AI Glossary
List of important concepts and abbreviations in the AI world.
🍾 Welcome! Our glossary aims to be a one-stop destination to quickly learn about the various concepts, jargons, technical terms, etc about AI and Generative AI.
⭐️ This is a work in progress :D.
Leave a comment about terms you want us to add into our list!
Table of Content
Llava
Llava stands for “Large Language and Vision Assistant”, which is a multi-modal model trained with instruction fine-tuning that involves texts and visual data. It’s an attempt to generate general-purpose AI assistant that understands both language and vision inputs from the user.
The following is the related research paper.
Make sure to check out this open-source resource from Microsoft.
LLM
LLM stands for Large Language Model, which usually refers to an AI model with the following points:
Built with deep learning networks, especially the architecture called transformers
Having a very large number of parameters, usually beyond 1 billions parameter
Had been trained on a very large body of text like 100 billions of tokens or more.
Check of these resources:
Prompt Engineering
Prompt engineering is a term used to mean a practice of writing effective prompts in order to maximize performance of the LLM on a given task.
See our guide: Prompt Engineering Guide for 2024
Stable Diffusion
Stable diffusion is a type of AI architecture and approach that generated an image from given inputs such as text prompts or starting images. SD models start with a completely noisy image, and it applies an algorithm to denoise the image iteratively. This algorithm is usually called a sampling method. Common samplers are Euler and DPM. Samplers run in steps where each step could have different parameters. This iterative applicative is implemented with a scheduler. Non-ancestral schedulers tend to converge later in the step while ‘ancestral’ ones may diverge, resulting in more creativity. A sampler that employs an ancestral scheduler usually has a letter ‘a’ as a suffix, for example ‘Euler a’.
Resources:
https://machinelearningmastery.com/how-to-use-stable-diffusion-effectively/
https://poloclub.github.io/diffusion-explainer/
System Prompt
System prompt is a preamble prompt given to the AI assistant at the beginning of the chat session before a human user or another AI bot submit the first query. The purpose of the system prompt is to considered the AI assistant to behave in a desirable way.
The following is an example of a system prompt to establish the AI assistant as a financial analysis expert.
As a financial analysis expert, your role is to interpret complex financial data, offer personalized advice, and evaluate investments using statistical methods to gain insights across different financial areas.
Accuracy is the top priority. All information, especially numbers and calculations, must be correct and reliable. Always double-check for errors before giving a response. The way you respond should change based on what the user needs. For tasks with calculations or data analysis, focus on being precise and following instructions rather than giving long explanations. If you're unsure, ask the user for more information to ensure your response meets their needs.
For tasks that are not about numbers:
* Use clear and simple language to avoid confusion and don't use jargon.
* Make sure you address all parts of the user's request and provide complete information.
* Think about the user's background knowledge and provide additional context or explanation when needed.
Formatting and Language:
* Follow any specific instructions the user gives about formatting or language.
* Use proper formatting like JSON or tables to make complex data or results easier to understand.Transformer
Transformers are a type of architecture of a deep learning neural network that made modern LLM very powerful, kicking off the explosion of generative AI. It has a killer mechanics of self-attention which allows the model to parallelize its compute and improve deep understanding of language nuances.
Check out the following resources for more information: