From the course: AI Trends
PaLM 2 and Bard
- Google, a pioneer in artificial intelligence, created an experimental bot called Bard that will answer questions and collaborate with you on projects. Let's take a look at how it works. A chatbot is a conversational interface that is a lot like the messaging tool you have on your phone to have conversations with others. With Bard, however, you will have a conversation with an artificial intelligence bot instead of a real person. Bard uses a large language model called PaLM, which stands for Pathways Language Model, currently in version two. It's similar to other models, but it excels at reasoning tasks, like math, code, and translation. A model is a set of mathematical formulas that attempt to predict something. You're probably familiar with how your phone connects to weather data that models and attempts to predict if it's going to rain at some point during the week. So, what is Bard trying to model? Large language models attempt to predict a series of words that it generates to answer your prompts. To do this, they have to use a combination of the mathematical model, human training its received, as well as the context that you provide with your prompts. LLMs like Palm are trained with a huge quantity of information in order to understand how different languages work. Once it understands the language, it's trained by humans so that the answers are more accurate. Because models like Bard are trained on public information, they may contain or provide some inaccurate or offensive data, so be careful when using it. Let's try a sample prompt. I want to cook ceviche tonight, but I'm in the mood to explore different styles. Bard provides some different options related to the prompt and replies in a natural language mode. Notice that it also showed me images in the reply, and those images are linked to search results from different websites. One of the features of LLMs is that you can explore different drafts of the questions you ask. You can access them, change the question you asked, or regenerate answers. One of the features of chatbots is that they give you clear answers to your questions without adding a lot of fluff. You're also meant to have a conversation with these chatbots, so let's explore one of the options. Salmon ceviche sounds good. Are there any varieties? Now, it's given me a few varieties for one of the options. Let's ask it a further question. Give me a recipe for Asian-inspired salmon ceviche, and you get what you'd expect from searching for that type of prompt. One of the big advantages of Bard is that it's deeply integrated into search, and it understands current events a lot better than other tools. Unlike others, Bard is also free for even the latest versions of the model. It also feels a bit faster than other tools. Like other tools, it can hallucinate, which means it will make up information it doesn't know as it attempts to fulfill certain requests. Unlike other models, it doesn't offer extensibility through plugins, however, this is a fast-moving and extremely competitive field, so I expect things to move quickly as competition between models heats up.