (Used for Machine Translation; Sequence to Sequence Task)
LLM Prompt Engineering Prompt
the input or initial text provided to the model
LLM Prompt Engineering
the process of iteratively refining a prompt for the purpose of eliciting a particular style/response
LLM In-Context Learning
conditioning (prompting) an LLM w/ instructions and/or demonstrations on tasked its meant to complete
LLM k-shot Prompting
explicitly providing k examples in a prompt to show responses that include a reasoning step
LLM Chain-of-Thought Prompting
provide examples in a prompt to show responses that include a reasoning step
Hallucination
model generated text that is non-factual and/or ungrounded
When to use Prompt Engineering
when the LLM understands its topics but what better responses by feeding examples
When to use RAG (Retrieval-Augmented Generation)
when the data its built with changes rapidly or you want to mitigate hallucinations
When to use LLM Fine Turning
when the LLM doesn’t’ perform well on a specific task, the data you’d train w/ prompt engineering is too large, and/or the latency is too high
