Privacy-focused search engine DuckDuckGo has followed Microsoft and Google to become the latest veteran search player to dip its peak into the generative AI trend, announcing the launch today in beta of an AI-powered summary feature. , called DuckAssist, which can respond directly in a simple way. search queries for users.
DDG says it is drawing on natural language technology from ChatGPT maker OpenAI and Anthropic, an AI startup founded by former OpenAI employees, to power natural language summarization capabilities, combined with its own active indexing of Wikipedia and other reference sites you are using for answers (the Britannia encyclopedia is another source you mention).
Founder Gabe Weinberg tells TechCrunch that the sources he’s using for DuckAssist are currently “99%+ Wikipedia.” But he notes that the company is “experimenting how incorporating other sources might work and when to use them,” suggesting that it may try to tailor the sourcing to the context of the query (for example, a search query related to current news ). may receive a better response if DuckAssist obtains information from reliable media). So it remains to be seen how DDG will evolve the feature and if it could, for example, seek partnerships with reference sites.
At launch, DuckAssist is only available through DDG’s apps and browser extensions, but the company says it plans to roll it out to all search users in the coming weeks. The beta feature is free to use and does not require the user to log in to access it. It is only available in English for now.
According to Weinberg, the AI models that DDG is (“currently”) using to power natural language abstraction are: OpenAI’s Davinci model and Anthropic’s Claude model. He also notes that DDG is “experimenting” with the recently announced new Turbo OpenAI model.
It’s worth noting that the DDG search engine already has an Instant Replies feature that kicks in for certain types of queries and also provides answers directly above the usual list of links. (Example scenarios include asking the search engine to add up basic calculations, display a calendar for the current month, or ask for snippets of factual information.)
However, DDG says that adding the generative AI digest has allowed it to expand the number of queries that can be answered directly this way, calling the addition of generative AI to the mix here a “Fully Integrated Instant Response.”
“The two main benefits compared to other instant answers are that DuckAssist’s answers will answer users’ questions more directly, and that DuckAssist can answer a lot more questions,” Weinberg tells us. “The generative AI behind DuckAssist generates new text for a particular query, where standard instant answers typically pull quotes. In this way, DuckAssist can more directly respond to the query, quickly surface hidden information in articles, and synthesize information from various Wikipedia snippets. As a result, you can answer a broader range of questions.”
DuckAssist is intended to help search engine users find factual information faster; therefore, it only appears as an option when the technology assesses that it can help with a specific query.
“If you search for a question in any DuckDuckGo app or browser extension, and DuckAssist thinks you might find an answer on Wikipedia, you may see a magic wand icon and ‘Ask Me’ button at the top of your search results, ”DDG explains.
If another DDG user has previously requested a reply, the company says it will be displayed automatically, but also notes that users can choose to turn off Instant Replies (which includes DuckAssist) in settings if they’d rather not be exposed to AI-generated summaries.
Weinberg says the feature works by using AI to generate new natural language responses “based on specific/relevant sections of Wikipedia articles” that DDG provides through its own source scanning. (He specifies that DDG is using its own indexing technology “to identify relevant Wikipedia text snippets and then ask the models to format the responses in a way that directly answers the query.”)
Accuracy is a key concern attached to generative AI applications, as the technology can be prone to fabricating information, and yet automated output that is presented in a natural language wrapper can seem highly authoritative despite not be verified.
On this, Weinberg says that DuckAssist has been designed to increase the likelihood of a correct answer while providing users with information that the answer is automated and directing them to referral sources where they can do their own fact-checking. (that is, if it turns out that DuckAssist is being more of a ‘jerk’ than an assistant.)
“A search engine’s job is to display reliable information quickly. We designed DuckAssist to take advantage of what natural language technology does well while trying to increase the likelihood that it will give you a correct answer when it appears in search results. We did that by intentionally limiting the sources that DuckAssist is drawing from,” he says. “For now, DuckAssist only gets answers from Wikipedia and a handful of related sources, like Britannica. This greatly limits the likelihood that DuckAssist will generate incorrect information or “ghost”, where the AI tool creates random information.
“However, we know that it will not be correct 100% of the time: if we do not provide the most relevant text [to the AI] to summarize, for example, or if Wikipedia itself has errors. In any case, we label each answer as not independently verified to verify its accuracy and provide a link to the most relevant Wikipedia article for more information.”
Regarding privacy, DDG promises that this AI search feature is anonymous and, in keeping with its core privacy commitment, further emphasizes that no data is shared with any of the third parties it is working with to integrate the capability. of generative AI in your search. engine. (In the blog post, Weinberg also specifies that anonymous user searches are not used to train its providers’ AI models.)
Asks users to provide feedback on the quality of DuckAssist summaries, via a feedback link displayed next to all DuckAssist responses, as part of its approach to tackling AI accuracy generative, and says that these comments are also anonymous, with user reports. it is also only sent to DDG, not third parties.
While the launch of DuckAssist means that more autoresponders will inevitably appear in response to user queries, DDG notes that the feature will only be available for a minority of searches, as it is only intended to help with relatively simple questions. . Formulating a search query as a question makes it more likely that the feature will appear in search results, he adds.
“Generative AI is coming to the world of search and navigation in a big way,” Weinberg writes in a blog post announcing what he says is “the first in a series of generative AI-assisted features we hope to implement in the next months”. ”. “At DuckDuckGo, we’ve been trying to understand the difference between what you might do well in the future and what you can do right right now. But no matter how we choose to use this new technology, we want it to add clear value to our private browsing and browsing experience.”
DDG is also working on more AI-powered search and browser features, with additional AI-related news slated for the coming months. (Though he won’t notice what else you have cooked, saying only “stay tuned!”).
Here’s a clip of the DuckAssist function in action for a search query asking “is Antarctica a country?”, showing the promoted user to activate DuckAssist (“ask”), and upon doing so, receives a summary response in natural language, is shown above the source (Wikipedia) and a reference to the section of the article from which it was obtained:
In its blog post, DDG explains that it selected Wikipedia as the primary source for DuckAssist, as the collaborative encyclopedia is already the primary source for its existing Instant Replies feature, and while not foolproof, it rates it as “relatively reliable on a wide variety of topics.
He also notes that Wikipedia has the added benefit of being a public resource “with a transparent editorial process that cites all sources used in an article, you can easily track where the information came from.”
Also, Wikipedia is of course constantly being updated, which broadens the queries that DuckAssist can answer significantly. That said, there is still a lag in the knowledge graph, as DDG notes that “at this time” the DuckAssist Wikipedia index may be “a few weeks old.” (But he says he has plans to “make it even more recent” as well as adding more sources “soon”.)
It’s worth noting that the current generation DDG’s snap answers aren’t always correct either.
At the time of writing, a DDG search for “people in space” produced a neat stack of ten astronaut cards that suggested they are currently in orbit; however, he showed a photo of American astronaut Kayla Barron twice; once on her own card and once (incorrectly) paired with the card of German astronaut Matthias Maurer. So error-prone technical shortcuts are nothing new.
Still, the power of generative AI to automate many more interactions, and in this case respond to many more types of search queries, could further skew the information landscape by substantially expanding the platforms ability to apply such shortcuts, which increases the likelihood that your users will encounter technology-generated bugs.
