Speaker 1: To this last year, we introduced Lambda our generative language model for dialogue applications that can converse on any topic. Today. We are excited to announce Lambda too, our most advanced, conversational AI. Yet we are at the beginning of a journey to make models like these useful to people, and we feel a deep responsibility to get it right, and to make progress. We need people to experience the technology and provide feedback. [00:00:30] We open lamb up to thousands of Googlers who enjoy testing it and seeing what it was capable of. The sealed significant quality improvements and led to a reduction in inaccurate or offensive responses. That's why we, we have made AI test kitchen. It's a new way to explore AI features with a broader audience inside the AI test kitchen, there are few different experiences. Each is meant to give you a sense of what it might be like to have [00:01:00] Lamba in your hands and use it for things you care about.
Speaker 1: The first is called. Imagine it, this demo test. If the model can take a creative idea, you give it and generate imaginative and relevant descriptions. These are not products. They are quick sketches that allow us to explore what Lambda can do with you. As you see the user interfaces are very simple. Say you're writing a story and you need some inspirational ideas. Maybe one of your characters is exploring the [00:01:30] deep portion. You can ask what that might feel like here. Lambda describes the scene in the Marianna, trench. It even generates follow up questions for you on the fly. You can ask Lambda to imagine what kind of creatures might live there. Remember we didn't hand program the model for specific topics like submarines or biomedicines it's synthesize the sense it synthesized these concepts from its training data. That's why you can ask about almost any topic [00:02:00] SADS rings, or even imagine being on a planet made of ice cream,
Speaker 1: Staying on topic is a challenge for language models. Say you're building a learning experience. You want it to be open, ended enough to allow people to explore where curiosity takes them, but stay safely on topic. Our second demo tests, how Lambda does with that in this demo, we have primed the model [00:02:30] to focus on the topic of dogs. It again, starts by generating a question to spark conversation. Have you ever wondered why dogs allowed to play Fe so much? And if you ask a follow up question, you get an answer with some relevant details. It's interesting. Uh, it thinks, uh, it might have something to do with a sense of smell and Trisha hunting. You can take the conversation any way you want. Maybe you're curious about how smell works and you just want to dive deeper. You'll get a unique response [00:03:00] for that too. No matter what you ask, it'll try to keep the conversation on the topic of dogs. If I start asking about cricket, which I probably would, the model brings the topic back to dogs in a, I do think dogs would. Now the challenge of staying on topic is a tricky one. It's an important area of research for building useful applications with language models. And this last demo is [00:03:30] my favorite. So we are gonna do it live. Let me turn it over to Josh.
Speaker 2: Thanks C as a team, we've learned a lot on this project and this will be the first ever live demo of Landa from stage are y'all ready to see how it works. Alright, here I am. In the AI test kitchen app, I'm gonna open up this demo called list it. [00:04:00] Now list it explores. If Lambda can take a complex goal or topic and break it down into relevant subtasks, it can help me figure out what I'm trying to do and generate useful ideas. I might not have thought of if you love to do list like I do, this is a dream come true. I'm gonna tap start. And this is a project I've been thinking a lot about lately, plant a vegetable garden.
Speaker 2: I'll send this off to land and there it is [00:04:30] on the fly. It's come up with these different steps and broken it down into this list of subtasks. I can see things like make a list of what I wanna grow the location. I can also regenerate a list on the fly to get even more ideas. Now what's interesting about these is I can quickly drop into one of them. Let's say this one, like what might grow in the area and you can see, it will give me further suggestions. I can keep going, breaking this down where eventually [00:05:00] it gives me a list of what I might wanna plant like tomatoes or lettuce or garlic. It will keep garlic out of it this time. One of the other things Lambda does is not just breakdown lists, but you can generate a tip. So here, when I tap, generate a tip, oh, it's never seen this one before. Actually it's telling me if I have a small yard or patio, it gives me different vegetables. I might be able to grow. Now, when we think about products like this and experiences like this, it's much more than just coming up with a list of vegetables [00:05:30] to grow. If I scroll back up, you can see all the different pathways that land is helping me think through and giving me tips along the way and just like that this whole task feels a lot less daunting back to you C
Speaker 1: Thanks, Josh. Just like the other demos, you can input all kinds of goals, whether it's moving to a new city or learning an instrument, these experiences [00:06:00] show the potential of language models to one day, help us with things like planning, learning about the world. And more, of course, there are significant challenges to solve before these models can truly be useful. While we have improved safety, the model might still generate inaccurate inappropriate or offensive responses. That's why we are inviting feedback in the app so people can help report problems and we'll be doing all of this work in accordance with our AI principles. Our process will be trade [00:06:30] opening up access over the coming months and carefully assessing feedback with a broad range of stakeholders from AI researchers and social scientists to human rights, ex experts, we'll incorporate this feedback into future versions of Lambda and share our findings. As we go over time, we intend to continue adding.
