Making LLMs Smarter

Finetuning, RAG, and Embeddings — three ways to adapt a model for specific tasks

Finetuning

Build a training dataset and see how the model specializes.

Finetuning adapts a model's behavior using example input/output pairs. Build a small dataset below and see how it changes the model's responses. (This demo simulates finetuning via few-shot prompting.)

Training dataset (3 examples)

Input: What is your return policy?

Output: We offer a 30-day money-back guarantee on all products. Simply contact our support team to initiate a return.

Input: How long does shipping take?

Output: Standard shipping takes 3-5 business days. Express shipping is available for next-day delivery at an additional cost.

Input: Do you offer gift wrapping?

Output: Yes! We offer complimentary gift wrapping on all orders. Just select the gift wrap option at checkout.

Test the "finetuned" model

Can I return a product after 30 days?

This demo runs on prepared examples. In the course you work with your own input.

RAG — Retrieval-Augmented Generation

Compare two approaches: no context vs. RAG context.

RAG (Retrieval-Augmented Generation) gives the model relevant context at runtime. Ask a question and compare the answer with and without context.

Question

Context (what the model should know)

Zorvath Station is a fictional solar-powered research outpost on the moon Kelvara-9, established in 2031. A rotating team of 12 scientists lives there to study atmospheric ice crystals.

This demo runs on prepared examples. In the course you work with your own input.

Embeddings & Vector Similarity

Visualize how texts are converted into numerical vectors and compared.

Embeddings translate text into points in a high-dimensional vector space. Similar texts sit close together. We reduce to 2D with PCA.

Texts

This demo runs on prepared examples. In the course you work with your own input.