Summary of the first year of artificial intelligence in 2023: the thunder is loud, the rain is small, and the future is promising

author:Business breakfast

Since the birth of ChatGPT, the entrepreneurial nerves of China and even the world have been vigorously stirred. I remember at the beginning of the year, when people didn't mention chatgpt or AI, they seemed to be outdated. China's "innovation" bigwigs who copy to c have entered the AI model, the vice president of a certain regiment, a certain Sichuan Intelligence, and the stalls of coal mining for a while have started AI entrepreneurship, and the 100 model war is reminiscent of the 100 regiment war of the year, and some people call 2023 the first year of artificial intelligence, which is not an exaggeration.

Summary of the first year of artificial intelligence in 2023: the thunder is loud, the rain is small, and the future is promising

2023 is almost over, and it's time to summarize this year, the so-called first year of artificial intelligence.

Thunder: The wave of AI is coming, and everyone is saying that the new industrial revolution has arrived

The gears of time are turning to the end of 2022, and at the beginning of 2023, no one would have thought that a sensational AI curtain would be opened so quickly. After the launch of ChatGPT 3.5, the number of registered users exceeded one million in just a few days, and by the end of January 2023, the number of people had reached a terrifying 100 million, making it the fastest-growing consumer app in history. Its anti-heaven chat function breaks the traditional inherent machine dialogue mode, so that the so-called mechanical has a real human taste, many people smell the smell of new technological revolution, all kinds of bigwigs began to stand on the platform (grab the microphone) into the AI, and began to talk about large models when they meet people.

In March 2023, Meituan founder Wang Huiwen, who has been retired for three years, announced that he would build a Chinese version of ChatGPT with $50 million.

In March 2023, Baidu released Wenxin Yiyan, which Baidu calls it: a knowledge-enhanced large language model.

In April 2023, Wang Xiaochuan established Hyakukawa Chino.

What other bigwigs are in the back or for attention?

Ali has AI Qianwen、Tencent mixed yuan model、Jingdong launched the Yanxi model、iFLYTEK released the Xinghuo cognitive model, etc.,In addition to these,There are many more to list one by one,Let's look at the following picture to know how crazy the 100-model war is。

Summary of the first year of artificial intelligence in 2023: the thunder is loud, the rain is small, and the future is promising

When it comes to thunder, I have to say one thing, that is, one thing that Chinese people love to do: brush the list.

So many large models with large models appeared, who is powerful in the end, then hit the list, superclue made a Chinese large model list, which gave everyone the opportunity to compete on the same stage, superclue supports multi-dimensional ability evaluation, don't they all say that they are powerful? Come and try, as a result, this group of domestic problem makers tried to try. But they played a hand, which was to brush the questions.

Every time you look at the update of the superclue list, you will find that the first brother on the list often changes, once you get the position of the first brother on the list, one thing you immediately do is to promote chatgpt, especially chatgpt4.0, taking SuperCLUE, CMMLU and C-Eval three lists as examples, chatgpt almost did not get the first position on the list, and many times the ranking is close to tenth.

The reason why the large model running score list presents the chaos of "alternating between the first and largest brothers" is that many large models respond to the list test by brushing questions. There is a question bank for the large model list test, you read that right, it is a question bank. The question banks of the major model lists are almost one-way transparent to manufacturers, that is, there is a so-called "benchmark leakage". For example, the C-Eval list had 13,948 questions at the beginning of its launch, and due to the limited question bank, there was a situation where some unknown large models were directly asked to "clear the customs" by brushing questions.

This gives the domestic large model a chance, and it must not be able to beat ChatGPT, but the ability to be on the list is definitely more than ChatGPT. As a result, the big brother on various lists kept being brushed out.

A big reason why everyone is keen on brushing the list is that at the moment when artificial intelligence is popular, publicity is very important, and the first is traffic. As for whether you can really play ChatGPT, it's not something you need to pay attention to.

No matter what the reason for swiping the list is, it can be seen from everyone's enthusiasm for swiping the list how popular the large model is.

With so many bigwigs on the platform, so many companies entering the big model, and so many people brushing the list, artificial intelligence has earned enough attention as soon as it appears.

Light rain: After a busy year, the movement is getting smaller and smaller, and the garbage is gradually cleared

From the beginning of the year to the end of the year, as an AI observer and practitioner, the author has experienced a small AI cycle. At the beginning of the year, a bunch of people poured in, bringing money and vowing to build Chinese large models, and by the end of the year, these bullish people seemed to have disappeared a little, and many large models were no longer publicized, and the movement was getting smaller and smaller. In addition to Baidu, the Chinese large model seems to be ironclad and persistent, and other small garbage has gradually been cleaned up.

Baidu's determination and action to make a big model is still very large, and Wenxin has begun to gradually embed it in Baidu applications, for example.

Picture intelligence app

As shown in the figure below, Baidu's AI picture assistant provides functions such as removing watermarks, and the author tested it, and the effect is not bad.

Summary of the first year of artificial intelligence in 2023: the thunder is loud, the rain is small, and the future is promising

At present, except for some pictures with copyright that cannot be modified, other pictures can be modified for free using Baidu Image Assistant, which is very convenient.

Let's take a look at operai, in addition to the launch of ChatGPT 4.0, there is a GPTS similar to Apple's appstore, and then there is a wave of palace fighting, there is a problem in the large models at home and abroad, the rain is small, and the landing application scenarios are scarce, most of which are concentrated in:

Chat to relieve boredom, write copywriting (ad generation), mapping, customer service (more applications), video generation (the more popular Guo Degang speaks English), etc.

Although the large model data is becoming richer and more and more applications, there are still more and more applications in the field of speaking and writing, which makes its application scenarios narrower. If you ask a person in the tech circle who knows ChatGPT and knows about large models, you ask an ordinary person and he doesn't know at all. Unlike Alipay, adults and children know it.

Many people have gradually realized this problem and begun to look at the future of large models from the perspective of industrial applications.

The future is promising: The future of AI is promising

Many people talk about the power of the Industrial Revolution as an example, the competition between horse-drawn carriages and automobiles. The automobile is a revolutionary application for the horse-drawn carriage, but the engine technology of the automobile is definitely not able to compete with the horse-drawn carriage alone. It was the implementation of engine technology that built the automotive application and defeated the carriage. The same is true for artificial intelligence, which must be implemented in application scenarios and form industrial products in order to stand out in the competition between new and old ones. In the future, the application of AI will be the support for the healthy development of AI. Here's an example:

The author recently contacted a Shanghai-based artificial intelligence company, BodyAPT, to do AI clothing custom generation, through their technology, you can use your mobile phone to scan the front and side of the human body, only two body scanning photos can be used to calculate the human body net data according to their AI algorithm, and the accuracy is extremely high. The author experienced that the shoulder width data BodyAPT measured 46 cm, and the manual measurement was also 46. Not only that, they use the collected data to publish directly, which means that clothing customization is a term, and the human body data is designed into a pattern, and the clothing factory can make clothes according to the pattern.

Summary of the first year of artificial intelligence in 2023: the thunder is loud, the rain is small, and the future is promising

BodyAPT is directly delivered to the factory with a digital version of the core, and the whole process is all intelligent without manual intervention. From volume to plate making, it only takes a few seconds. Customizing a shirt takes only 2 hours from measurement to factory production. BodyAPT is an AI application that has broken the previous stage that AI could only be applied to speaking and singing, and has been applied to the actual process of the industry, and has carried out an intelligent transformation of the traditional clothing customization process. Truly using AI technology to disrupt the industry.

Recently, I contacted some top scientists in artificial intelligence, talked to them about AI, and found that they no longer purely talk about technology, but focus on landing applications. This is the same as the author thinks, in good engine technology, if you don't make a car, you can't compete with a horse-drawn carriage.

These top AI scientists also mentioned that data is that AI itself is a collection of algorithms, but the optimization of these algorithm models needs to be fed with actual application data, so as to constantly adjust, align, and optimize according to actual data. After the development of BodyAPT, the technology has been continuously verified in the market, and the BodyAPT model has been optimized by various human body shape and posture data, and the actual data has been adjusted to make the human body data measured by BodyAPT extremely accurate.

From BodyAPT, it can be seen that AI large model + specific industry application will be the road to the benign development of artificial intelligence in the future. What can be met is that in the future, there will be more and more AI applications in industries like BodyAPT, which will truly get out of the current bottleneck of "talking, learning and singing" applications.

There is no doubt that the future of AI can be expected, and whether it can be like the transformation of human society by Internet technology in the past 20 years depends on how the actual application of AI in the industry will be in the future, let us wait and see.

Read on