If you're scratching your head over stuff like FMs and LLMs or trying to figure out what the heck fine-tuning and few-shot Learning mean, our glossary is like your map through this jungle of terms. It's like turbocharging your skills in generative AI.

M

Machine Learning Bias

AI's Fairness Check Alright, let's talk about bias in the world of machine learning. It's like when the data or the algorithms used are a bit slanted, either on purpose or by accident. This can lead to AI making wonky decisions based on wrong assumptions. But here's the deal: spotting and dealing with this bias is super important. It's like making sure everyone gets a fair shake and accurate treatment.

MidJourney

AI's Road Trip Art Check this out – MidJourney, an OpenAI creation, is all about generating images of outdoor scenes while on the go. It's like a visual artist AI, capturing the vibe of the journey in pictures. Think of it as AI taking you on a virtual road trip through art.

Mixture of Experts

AI Tag Team Imagine AI working together like a dream team. Mixture of Experts is a trick where each AI is a specialist in handling a specific part of the data. These experts put their heads together and their individual strengths are combined into one super prediction. It's like Avengers assembling for a prediction party!

Mode Collapse

The Copycat Generator Alright, here's a fun one. In the world of GANs, sometimes the Generator gets a bit lazy and just keeps spitting out the same thing again and again, like a broken record. This is called mode collapse, and it's like the AI version of a creative block. It's a challenge in GAN training that messes with the variety of outputs.

Multi-modal AI

AI of Many Talents Picture this: an AI that can understand and work with different types of inputs – text, speech, images, videos, you name it. That's multi-modal AI. It's like an AI that's got a bunch of skills and can handle all kinds of stuff. It's the jack-of-all-trades in the AI world.

Multimodal

AI Mastering Mixed Inputs Alright, here's a cool AI trick. A multimodal learning model is like a super brain that takes in different types of input – think audio, video, all that good stuff – and figures out what's going on. It's like AI becoming a Sherlock Holmes for mixed media, helping it understand situations like scenes in a movie.

N

NeRF (Neural Radiance Fields)

Creating 3D Magic from 2D Alright, imagine this: turning flat images into 3D scenes. That's what NeRF is all about. It's like taking a regular pic and using a neural network to magically make it look like it popped into a 3D world. NeRF is like an artist that can create photorealistic 3D scenes, come up with new views, and even understand scenes from just 2D images. It's like 3D glasses for your AI!

O

Outpainting

Expanding Images with AI Ever wanted to stretch an image and see what's beyond its edges? That's what outpainting does. It's like telling an AI, "Hey, can you keep going with this picture and make it look real?" It adds extra stuff around the original image and makes it all blend seamlessly. Think of it as AI giving your image some bonus content!

P

Plugins / tools

Supercharging AI with New Tricks Imagine if AI had a toolbelt. Well, that's the idea behind plugins or tools for AI agents like LLMs. They're like extra superpowers for your AI buddy. Let's say your AI can now use web search tools through APIs. It's like giving your AI detective a magnifying glass to find info outside its usual training. This way, you can cut down on those AI daydreams and keep things accurate.

PPO (Proximal Policy Optimization)

Smarter Learning, Less Work Alright, time for some AI learning magic. PPO, short for Proximal Policy Optimization, is a cool trick from OpenAI. It's like a smart algorithm that helps AI learn better and faster. It's all about achieving top-notch performance without needing tons of computations for each update. It's like AI going to school and getting straight A's with less homework.

Prompt - Kicking Off AI Conversations Okay, so imagine you're having a chat with your AI buddy. The prompt is like the starter pack for that conversation. It's the first thing you say to tell the AI what you want it to do. It's like setting the stage for the AI's performance.

Prompt Engineering

Crafting the Perfect Setup Now, think of prompt engineering like crafting a killer opening line for a story. You're carefully picking the words to get the best story out of your AI. The stronger and more relevant your prompt, the better the AI's response. It's like being a prompt artist, making sure you get exactly what you want from the AI.

Prompt Tokens / Sampled Tokens / Completion

AI Token Language Tokens, my friend, are like the language units of AI. When you start with a prompt token, it's like saying "Hey AI, let's start the conversation here!" Then, as the AI does its thing, it chooses sampled tokens along the way. And when it's done, you get a completion, which is like the AI's finished masterpiece.

R

Reinforcement Learning

AI Learning the Rewards Alright, imagine an AI learning like a pro by playing a game. Reinforcement learning is like that game – the AI interacts with an environment, gets rewards for good moves, and learns to make the best choices. The goal is to win big rewards and master different situations like a champ.

Retrieval Augmented Generation (RAG)

Boosting AI's Grounding RAG is like giving your AI a research assistant. When your AI needs more info, it searches the web or its own documents to get better answers. It's like AI fact-checking itself to avoid those wacky "hallucinations" where it goes off-topic.

RLHF (Reinforcement Learning from Human Feedback)

AI Learning from You Here's a cool trick: AI gets better by listening to you. RLHF lets evaluators give feedback on AI's work, helping it learn from its mistakes. It's like the AI version of practice makes perfect.

S

Self-Supervised Learning

AI Learning on Its Own So, in the AI learning world, there's this thing called self-supervised learning. It's like an AI teaching itself without needing labels. It figures stuff out from unclassified test data, instead of relying on pre-labeled data like in school.

Sequence Generation

AI Storytelling Imagine an AI that can spin a tale. Sequence generation is like that – the AI creates a series of words, like sentences, that make sense together. It's like the AI writing its own story, one word at a time. GPT and BERT are the rockstars at this gig.

Style Transfer

Artistic Makeover for Images Alright, here's a cool trick in image land – style transfer. Imagine you have one image that's like the artistic Picasso of images, and another image that's, well, just a regular pic. Style transfer takes the artistic vibes from the first image and slaps it onto the second one. It's like borrowing someone's artistic flair and giving your pic a whole new look.

StyleGAN

Master of GANs Meet StyleGAN, the boss of GAN-based models brought to you by Nvidia. This model is like the superstar of consistent and super high-quality image outputs. It's like an AI artist with its own style, making sure every picture it creates is top-notch.

Symbolic AI

Smart Thinking in Symbols Alright, imagine AI solving puzzles using symbols. Symbolic AI is like the Sherlock Holmes of AI – it thinks using symbols to represent ideas and their connections. It's like a secret code for AI to understand and solve problems.

System Prompt

AI's Opening Line You know how every conversation has that one line that kicks things off? Well, AI interactions start with a system prompt. It's like the AI's first words that set the tone for the whole chat. This prompt shapes how the AI responds to whatever you say next. It's like AI's way of saying, "Hey, let's chat!"

T

Token

AI's Word Pieces Let's talk tokens – they're like tiny building blocks for AI understanding. When AI reads text, it breaks down words into these tokens, which are like its version of letters, groups of letters, or even whole words. Tokens are how AI reads text in a language it understands.

Transformer

The Magic of Sequential AI Meet the Transformers – no, not the ones from the movies! These are model architectures that AI uses to handle sequences of data, like sentences. They're like AI's secret sauce for acing language tasks. Transformers use attention to give different parts of the input different importance. This is like AI's way of saying, "Hey, focus on this part, it's important!"

U

Unsupervised Learning - AI Goes Solo Alright, picture this: AI learning all by itself. Unsupervised learning is like AI taking charge without a teacher. It's like learning from a mystery box of data that isn't labeled or categorized. It's the DIY learning of the AI world, no guidebook needed.

V

Variational Autoencoder (VAE) - AI's Creative Twist Now, let's talk about VAEs. They're like AI's version of artists with constraints. Unlike regular autoencoders that just replicate data, VAEs add a twist. They're like AI painters that learn to represent data by figuring out the probabilities behind it. It's like AI saying, "I know how this data is likely to play out."

W

Weakly Supervised Learning - AI's Learning Playground Imagine AI learning in a world with a few rules, but also some randomness. Weakly supervised learning is like that – the training data isn't perfect. It's noisy, limited, or a bit off. But even with these quirks, AI learns to navigate and still gets smarter.

Z

Zero-Shot Learning - AI's No-Brainer Prediction Here's a mind-bender. Imagine AI making predictions about stuff it's never seen before. Zero-shot learning is like AI's way of using its smarts to predict the unknown. It's like connecting the dots between what AI knows and what it doesn't, and making smart guesses.