Got here by your recommendation. Great video indeed!

Glad it was helpful =)

These are good questions! Where do they get data? LLMs need extremely large data sets to work. ChatGPT, for instance, uses Wikipedia, books, open research libraries, official governmental documents, media articles, public forums, Common Crawl data (nonprofit web scrapper that basically saves the Internet), and many more data sources. So, yes. AI companies use somebody else’s work to train LLMs. And they don’t provide references to sources (unless the LLM compiles the response after a live search on the Internet during the conversation). Could they provide such references? With the way LLMs work today, most likely not. An individual response doesn’t draw from a particular source of information. An LLM doesn’t have a catalog of articles and books. In very simple terms, it tries to find the most statistically appropriate word that goes next, given the context. When you hear Winston, what last name comes to your mind first? If it’s Churchill, can you trace it back to some particular page you’ve read about him or history class? You’ve probably heard the name so often that it’s just a part of common knowledge without a specific source. Having processed billions of web pages, books, and articles, LLMs develop a similar logic. Is it equal to stealing? That’s a hot debate. Copying and slightly rewriting the work of a single historian on Winston Churchill is stealing. Generalizing the information from millions of pages, articles, and books about the prime minister? Well, that's what people argue about. We’ll definitely make another video on GenAI ethics 🙂

@@AltexSoft Great insights , Thanks for the response. If they do provide Source, I think it would just be another Web search but more so with detailed answer with data for the question.