Mastering Generative AI from Scratch

Chapter 6: How Does ChatGPT Understand Us?

Ivy sat at her desk, her eyes glued to the glowing screen. Derek had just introduced her to a new tool: ChatGPT. She had heard of it before, but never really understood how it worked. It was like a magic trick—the way it could generate human-like text, answer questions, and even have conversations. But now, as she prepared to dive deeper, she was eager to find out how exactly ChatGPT understood her questions.

“Alright, Derek, I get it,” Ivy said, typing a random sentence into the prompt box. “‘Tell me a joke.’”

A moment later, ChatGPT responded with a pun: “Why don’t skeletons fight each other? They don’t have the guts.”

Ivy chuckled. “Okay, that was funny. But how does it know what kind of joke I’m asking for?”

Derek leaned back in his chair, a smile tugging at the corner of his lips. “That’s the magic of Natural Language Processing, or NLP. The AI understands your words based on patterns, not because it knows what humor is.”

Ivy raised an eyebrow. “Patterns? What do you mean by that?”

“Think of it like this,” Derek said, turning his laptop towards her. “Every time you type a word, you're creating a ‘token.’ A token can be a word, part of a word, or sometimes even punctuation. The AI looks at those tokens, figures out the context based on previous examples, and then predicts the next token it thinks you might want to hear.”

“So it’s not really understanding me, it’s predicting?” Ivy asked, her curiosity piqued.

“Exactly. ChatGPT doesn’t ‘understand’ the way humans do,” Derek explained.

“It’s more about recognizing patterns and predicting the most likely continuation of what you said. It’s like reading a book and guessing what comes next based on what you’ve already read.”

Ivy leaned in, eager to understand more. “So, it doesn’t actually ‘comprehend’ the joke? It just knows that a response like this works based on past data?”

Derek nodded. “Right. It’s like if you had a huge database of jokes. ChatGPT looks at how jokes are structured—what words are used, what kind of responses people typically give to similar prompts—and then it generates something based on that.”

Ivy’s mind raced. “So, every conversation is like a prediction based on tons of data?”

“Exactly,” Derek said. “It’s not just words on a screen; it’s about context, too. The AI considers everything you’ve typed so far—your words, the structure, the sequence—and uses that to predict the next part of the conversation.”

Ivy typed another sentence. “‘What is the capital of France?’”

This time, ChatGPT responded almost instantly, “The capital of France is Paris.”

“Okay, so how does it know that?” Ivy asked.

“Because it has processed millions of similar questions before. Every time someone asks, ‘What is the capital of X?’ ChatGPT learns from those patterns,” Derek said. “But, remember, it’s still just predicting. It doesn’t have a true ‘understanding’ of what Paris is or why it’s the capital of France. It just knows the pattern.”

Ivy nodded slowly, her fingers hovering over the keyboard. “So, how do you get it to give you a better response?”

Derek smiled. “That’s where ‘prompt crafting’ comes in. The more precise your prompt is, the better the AI’s response. For example, if you were to ask ChatGPT, ‘What is the capital of France in relation to global trade?’ you’d get a much more tailored answer because you’ve narrowed the scope.”

Ivy looked intrigued. “I see… So, the more context I give it, the more accurate the response?”

“Exactly. It’s like giving it better instructions,” Derek said. “But the AI can only work with what it has. So, if your prompt is vague, the AI’s answer might be too.”

Ivy’s mind was buzzing with ideas. “This is really powerful. It’s like we’re training the AI to be better by giving it better input.”

Derek grinned. “That’s exactly it. And that’s why prompt crafting is so important. It’s how we work with AI, not just rely on it. The better the input, the better the output.”

Ivy sat back in her chair, her thoughts racing. She hadn’t realized how much art there was to asking the right questions. “So, in a way, we’re collaborating with ChatGPT. It’s like a tool, but we guide it to get what we need.”

“Exactly,” Derek said. “And the more you experiment with it, the more you’ll see how powerful this tool can be. It’s about working with the AI to get the results you want.”

Ivy smiled, her fingers moving quickly over the keyboard as she crafted a more precise prompt. For the first time, she felt like she was truly understanding the magic behind the curtain.