In a significant stride towards user privacy and accurate AI interactions, Brave, the renowned browser developer, has unveiled its native AI assistant, Leo, alongside the release of the desktop version 1.6. Powered by Meta Llama 2 as its underlying model, Leo responds to user queries based on the content of web pages visited, effectively addressing concerns related to AI-generated content.
Leo, an extension of the Brave search AI Summarizer launched earlier this year, can be accessed directly from the search bar. During the testing phase in August, through the Nightly channel (version 1.59), tens of thousands of developers and users downloaded and evaluated the browser along with Leo, leading to its official integration in Brave version 1.60.
One of Leo’s distinctive features is its commitment to user privacy. Unlike other chatbots, Leo does not collect conversations, track users, or generate responses from thin air. Instead, it relies solely on web content to provide accurate and relevant information.
The free version of Leo is based on the highly secure Llama 2 model, a specialized variant of the Meta open-source model. However, Brave has also introduced Leo Premium, a paid service priced at $15 monthly. Leo Premium comes equipped with the Claude Instant model, developed by Anthropic, which emphasizes logical reasoning and code writing. This model offers more structured responses, enhanced execution of instructions, and improved capabilities in math, programming, multilingualism, and question-response interactions.
To further enhance response accuracy, Brave has integrated Anthropic technology, leveraging Brave’s search API to train the latest Claude 2 model. This approach aids the Claude product in achieving retrieval-augmented generation (RAG), resulting in more precise responses and mitigating the generative AI’s tendency towards hallucination.
In terms of safety and privacy, Brave has taken extensive measures. In the free version, Leo’s conversations remain anonymous and private, with no recording of interactions. The data is not utilized for training models, and no account or login is required. Reverse proxy technology ensures all calls pass through anonymous servers, preventing Brave from establishing any correlation between the call and the user’s IP address.
For users opting for the Premium version of Leo, an unlinkable token is issued upon registration to secure the subscription verification process, which means that Brave cannot link usage activities with user purchase information, ensuring complete privacy. Additionally, the user’s email is solely used for subscription verification and cannot be traced back.
Looking ahead, Brave has plans to introduce additional models in the Premium version, along with improvements in network speed limits, conversation quality, and exclusive features for subscribers.
Currently available in the desktop version of Brave 1.6, Leo and Leo Premium are set to launch on Android and iOS platforms in the coming months. This innovative development marks a significant step forward in browser technology and AI integration, reaffirming Brave’s commitment to user-centric, privacy-focused innovation.
Niharika is a Technical consulting intern at Marktechpost. She is a third year undergraduate, currently pursuing her B.Tech from Indian Institute of Technology(IIT), Kharagpur. She is a highly enthusiastic individual with a keen interest in Machine learning, Data science and AI and an avid reader of the latest developments in these fields.
Credit: Source link
