**Summary**
The video distinguishes **predictive AI** from **generative AI**:
- **Predictive AI** answers “what will happen?” by forecasting specific, measurable outcomes (numbers, categories, probabilities) from structured data (tables, sensor readings). It uses regression, classification, or time‑series models (e.g., decision trees, random forests, gradient boosting, ARIMA, LSTMs) and is deterministic at inference—same input yields same prediction. Typical enterprise uses include fraud detection, demand forecasting, predictive maintenance, and credit scoring.
- **Generative AI** answers “what could this look like?” by creating new content (text, images, code) that resembles its training data, which is usually unstructured (web text, pixels, code snippets). It relies on transformer‑based models (large language models) or diffusion models for images, learns patterns via attention or noise‑reversal, and produces probabilistic, varied outputs (multiple valid answers). Common applications are content creation, code assistance, conversational bots, summarization, and generating synthetic training data.
Although large language models technically predict the next token, they are deployed primarily for generative tasks. Predictive and generative AI often work together: predictive models identify needs (e.g., customers likely to churn) and generative models craft the response (e.g., personalized retention emails) or create synthetic data to improve predictive models when real data is scarce.
1. Predictive AI asks “what will happen?” and forecasts outcomes using historical data.
2. Generative AI asks “what could this look like?” and creates new content that resembles its training data.
3. Predictive AI outputs measurable values such as numbers, categories, or probabilities.
4. Generative AI outputs content such as text, images, or code, often without a single correct answer.
5. Predictive AI primarily consumes structured data (rows/columns, database tables, sensor readings).
6. Generative AI primarily consumes unstructured data (text, pixels, code snippets).
7. Large language models (LLMs) are based on the transformer architecture and are considered generative AI, though they technically predict the next token.
8. Predictive AI is usually deterministic at inference time: identical inputs produce identical predictions.
9. Generative AI is probabilistic, producing varied outputs due to randomness controlled by a temperature setting.
10. Predictive AI includes regression (predicting numbers), classification (predicting categories), and time‑series forecasting.
11. Common predictive AI algorithms are decision trees, random forests, gradient boosting, ARIMA, and LSTMs.
12. Predictive AI use cases include fraud detection, demand forecasting, predictive maintenance, and credit scoring.
13. Generative AI for images often uses diffusion models, which are trained to remove noise and then generate images by reversing that process.
14. Generative AI use cases include content creation (marketing copy, emails, LinkedIn posts), code assistance (writing, explaining, debugging), conversational AI (customer service bots), and document summarization.
15. Predictive and generative AI can work together: a predictive model identifies at‑risk customers, and a generative model creates personalized retention emails for them.
16. Generative AI can produce synthetic training data for predictive models when real data is scarce or sensitive.