There are several pieces of information worthy of attention regarding the opening of Baidu Wenxin Yiyan to the whole society and the newly reconstructed AI native applications:

1. The opening of Baidu Wenxin Yiyan: Baidu Wenxin Yiyan is a semantic understanding model developed by Baidu, which understands user input and automatically generates text responses. This openness means that Baidu will allow a wide range of users and developers to use Wenxinyiyan’s functions, thereby applying semantic understanding technology to more scenarios.

2. Newly reconstructed AI native applications: This means that Baidu has comprehensively improved and upgraded its AI applications to improve user experience and functional performance. Specifically, applications may undergo improvements in interface design, function expansion, performance optimization, etc. to better meet user needs.

Let me talk about some practical experience:

Picture-reading answers

When I first opened Wen Xin Yiyan, the first feature that attracted me was chat. There was an icon to upload pictures on the bar, and I immediately selected a photo to upload. After that, some function tips will pop up, including: looking at pictures and writing poems, V50 copywriting, text assistant for Moments, etc. I chose a text assistant and entered a light and humorous style.

Hey guys, I was shocked by this exaggerated copywriting. This accompanying text is simply letting myself go. I couldn’t imagine what people around me would think of me if I sent this copy to Moments. I was so frightened that I had to tell them to keep it cold, concise and with fewer words. It also understood my embarrassment and responded with eight words: enjoy the food and feel the beauty. Wonderful, wonderful!

Discovery: More built-in answer templates

The discovery function on the mobile terminal should be said to be the biggest highlight of Wen Xinyiyan, which can be said to have lowered the threshold for using AI a lot. I used to share the website I built with the interface with my friends. The biggest problem was that others didn’t know what to ask and how to ask. Wen Xin solved this problem very well and put the capabilities of many models into scenarios. .

From the perspective of modal capabilities, there are mainly three types: text-based text, text-based text, and picture-based text. First, let’s try out the copywriting of Xiaohongshu store to try out the effect.

This result is quite surprising overall. It can be seen that Wen Xinyiyan has worked hard on the Prompt project. It basically meets the basic elements of Xiaohongshu Tandian copywriting: format, expressions, tags, and the ability to extract and analyze keywords is pretty good. However, I also tried using this prompt to write other types of Xiaohongshu copywriting, such as picnics, but the effect was completely ineffective. It means that this prompt is specially optimized for the scene of visiting stores and does not have the ability to migrate. Overall, the quality of prompts is pretty good, and the AI’s answering capabilities are pretty good.

AI Painting

Next, let’s test the ability of AI painting. Let’s start with two painting words that have been complained about for a long time: busy traffic and braised lion’s head. It was said that I could draw cars, water, horses, dragons, and real lion heads before. Let’s see if my ability has improved now.

What’s surprising is that both of them are drawn correctly. I don’t know whether it’s because the translation function has been upgraded, the drawing ability has been upgraded, or it’s been optimized specifically. In short, there are a lot less flaws now. (It feels less fun, haha).

From the understanding of painting and the effect of drawing, it should be said that I can get a passing grade. Now, I don’t expect it to draw well. If there is any article that needs pictures, use this It is also a simple and easy choice to draw. However, what is worth complaining about is that the lower right corner of each painting will be marked: AI painting. This is a bit uncomfortable to use, and you have to edit it yourself.

In addition, the painting does not support adjustment through continuous dialogue. You can only enter a complete description to repaint.

Coding ability

Finally, let’s compare Xinyiyan’s coding ability. This is an important manifestation of AI ability and can also be an important part of productivity.

Here is a requirement for writing Python code for merging Excel files. Wen Xinyiyan's writing is relatively concise and elegant overall. The quality of his answers is not much different from ChatGPT, and it is still usable. level.

(I will not compare it with GPT4 here, there must be some differences)

Summary

Before I planned to write this article, I read a lot about Wen Xinyiyan The articles used include many ironic stories. I thought that after using it, there would be many flaws, but it didn't. Although there is still some gap between Wen Xinyiyan's ability and large foreign models, this mobile app is relatively sincere in general. The overall use is very smooth, the answers are very fast, and it has enough high-quality Prompt template, there will be many thoughtful tips between questions and answers, and the overall experience will be good.