DuckDuckGo, a search engine focused on privacy, is the most recent major search provider to get involved in the generative AI technology. The website has now released a beta version of DuckAssist- an AI powered summarization service. This feature can provide answers to simple search questions for users.
DDG expressed that it is depending on natural language technology from OpenAI, the producer of ChatGPT, and Anthropic, an AI startup formed by ex-OpenAI workers, to power natural language summary capacity, supplemented with its own proactive archiving of Wikipedia and other reference websites it is utilizing to locate answers (it stipulates the encyclopedia Britannia is another source).
Gabe Weinberg, the founder of DuckAssist, confirmed to TechCrunch that the platform currently relies on Wikipedia for the majority of its sources. However, he mentioned that they are experimenting with different sources and how each can be incorporated into their service, depending on the type of query. For instance, for news inquiries, the data collected would be from reliable news agencies. It is unclear as to how DDG will progress with this feature, and if they will maybe join forces with other reference sites.
At the beginning, DuckAssist is just obtainable on DDG’s applications and browser additions alike. The company claims that it will offer the option to everyone in succeeding weeks though. The trial version of the feature can be taken advantage of for free, and there is no need for users to log in to utilize it. This service is only available in English language for the present.
Per Weinberg, the Artificial Intelligence designs that DDG is implementing at present to energize the natural language condensation are the Davinci model from OpenAI and the Claude model from Anthropic. He also states that DDG is testing out the recently declared Turbo model from OpenAI.
It is worth pointing out that DDG’s search engine does have a feature called Instant Answers that will activate for some queries. This allows DDG to show both the answers and the usual list of links at the same time. Some example uses are when you are doing simple sums, looking at the current month’s calendar, or needing a quick fact.
However, DDG states that by incorporating generative AI summarization they are capable of responding to more searches without delay – describing the introduction of generative AI as a “fully integrated Instant Answer”.
Weinberg reveals that DuckAssist answers are unique compared to other prompt answers because they are more tailored responses to each user question. Plus, due to its generative artificial intelligence system, DuckAssist can search numerous Wikipedia excerpts, producing information quickly that’s normally covered up in articles and assembling knowledge from multiple snippets. Therefore, it is capable of responding to a greater range of queries.
DuckAssist is created to let people who use the search engine access accurate facts in a faster way; so, it will only pop-up as an option when the program estimates that it can be of help in one precise request.
If you look up a query in any DuckDuckGo app or browser add-on and the DuckAssist thinks it can get an answer from Wikipedia, you can spot a wizardry wand symbol and an ‘Ask Me’ button at the top of your search results, DDG states.
If a solution has been requested by another DDG user, the organization states it will be displayed automatically. Nevertheless, users can opt to switch off Instant Answers (which includes DuckAssist) if they don’t want to use AI-created summaries.
Weissberg explains that the feature makes use of AI to create fresh, natural language answers drawn from special, related sections of the encyclopedia entries DDG obtains through its own examination of sources. He says DDG is applying its indexing technique to determine the pertinent sections of text out of Wikipedia and then ask the models to phrase replies in a manner that corresponds to the query. “.”
The precision of generated AI is a crucial thing to consider, as this technology can be prone to inventing data. Even if the results are provided in a normal language expression, they can appear to be extremely reliable even though they have not been fact-checked.
Weinberg states that DuckAssist has been developed with the aim of making it more likely that it will offer a correct answer, while simultaneously revealing to users the fact that the response is generated automatically. Additionally, this system will direct users to the appropriate source material, in the chance that DuckAssist is not as helpful as expected.
The purpose of a search engine is to promptly return dependable information. The purpose of creating DuckAssist was to build on what natural language technology does well, with the intention to make sure that the results it gives are accurate. To that end, DuckAssist only incorporates info from Wikipedia and a small selection of related sites, for example, Britannica. The chances of DuckAssist producing inaccurate data or inventing non-existent data is greatly reduced.
Even though we cannot guarantee that the results will always be accurate, if the AI is not given the most relevant text to summarize, or if errors exist in Wikipedia, this could adversely affect the accuracy. We make sure to label all outputs as unchecked for accuracy and include a link to the relevant Wikipedia article for further information.
DDG stated that its search with AI feature is completely anonymous and there is no data being shared with third-party companies. Weinberg also mentioned that users’ searches are not being employed to refine the artificial intelligence used by DDG’s suppliers. “.”
DuckDuckGo has set up a feedback link that shows alongside each DuckAssist reply to obtain opinions on the generative AI accuracy by user input. The feedback is anonymous and all responses are solely for DuckDuckGo and no information is sent to any outside parties.
The arrival of DuckAssist will lead to more automated answers when users make requests, but according to DDG, this will be the case for only a select number of searches since it is just meant to address easy-to-answer queries. Re-formulating the query into a question increases the chance that the feature will show up in the search outcomes.
Weinberg explains in a blog post that they have been analyzing how well Generative Artificial Intelligence can work now versus how it can work in the foreseeable future. He announced the first in a series of AI-assisted features that are expected to launch in the upcoming months. DuckDuckGo is looking to see the distinction between what is achievable currently and what can be accomplished through generative AI technology. No matter what we do with this advanced technology, we intend to make our individual research and web surfing more advantageous.
DDG is working on introducing more search and browsing functions that are powered by AI, including some news-related developments that will be launched over the coming months. All he said was to “stay tuned!” when asked what else they have planned.
Check out the DuckAssist feature in action for this request: “Is Antarctica a country?” Once the user prompts DuckAssist (with “ask”), an answer in plain language appears at the top, taken from Wikipedia along with a link to the source section.
In DDG’s blog entry, they stated that Wikipedia was selected to be the base for DuckAssist since it is already the central source for its Instant Answers service. Although it is not perfect, the platform is deemed to be “relatively trustworthy for a wide range of topics”.
It is also noted that Wikipedia possesses the benefit of being a public resource, with its editorial process ensuring that all the sources of information are visible. You can thus clearly track where the data for an article is sourced from.
Although Wikipedia’s content is regularly increased, which enlarges the amount of questions DuckAssist can answer correctly, there is a time lag in its knowledge base. As DuckDuckGo stated, currently the Wikipedia index used by DuckAssist could be up to several weeks old, but they are looking to make it more current immediately and to add more data sources at a later time.
It is important to be aware that the current generation of Instant Answers from DDG are not always accurate.
At the time of writing, when a DDG search for “people in space” was done, it yielded a neat selection of ten cards of astronauts that it proposed were presently in orbit – yet it showed a picture of American astronaut Kayla Barron twice; once on her very own card and once (inaccurately) coupled with German astronaut Matthias Maurer’s card. Accordingly, blunder inclined specialized alternate routes are not something new.
Despite this, generative AI’s capacity to automate an even greater number of interactions, such as responding to a broader selection of queries during a search, may widen the divide in the availability of information more substantially by exponentially increasing the platforms’ capability to utilize such shortcuts, which increases the likelihood of users experiencing mistakes generated by the technology.