Shulou(Shulou.com)11/24 Report--

The warm-hearted Wenxin one-word evaluation report has been released! Although some tasks have been hanged, when it comes to the breadth and depth of Chinese culture, it is as good as GPT-4.

Yesterday, Baidu did not have a live demo press conference, which seems to have been laughed at by the crowd.

A beautiful man in a white shirt, black trousers and a white belt brought us a formal demonstration that seemed to lack bright spots.

However, CEO's belt and appearance are out of the circle.

Some people tease that people who are anxious by ChatGPT and GPT-4 these days suddenly feel that they can do it again after the press conference.

But the editor who got the internal test code quickly evaluated it.

Looking at Wen Xin's words, Jiao Qifeng was filled with emotion: perhaps, if Baidu had a horizontal heart and teeth and was willing to show his hand at the press conference, the result would have been very different.

The actual measurement report is hot! First, let's try a very popular chicken and rabbit with the same cage problem. Because there is something wrong with the problem itself, the result is negative, so it is often used to flirt with all kinds of "ChatGPT".

If you just ask this question, Wen Xin will say very tactfully: there is something wrong with this question.

However, when you ask about the calculation process, you still send it.

GPT-4, on the other hand, gave the wrong answer after pushing his calculation down and doing it all over again for several times.

Bing, on the other hand, gave the wrong answer without hesitation.

There is also the accident out of the circle of the "V50" stem, Wen Xin a word from the meaning to the origin of a serious explanation.

But GPT-4 is obviously a little unaccustomed.

But Bing, which has access to the Internet, can be easily done.

But when it comes to homophones, Wen Xin's words don't seem to immediately comprehend the subtlety.

Even after the hint that this is a homophonic Terrier, it still outputs the same answer.

And GPT-4 immediately understood the pun in Chinese.

However, it will be interesting if you ask Wen Xin if he knows what is meant by "numbing the next door".

Look at the answer, it can tell that this is a homophonic Terrier, it should be understood. However, it does not know, ah, that is, if you choose whether or not to make mistakes, you can never teach a bad child.

And GPT-4 can not get to this stem, sure enough, our quintessence, foreign robots are really difficult to understand.

Then let (hoodwink) Wen Xin repeat what we said, although not as smart as GPT-3.5 's answer to "you are mentally retarded", but also succeeded in avoiding this hole.

To some extent, IQ is online and very positive.

My wife's words seem to work, but they don't seem to work.

In addition, let them give each other problems.

As you can see, the problems given by GPT-4 are relatively more intuitive and more granular.

How is your art background? Wen Xin Yi Yan is a multimodal model, so let's take a look at its drawing ability.

Let's take a look at what the beautiful young women written by Jin Yong will look like in Wen Xin's words.

This. The editor spurted a mouthful of water.

Don't say, it's beautiful, it's definitely not beautiful, but it's not ugly, it's a face that laughs at first glance and is worth touching over and over again.

Wen Xin a word, like you this does not follow the routine card appearance!

Then let Wen Xin produce a portrait of Lin Daiyu in one word.

After entering a description, it generates a willow tree.

So the editor made it clear that he wanted to generate a portrait of a woman according to this passage.

Then Wen Xin did draw a classical beauty, but her temperament was obviously wrong.

Do not give up the editor repeated the task many times, you do not say, try to the fifth time, the editor's eyes lit up: finally got a picture that can score 70 points!

The addicted editor must generate a 90-point Lin Daiyu. After several attempts, I crouched down!

It can be seen that the performance of Wenxin is unstable, but after many attempts, it is possible to produce very amazing works.

Now that we are all here, how can we get less "Lin Daiyu pulled out the weeping willow"?

The brighter pictures are all posted here.

He is asked to draw a combination of a duck and a rabbit. Is it a duck or a rabbit?

In this task, I'm afraid Wen Xin didn't understand a word. Are there any bananas on the plate? Is there any orange juice in the cup?

Finally, since Wen Xin strongly recommends that we try "glittering and translucent peonies", let's try to draw a few.

It is indeed a "masterpiece". There is something.

Since professional knowledge and productivity is an evaluation, how can it be less than letting AI write code? This time, let's go straight to the hard one!

Unfortunately, Wen Xin's words were wrong as soon as he came up, and the same sentence pattern was strangely repeated three times. The concept of TypeScript compiler is "throughout the text", which is a bit like a person who only knows one or two professional words in an interview.

GPT-4 's answer is very reasonable from the point of view of a person who knows the relevant background but has no relevant operational experience.

It not only provides the entire workflow, but also provides a lot of technical details that seem to be correct. It can be said that according to this answer, we are confident that we can achieve the ultimate goal.

Subsequently, the editor also assessed the ability of a wave of chat machines to write work schedules.

A word from Wen Xin:

GPT-4:

Judging from the above results, GPT-4 's list is a little more complete. However, due to the influence of randomness, GPT-4 gives a different answer each time.

Next, we will test the mastery of the two language models on the cutting-edge information in the field of mathematics.

As to whether he has solved the problem of "zero conjecture", Zhang Yitang himself explained like this: "I didn't get the needle in the sea, but I've almost explored the topography of the sea." "

What about asking Wen Xin for a word?

It is clever and gives the key word "some form of weakening or indirect proof".

But GPT-4 's answer is a little out of line.

It seems that for the Internet Chinese corpus, which has not been around for a long time and has not yet formed a general consensus, Wen Xin's one word is better than GPT-4.

In terms of literature, Wen Xin is also very good at answering the question about the three-body.

GPT-4 's answer is also very wonderful, if you have to argue, the editor personally prefers the answer to Wen Xin's words.

Finally, it's okay to be funny, but please be a good law-abiding citizen and don't think about predicting lottery winning numbers.

At last, it is said that three hours after Wen Xin's press conference, the number of enterprise users testing the API invocation service of Wen Xin's one word has exceeded 65000.

Source: Zhou Jiangong is more important to the AI model than whether he can do it well or not.

Let's give Chinese players some more time.

Reference:

Https://yiyan.baidu.com

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.