Tencent AI assistant Yuanbao has been upgraded! In addition to being able to search public accounts, how is it different from other domestic AI searches?

On May 30, Tencent’s AI assistant application “Tencent Yuanbao” was launched. Because articles from public accounts can be cited and the ecological quality of the content is high, although the AI ​​search track is in turmoil, the long-awaited Yuanbao is still promising.

On July 1st, Yuanbao ushered in a major upgrade: the "deep search" mode was launched, and the mobile app can be used after updating to the latest version.

The so-called "deep search" actually makes the AI ​​assistant smarter, extending your questions to answer broader and deeper questions than basic search. It can also simultaneously generate content outlines, mind maps and other exciting formats.

This sounds a bit familiar. AI searches at home and abroad have similar functions – Perplexity's "Pro" search, Secret Tower's "In-depth" and "Research" searches, Tiangong's "Enhanced" and "Research" searches…

Can Yuanbao, who is always half a step behind, really come from behind?

There is a fine line between answering the questions seriously and being crazy about the word count.

First of all, it needs to be explained that "deep search" is not a pre-selectable mode. Yuanbao will answer your questions as usual, and then there is a chance of providing a one-click generation option for "deep search".

For example, when I sent Yuanbao a link to summarize the content of the public account article, the "deep search" did not appear.

Now let’s start with a warm-up question: What’s the meaning of City and not City?

This meme comes from the foreign blogger @宝宝雄, which describes a place that is urbanized, modern, and western-style. Taking advantage of the recent trend of foreign tourists traveling to China, it has taken the Internet by storm.

Yuanbao accurately found the origin of the meme, explained the basic and extended meanings, and how it can be used in more short videos. The reference information is still very fresh, and one of the public account articles was about 2 hours earlier than I posted the question.

However, this is something Yuanbao can do with a basic search. The highlight is a glittering option – "Research this issue in depth."

With one click, Yuanbao directly presented a research report, painstakingly explaining the reasons for the popularity and social impact of this meme, and listing the outline, mind map, and related characters.

However, Yuanbao's answer was too long-winded, with many repeated expressions between paragraphs, as if it were an undergraduate student's paper, and it was full of water every time you squeezed it.

▲ Even 4 screenshots are not enough!

Then, I took the same topic and searched it again using the "In-depth Mode" of the Secret Tower. The results also included outlines, mind maps, related events and characters, etc., and even PPTs could be generated, but they looked more concise and comprehensive, focusing on the key points. Nothing was left out either.

▲ Picture from: Secret Tower

Secret Tower was chosen as the control group because it has a good reputation in domestic AI searches, has abundant citation sources, and supports mind maps and outlines. The confrontation between Nami Tower's "Deep Mode" and Yuanbao's "Deep Search" is relatively fair.

Here’s another advanced question: What major developments will be made in AI-generated video tools at home and abroad in June 2024?

Yuan Bao refused to answer at first, saying that this was a matter for the "future" and told me to do it myself. Even though the training data is not that new, it can be searched online! When I reminded Yuanbao, it finally responded and listed tools such as Keling and Luma.

Then we conducted "in-depth research". Yuanbao gave random answers, probably due to confusion about the prompt words and context, and answered two of my questions at the same time: which AIs can be searched online, and the development of AI-generated video tools.

The understanding ability is relatively lax, but the extended angle makes up for this part. Yuanbao introduced the specific use methods of some of the tools, and thought about the possible risks of AI-generated videos, which is both practical and sublime.

Then use the in-depth mode search of Secret Tower, and the answers outputted in one go are very good. The release time of each tool is marked, and the outline next to it extends to more tools not released in June, as well as the outlook for the entire industry, which is detailed. appropriate.

▲ Picture from: Secret Tower

In fact, during the experience, Yuanbao's "read error" appeared more than once.

When I asked about the world's first smart glasses integrated with GPT-4o, in the "in-depth search" mode, in addition to the glasses themselves, Yuanbao also introduced the applications of GPT-4o in other fields, and even mentioned GPT-4o's competitor Gemini.

If I were a marking teacher, I would feel that the AI ​​is piling up materials and is suspected of missing the topic. Friends, our core is smart glasses. We should either introduce the development history vertically or compare competing products horizontally, instead of crazy expansion of side plots.

Next try Yuanbao's comfort zone. Tencent said that in professional scenarios such as scientific research and finance, the effect of deep search is more prominent. So, here’s an academic question: What sociological theories can be used to explain the recent conflicts between Manner Coffee employees and customers?

This time, Yuanbao answered well, mentioning theories such as alienation, emotional labor, capital surveillance, and workers’ dilemma, and leading us to think: Does the Manner incident reflect broader social issues?

Then I gave the question to the in-depth mode of Secret Tower, and selected academic as the search scope. The answer was more open-minded and more nonsense than Yuanbao. It seemed that the academic content was higher.

▲ Picture from: Secret Tower

Based on the test results of the above questions and combined with the user experience, we can summarize the characteristics of deep search:

1. Not every question requires in-depth search. Sometimes it appears that the information is redundant, the logic is messy, and the focus is not highlighted.

2. Within a limited range of topics, Yuanbao’s in-depth search performance is not as good as Secret Tower’s in-depth mode. (Everyone is welcome to put forward their own opinions)

3. Yuanbao’s reference materials are a mix of third-party websites and public account articles. Within a limited range of topics, the public account ecology does not show obvious advantages. High-quality information sources are important, but model capabilities are also important, and the context is If there is an error in understanding, the output result will be directly lowered.

4. The loading speed of deep search is slow and the number of words is large. It takes more than a minute to generate a complete answer of nearly 3,000 words, which makes people anxious. If you want to read the content while waiting for the loading to complete, the operation is not convenient.

AI search allows fewer questions to be thought about more

Invariably, AI search products are moving in a deeper direction, including "deep search", "research", and "enhancement"…

These unclear prefixes clarify the positioning of AI search. They do not completely replace traditional search, but have their own track, have a deeper understanding of users' query intentions, and provide more accurate and complete search results.

Data shows that in the current use of large model-related products, more than 65% of user needs are focused on improving work and learning efficiency, of which "search question and answer" needs account for as high as 45%.

If you just find a certain website or a certain piece of news on your mobile phone, there is not much difference between AI search and traditional search.

For example, compared to entering "July 4th Generative AI Big Events" and understanding today's big news on Tencent Yuanbao, I think the morning newspapers of Aifaner and APPSO are more valuable.

The human editor's information screening, summarization and even layout aesthetics are better than the AI ​​that fills the screen with words.

But finding articles, videos, pictures, and data related to a certain topic, conducting targeted searches for podcasts, papers, etc., and generating survey reports, mind maps, and PPTs are the strengths of AI search.

AI search is not only search, but also research and summary. It is somewhat like the process in which our editors expand the selected topic into an outline and then write it into an article.

Based on this perspective, I asked Yuanbao: Introduce several domestic and overseas AI spoken language teaching apps similar to NetEase Youdao Echo.

The question is about the fact itself. I would like to know what the extension angle of the ingot will be after an in-depth search. Can some usable viewpoints be drawn out? The results are relatively conventional, still focusing on the features and charging standards of a certain app.

I decided to ask further and relate these apps: What are the common characteristics of these AI speaking teaching apps? What needs does it reflect for users in spoken language learning?

This time, Yuanbao gave me a much better answer, summarizing the characteristics of personalized learning, instant feedback, simulated real conversation scenarios, and rich learning resources.

It is also interesting to extend the perspective, becoming more concrete and perceptible, paying attention to timeliness and user groups: What are the new technology applications of the latest AI speaking teaching app in 2024? Which AI speaking teaching apps are suitable for beginners?

AI's ability to search and find information is certainly powerful, but it also requires our guidance to provide useful "you may also want to know" and "you may also want to ask." During this process, we are also thinking.

The CEO of Perplexity, the leader in AI search, said in an exclusive interview that most problems can be solved by just searching on the Internet, while a few problems require additional thinking.

This point of view is very suitable for solving a college entrance examination essay question in 2024: With the popularization of the Internet and the application of artificial intelligence, more and more questions can be answered quickly. So, are we going to have fewer problems?

The ancient Greek philosopher Zeno compared human knowledge to a circle. Inside the circle is known, and outside the circle is unknown. The more things we know, the larger the circle will be, and the longer the outer edge will be. We don’t know There will be more.

Because of AI, the process of getting answers is faster, and more questions arise. It is also because of AI that we can go further on those few problems that require additional thinking.

It is as sharp as autumn frost and can ward off evil disasters. Work email: [email protected]

# Welcome to follow the official WeChat public account of aifaner: aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo