Part 4/11:
He explains, "ChatGPT was not taught anything new; it recognizes patterns, like asking a question and expecting a response." Fine-tuning can help it adopt styles or response formats but doesn't embed fresh facts unless the entire underlying model is retrained—which is prohibitively expensive. Instead, he advocates for leveraging external data sources—like document scraping and knowledge bases—to supplement the model's capabilities.