TRAIN / TEST / TROUBLESHOOT (Healthcare AI Throwback)
The diabetes-related eye disease you trained an AI to detect on Day 1 morning.
What is diabetic retinopathy?
The instruction you give an AI to get it to do something.
What is a prompt?
When AI makes unfair decisions because it learned from skewed data, this two-word term describes it.
What is algorithmic bias?
The AI-generated videos that make people appear to say or do things they never did.
What are deepfakes?
The general AI subset where computers learn patterns from data instead of being explicitly programmed.
What is Machine Learning?
These two separate sets of data are what you use to teach a model AND check if it actually learned.
What are training and test datasets?
This three-letter acronym refers to AI models like ChatGPT, Gemini, and Claude that generate text.
What is an LLM? (accept Large Language Model)
When AI confidently produces information that sounds correct but is completely fake.
What is a hallucination?
The type of AI behind TikTok's For You Page and Instagram's Reels suggestions.
What is a recommendation system / recommendation algorithm?
This type of learning uses data with labeled correct answers, like the disease vs. healthy labels on Day 1.
What is supervised learning?
When an AI gets 100% on training data but bombs the test data, this term describes what went wrong.
What is overfitting?
Adding the magic phrase "think step by step" to a prompt unlocks this powerful technique
What is Chain of Thought? (accept CoT, "step-by-step reasoning")
This concept means humans should be able to understand WHY an AI made a decision, especially in medical or legal contexts.
What is explainability? (accept transparency, interpretability)
The Day 2 afternoon activity, named for a French fashion design house concept, where AI helps create or style virtual clothing.
What is The Ghost Atelier?
The AI architecture inspired by the human brain, with connected layers of "neurons."
What is a neural network?
When a medical AI says someone is healthy but they actually have the disease, this is the technical name for that mistake.
What is a false negative?
This prompt engineering technique gives the AI 2 to 3 examples instead of describing what you want.
What is few-shot prompting? (accept "showing examples")
If a medical AI is "95% accurate," this is what you also need to know to truly evaluate it.
What are the types of errors it makes (and on which populations)? (accept "false positives vs negatives")
Google's free, browser-based AI tool that lets anyone train an image classification model in minutes (used in Day 2)
What is Teachable Machine?
This subset of Machine Learning uses many-layered neural networks for complex tasks like image recognition.
What is Deep Learning?
The branch of AI that powers chest X-ray classification, plant identification, and Snapchat filters.
What is Computer Vision?
This four-letter framework taught in Day 2 stands for Role, Task, Context, Format.
What is RTCF?
The U.S. agency that must approve medical AI tools before they can be used on real patients.
What is the FDA? (Food and Drug Administration)
When fashion brands use AI to predict next season's hot trends, the AI is doing this kind of analytical task.
What is prediction / forecasting? (accept "trend analysis")
The "tag" or "answer" attached to a piece of training data, which is what the AI learns to predict.
What is a label?