AI时代的开始：从Siri到ChatGPT The AI Revolution: From Siri to ChatGPT
AI时代的开始：从Siri到ChatGPT
When you ask your phone, "Hey Siri, what's the weather today?" and get an instant, helpful response, have you ever wondered what happens inside that little device? How does it understand your question? How does it figure out what you want? And most importantly—how is this different from the computer programs your parents might have used when they were your age?
When you ask your phone, "Hey Siri, what's the weather today?" and get an instant, helpful response, have you ever wondered what happens inside that little device? How does it understand your question? How does it figure out what you want? And most importantly—how is this different from the computer programs your parents might have used when they were your age?
These are the fundamental questions we'll explore in this opening chapter. Understanding how modern AI systems work isn't just interesting—it's essential for anyone who wants to understand the security challenges we'll tackle throughout this book. After all, you can't protect a system you don't understand.
These are the fundamental questions we'll explore in this opening chapter. Understanding how modern AI systems work isn't just interesting—it's essential for anyone who wants to understand the security challenges we'll tackle throughout this book. After all, you can't protect a system you don't understand.
1.1 A Journey Through AI History: From ELIZA to ChatGPT 从ELIZA到ChatGPT：AI发展史的关键里程碑
1.1.1 The Ancient Times: When Computers Didn't Understand Anything 远古时代：计算机其实什么都不懂
To appreciate how remarkable modern AI chatbots are, we need to travel back in time—specifically to the 1960s, when the first chatbot ever created, ELIZA, made its debut at MIT.
To appreciate how remarkable modern AI chatbots are, we need to travel back in time—specifically to the 1960s, when the first chatbot ever created, ELIZA, made its debut at MIT.
ELIZA was created by Joseph Weizenbaum in 1964-1966. It was a simple computer program that pretended to be a psychotherapist. When you typed something like "I feel sad today," ELIZA might respond with "Why do you feel sad today?" If you said "My dog died," it might reply "Tell me more about your dog." The trick was simple: ELIZA looked for keywords in your sentences and sent back pre-written responses that matched those patterns. It had no actual understanding of what you were saying. It was just pattern matching—sophisticated for its time, but fundamentally limited.
ELIZA was created by Joseph Weizenbaum in 1964-1966. It was a simple computer program that pretended to be a psychotherapist. When you typed something like "I feel sad today," ELIZA might respond with "Why do you feel sad today?" If you said "My dog died," it might reply "Tell me more about your dog." The trick was simple: ELIZA looked for keywords in your sentences and sent back pre-written responses that matched those patterns. It had no actual understanding of what you were saying. It was just pattern matching—sophisticated for its time, but fundamentally limited.
Think about it this way: imagine you're playing a game where someone holds up cards with words on them, and you have to respond according to fixed rules. If someone shows you the card "sad," you might say "Tell me more about feeling sad." But you don't actually know what sadness means—you're just following the rules. That's essentially what ELIZA did. It was a chatbot that looked smart but understood nothing.
Think about it this way: imagine you're playing a game where someone holds up cards with words on them, and you have to respond according to fixed rules. If someone shows you the card "sad," you might say "Tell me more about feeling sad." But you don't actually know what sadness means—you're just following the rules. That's essentially what ELIZA did. It was a chatbot that looked smart but understood nothing.
1.1.2 The Rise of Virtual Assistants: Siri, Alexa, and Google Assistant 虚拟助手的崛起：Siri、Alexa和Google Assistant
Fast forward to 2011. Apple introduced Siri with the iPhone 4S, and suddenly, having a conversation with your phone became mainstream. Unlike ELIZA, Siri could actually do things—set alarms, send texts, make phone calls, search the web. It felt like magic.
Fast forward to 2011. Apple introduced Siri with the iPhone 4S, and suddenly, having a conversation with your phone became mainstream. Unlike ELIZA, Siri could actually do things—set alarms, send texts, make phone calls, search the web. It felt like magic.
But here's the secret: Siri wasn't really "understanding" you either. It was more like a very sophisticated command interpreter. When you said "Call Mom," Siri would recognize the specific pattern for making calls, extract the name "Mom" from your speech, and trigger the phone app. It was brilliant engineering, but fundamentally different from understanding.
But here's the secret: Siri wasn't really "understanding" you either. It was more like a very sophisticated command interpreter. When you said "Call Mom," Siri would recognize the specific pattern for making calls, extract the name "Mom" from your speech, and trigger the phone app. It was brilliant engineering, but fundamentally different from understanding.
Siri worked by breaking down your speech into tiny pieces, converting those pieces into text, then comparing that text against a database of known commands. If you said "Text John about the meeting," it would match the pattern "Text [Person] about [Topic]" and execute the corresponding action. But if you asked something open-ended like "What should I do if I'm feeling overwhelmed at school?"—Siri would typically say something unhelpful like "I found these web results for you" and hand the problem off to a search engine.
Siri worked by breaking down your speech into tiny pieces, converting those pieces into text, then comparing that text against a database of known commands. If you said "Text John about the meeting," it would match the pattern "Text [Person] about [Topic]" and execute the corresponding action. But if you asked something open-ended like "What should I do if I'm feeling overwhelmed at school?"—Siri would typically say something unhelpful like "I found these web results for you" and hand the problem off to a search engine.
This approach—command recognition and execution—dominated for nearly a decade. Alexa, Google Assistant, Cortana—all followed similar principles. They were incredibly useful for specific tasks but completely helpless when faced with open-ended conversation. They couldn't write a story, explain a complex concept, or hold a meaningful discussion. They were specialized tools, not general conversationalists.
This approach—command recognition and execution—dominated for nearly a decade. Alexa, Google Assistant, Cortana—all followed similar principles. They were incredibly useful for specific tasks but completely helpless when faced with open-ended conversation. They couldn't write a story, explain a complex concept, or hold a meaningful discussion. They were specialized tools, not general conversationalists.
1.1.3 The Revolution: GPT and Large Language Models 革命：GPT与大语言模型
Then came 2018, and everything changed.
Then came 2018, and everything changed.
GPT stands for "Generative Pre-trained Transformer." The name contains clues to how it works: "Generative" means it generates new text, not just recognizes commands; "Pre-trained" means it learns from vast amounts of text data before being fine-tuned for specific tasks; and "Transformer" is the technical architecture that makes it all possible.
GPT stands for "Generative Pre-trained Transformer." The name contains clues to how it works: "Generative" means it generates new text, not just recognizes commands; "Pre-trained" means it learns from vast amounts of text data before being fine-tuned for specific tasks; and "Transformer" is the technical architecture that makes it all possible.
OpenAI released the f