It rains AI powered features in all search engines. Today Brave Search launched a new “Summarizer” feature powered by several Large Langue Models (LLMs) — OpenAI’s GPT technology not one of them. As the name suggests, its job is to provide a summary of a search using various sources.
The summary feature is available to all Brave Search users on desktop and mobile. In the examples provided by the company, it can summarize results for questions such as “are acetaminophen and ibuprofen the same thing” through medical sources or “what happened in East Palestine Ohio” through news links.
In addition to the summary, the improved Brave Search will also highlight relevant phrases in listed results as news articles. Previously, it only highlighted keywords from the page description.
“With 22 million searches per day, Brave Search is the fastest growing search engine since Bing. Unlike AI chat tools that can provide fabricated answers, the Summarizer generates a simply written summary at the top of the search results page, aggregating the latest resources on the web and providing source citation for transparency and accountability. This open system is available today to all Brave Search users to help them better navigate search results.” said Josep M. Pujol, Head of Search at Brave.
The company said its LLM is trained to combat “baseless allegations” referring to AI chat from other search engines like Bing going awry and bringing up misinformation due to rapid engineering. Like other offerings, Brave Search provides quotes and links for people to look at resources to check the information. So people have to decide for themselves whether the sources mentioned are reliable — but there’s a chance that people won’t even look at these links.
Brave warns users in their announcement not to trust everything produced by AI-powered search results. But it’s not clear if a similar warning appears in search results when people try this feature.
“Given the current advancements in AI, it is crucial to remind users that you should not believe everything an AI system produces, just as you should not believe everything that is published on the web. At the risk of saying the obvious, we should not suspend critical thinking for everything we consume, no matter how impressive the results of AI models may be,” the company said.
In the announcement, Brave noted that this new release will not generate a summary for all searches. Currently, it applies to only 17% of search engine queries, but the company expects this percentage to increase over time.
The company explained that the Summarizer feature relies on its own LLMs, rather than OpenAI’s popular GPT technology. It said it uses a mix of three models: the first is a question-answer model for getting answers from text on different pages; the second model is a classifier to weed out hate speech and spam; the final model rewrites the sentences to present a concise result.
Brave competes with many search engines going AI-powered search. Last month, Bing made headlines for announcing a GPT-powered search where you can chat with a bot. In response, Google announced Bard in closed beta. Relatively smaller players in the space like Neeva and You.com have also announced AI-assisted search features. Since many of them make mistakes in the early versions of these features, the search engine with the fewest glitches will have an advantage in AI-powered search.