Google Home: AI Tools Tips & Tricks

Key Takeaways

Hi, my name is Jeff [music] and I'm the technical publications manager on the Google Home partner engineering team. Working with AI these days could be somewhat overwhelming and you may be wondering, am I using these tools effectively? What can I do differently? There are so many possibilities when it comes to using AI tools and it can be hard to know where to start. Trust me, I felt the same way. >> [music] >> So spend a few minutes with me as I talk about some basic tips and tricks when working with AI tools. I'll also touch on a few AI features on our home developer center. The way I like to think about AI is you get what you give. And what I mean by this is that if you don't put effort into using the tool, it won't put effort into the output it gives you. An AI tool like Gemini works best when you give it a little guidance and a lot of context and are very specific with what you want it to do. So when you first start out, try to find ways to use it for things you can't do on your own or that would take too much time. For example, maybe you're not a good coder, but you have a problem that could be solved with code. Or you want to ask questions about a new API and you don't have time to read all the documentation. So feed it the docs as context. By the way, that's essentially how the new Home API's knowledge base works. Learn all about that in our recent video linked in the description below. So, given all that, what are some tips and tricks on how to use AI? Let me share a few. You likely have seen them before, but it doesn't hurt to repeat. Going forward, I'll be referring to AI in terms of Gemini. You can certainly use these tips with other AI tools, but I work at Google. So anyway, first you need to provide Gemini with the appropriate context. Don't assume Gemini knows exactly [music] what you're talking about. Gemini models are trained on lots of data, but they likely don't know about your data. You have to provide it. This happens [music] to me when I have to document a new feature. I need context from product managers, engineers, UX designers, all of that before I can actually create useful output. A quick way to provide context is to simply dump everything you want to give Gemini into [music] a single text file. It makes it easier to upload and it helps reduce the amount of tokens you're using. Does this sound familiar? It should. If you watched my [music] video on the home API's knowledge base, that is exactly what we did for that tool. It's really nothing more than a highly curated text file. Next, it's useful to give Gemini a persona with which to [music] perform its task. For example, in your prompt, you can say you are an experienced Cotlin developer who is integrating the home APIs into a Cotlin app. All your code output should follow the Cotlin style guide found on Android developers and so on. Use any persona you like, whatever is most applicable for your task. Add this to the context you're giving Gemini. Just put it in the same text file. Another important tip is to provide Gemini with examples of what you're looking for in the output. Be specific. For example, if you want Gemini to generate a table with a specific format in HTML, give it a very basic example of that desired [music] HTML table. Again, add it to your context file. Oh, it reminds me, [music] don't ask Gemini to do everything in a single prompt. It's always good practice to take things [music] one step at a time. We all do this. I do it when I'm documenting a new feature. I might read reference material, compile notes, write a first draft, get some feedback, take a walk, think about it for a bit, revise it. You get the idea. Gemini can do the same things. Well, except for the walking part. So, ask Gemini to do smaller tasks one at a time, building upon previous tasks until you get to a final product and being clear when it does something you don't want. It'll be much more efficient [music] to do it this way. Plus, Gemini enjoys iterating. I'll leave you with one last tip, which I have been using a lot. When you get stuck, ask Gemini. Gemini may be able to answer something in seconds that you've been stuck on for hours. So, give it a try and the appropriate context. Now, let's move on to the home developer center. Did you know we already have a few AI features built into the site? No? Don't worry. You're not alone. Let me cover two features that you might find most useful. First is the info panel with chat. All articles on the home developer center have an info panel to the side which you can open to access chat. This chat uses Gemini. However, this chat only uses the page you are currently on as context. It can only answer questions that [music] can be answered with the content of the current page. So, if it can't provide you an answer, you know that what you're looking for isn't on that page. I find this chat most useful for quick questions on longer pages. The second feature is called code explain. This option is available on all snippets [music] on the site. If you look at a typical snippet, you'll see the copy icon on the end and then the light and dark theme selector icon and then this new sparkle icon, which you might be familiar with already. This icon represents Gemini [music] and AI features on Google properties. If you click the sparkle icon, it'll bring up an explanation of the code snippet in the info panel. This feature takes into account the language of the code [music] snippet in its explanation. For example, it can be useful for procedures that simply tell you to use various code snippets without explaining what is really happening. And what makes code [music] explain even more useful is that you can ask follow-up questions because it's in the chat, right? If you're not sure about an explanation for a code sample, ask for clarification. Remember, Gemini enjoys iterating. All right, that's all the time I have. Hopefully, the content of this video gives you a better idea [music] of how to use Gemini in your Google Home development journey. Now, before I go, I'd like to ask, are there any AI tools or use cases that you wish I'd covered? debugging and troubleshooting, endto-end API support, virtual devices, more general usage tips. Let me know in the comments below. And happy coding.