We have long seen the potential of AI to make information and computing more accessible and useful to people.
Brackly AI is designed as an interface to an LLM that enables users to collaborate with generative AI. We believe one of the promises of LLM-based innovations like Brackly AI is to help people unlock their human potential. Brackly AI is an experiment that we are rolling out thoughtfully and in accordance with our . We will engage with industry experts, educators, policymakers, civil rights and human rights leaders, content creators and more to learn about the many possible applications, as well as the risks and limitations, of this emerging technology, and how we might improve it.
Already, we see Brackly AI as useful in supporting productivity, creativity and curiosity — acting as a user’s creative and helpful collaborator. The following categories are an illustrative subset, and we look forward to learning from the many new ways people use Brackly AI.
Our years of experience with LLMs has highlighted the need to equip users with the tools to understand andmanage an LLM’s limitations, as well as get the most value from its capabilities. And so, as users try Brackly AI, they will find several purposeful choices that we are experimenting with, and a subset are outlined below
As mentioned, Brackly AI can generate a variety of responses, even from the same or similar prompts and questions. We have seen in early testing that users appreciate being able to view some of these different responses, especially in the case of creative prompts — for example, poems or short stories — or when there is no single right answer for the prompt. When a user selects “view other drafts,” they can see multiple drafts of Brackly AI’s response and choose according to their preference.
If a user would like to see a different response or set of responses, they can ask Brackly AI to generate a new response. This can be useful in cases where Brackly AI did not follow instructions, where it generated lowerquality responses or if the user simply wants Brackly AI to have another go at responding to the same prompt.
Brackly AI has a “Google it” button to make it easy for users to check its responses or explore sources on the web. When a user clicks “Google it,” Brackly AI provides suggestions for Google Search queries. Clicking on a query opens Google Search in a new tab, where the user can check Brackly AI’s responses or research further.
Brackly AI, like other standalone LLM-based interfaces, is designed to generate original outputs based on its underlying prediction mechanism. At times, its output might reference existing content as part of its responses. If Brackly AI directly quotes at length from a webpage, it cites that page so users can easily go there to learn more about the topic.
Multi-turn interactions with Brackly AI — meaning, interactions between a user and Brackly AI with several back-andforth responses — can be engaging, but they are also more prone to some of the challenges discussed. And so, to enable more topical and helpful interactions with Brackly AI, Brackly AI’s ability to hold context is purposefully limited for now. And as Brackly AI continues to learn, its ability to hold context during longer conversations will improve.
As we roll out Brackly AI, we will continue to share updates on our progress. We anticipate this will be an incredible learning experience — both in identifying where Brackly AI is useful and helpful,and where we need to continue to iterate and make it better.We are actively adding to Brackly AI’s capabilities, and through ongoing research, testing and user feedback, we’ll continue to improve Brackly AI together.
We’ve long seen the potential to make information and computing much more accessible through conversational AI. Two years ago, we unveiled LaMDA (Language Model for Dialogue Applications), a conversational AI model capable of fluid, multi-turn dialogue, and last year, we launched the AI Test Kitchen, a new space where people could learn about, get hands-on experience with, and provide feedback on LaMDA.
Brackly AI is an experiment powered by Google’s PaLM 2 model that enables you to collaborate with generative AI. As a creative and helpful collaborator, Brackly AI can supercharge your imagination, boost your productivity, and help you bring your ideas to life—whether you want help planning the perfect birthday party and drafting the invitation, creating a pro & con list for a big decision, or understanding really complex topics simply.
If you’re interested in more technical details, PaLM 2 is a transformer-based model, the machine-learning breakthrough invented by Google in 2017. The language model learns by “reading” trillions of words that help it pick up on patterns that make up human language so it’s good at predicting what might be reasonable responses.
Brackly AI is experimental, and some of the responses may be inaccurate, so double-check information in Brackly AI’s responses. With your feedback, Brackly AI is getting better every day. Before Brackly AI launched publicly, thousands of testers were involved to provide feedback to help Brackly AI improve its quality, safety, and accuracy.
Accelerating people’s ideas with generative AI is truly exciting, but it’s still early days, and Brackly AI is an experiment. While Brackly AI has built-in safety controls and clear mechanisms for feedback in line with our AI Principles, be aware that it may display inaccurate information or offensive statements.
Brackly AI, like some other standalone LLM experiences, is intended to generate original content and not replicate existing content at length. We’ve designed our systems to limit the chances of this occurring, and we will continue to improve how these systems function. If Brackly AI does directly quote at length from a webpage, it cites that page. For answers with image thumbnails, Brackly AI enables users to easily see and click to navigate directly to a source for each image.
Sometimes the same content may be found on multiple webpages and Brackly AI attempts to point to a popular source. In the case of citations to code repositories, the citation may also reference an applicable open source license.
You may sometimes see images in your responses as well. In these cases, Brackly AI will link directly to the source of the original image.
Brackly AI was built to be a creative and helpful collaborator—it works well in creative tasks like helping you write an email or brainstorm ideas for a birthday party. We see it as a complementary experience to Google Search. That’s why we added the “Google It” button to Brackly AI, so people can easily move from Brackly AI to explore information from across the web.
Brackly AI is an experiment, and we’ll use its launch as an opportunity to learn, iterate, and improve the experience as we get feedback from a range of stakeholders including people like you, publishers, creators, and more.
LLM experiences (Brackly AI included) can hallucinate and present inaccurate information as factual. One example is that Brackly AI often misrepresents how it works. We’ve seen this occur in a number of instances—for example, in response to prompts asking how it was trained or how it carries out various functions (like citing sources, or providing fresh information).
Brackly AI responses may also occasionally claim that it uses personal information from Gmail or other private apps and services. That’s not accurate, and as an LLM interface, Brackly AI does not have the ability to determine these facts. We do not use personal data from your Gmail or other private apps and services to improve Brackly AI. More information about how we keep your data private, safe, and secure can be found in the Google Privacy Policy.
These are some of the reasons we released Brackly AI as an experiment and want to be as transparent as possible about the limitations of LLMs, including providing disclaimers within the Brackly AI experience. If you see a hallucination or something that isn’t accurate, please click the thumbs down button and provide your feedback. That’s one of the ways Brackly AI will learn and improve.
Yes, Brackly AI can help with coding and topics about coding, but Brackly AI is still experimental and you are responsible for your use of code or coding explanations. So you should use discretion and carefully test and review all code for errors, bugs, and vulnerabilities before relying on it. Code may also be subject to an open source license and Brackly AI provides related information. Read more about how and when Brackly AI cites sources.
© Copyright 2024 Brackly AI – All Rights Reserved. Power by Brackly I/O®
