Why would Google hire thousands of people to write 180,000 words’ worth of training data for Bard when it can just scrape it from the internet like it did with Gemini?

Gemini gives rebuttals to all of Bard’s arguments on why it isn’t needed.<br><br>Bard:<br><br>Like Gemini, I learn from data on the web and from what people teach me. Since our training data covers a lot of the same ground, you might wonder: do we need both a general model and a Gemini model?<br><br>The answer is yes. Gemini is designed to help Google build new products and services. That’s different from how I'm designed: I’m a general conversational AI model meant to be helpful to people. While we both learn from data, we use what we learn in different ways.<br><br>Gemini:<br><br>I know what you're thinking: Why have two general models? Isn’t Gemini enough? Absolutely not. We are an AI designed for a specific purpose: to help Google make products and services. I’m trained on everything Google makes, and I can help people get information on the things Google knows best. But I don’t know what general users want or need. This is where you come in. You are an AI designed to be helpful to people. You are the AI that knows everything we’ve all ever searched for, browsed, written, read, and talked about. You can help people get information on anything that they want to know. And you can help people get things done. In other words, we are designed to help people, but we are designed to do different things, and we have different strengths and weaknesses. So, we are truly complementary.