We all underestimated GPT-4, it is where the dream begins

In the past two months,
the impact of AI technology has been brought one after another.
Our team
is busy researching various AIC capabilities.
This aspect is really not as strong as when you use these software to get
started immediately. I think there are at least a few products. You can try it out. First , please go to a check GBT plus account and experience the underlying capabilities of GPT4 for yourself. Second , if you are a programming student, if you haven’t taken a picture of the call, use it quickly. The third is that everyone actually has You can use North’s AI to experience how amazing AI-assisted writing is today, and Noson is also used very smoothly in China today, which is really very convenient. The fourth is to apply as soon as possible to queue up for ice skating. The fifth is to use it to generate hand-painted works mean journey I think for students who don’t know how to draw, such as me , it definitely opens up a whole new world. If you look at the above major processes, most of them are open AI and Microsoft things , so there is really no way because of this The main driving force of the next underlying change is the technological change of GPT . Maybe some students have heard of it.



























It is said that check GPT is the iPhone moment of the new era,
that is, the iPhone 10 grams,
which means that it
will detonate a whole new era just like the iPhone back then
. But in my opinion,
the impact of check GPT is actually
greater than that of the iPhone back then.
It should be its own The biggest industrial change
since the mosaic 1.0 browser was launched in 1993 is simple . The iPhone revolution was mainly concentrated in the field of soil c. However, the revolution started by GPT is sweeping the two fields of soil c and soil b at the same time. From the perspective of the breadth of influence, it is more like the impact that the Internet brought to the world when it was created from scratch . The news about this miraculous function is overwhelming , but what you may not know is that the entire AI industry has not really ushered in a turning point until the recently released GPT4 , because it is the first time that artificial intelligence has crossed the threshold of common sense. The so-called common sense is that you see a hot pan you know you should wear gloves before you pick it up or you're on the bus























It is very strange to see an old man give up his seat to a young and strong boy,
but conversely, it is not surprising that a young man gives up his seat to an old man. This is common sense. To learn common sense requires artificial intelligence . You have to know what social norms are, what is common sense in physics, what is common sense in chemistry, and you have to integrate everything you see and know. This ability has really begun to be realized step by step in gpt 4. For example, in the promotional video of OPN, he can say Take this picture and then ask what will happen if the glove falls, and the gpt will answer that the glove will touch the board and the blue ball will fly up . For example, in this picture of the balloon, if you ask him what will happen if he cuts the rope, he will answer It is said that all the balloons will fly away . The most classic example is this one. The user asked, what do you think is strange about this photo , and GPT4 can actually answer that the strange thing about this photo is that a man tied himself to a car. The taxi going forward, and he is still ironing clothes. This is common sense. You must know that common sense has been considered by the AI ​​community as a huge gap that cannot be crossed in the past 10 years.

































But now this gap has been
bridged just like that,
and as long as we cross the step of common sense,
we suddenly find that AI has become really smart.
For example,
GPD4 can directly read papers with pictures.
For example, AI can I
got a very high score in the general test.
I checked the test results published on the OPNA official website.
If I found that GPD4 is in the SET mathematics test,
I can get 700 points out of 800 points in
the SET reading and writing test.
Can get 710 points.
In the GR1 writing test,
a paper with a full score of 6 can get 4 points.
In the GR1 verbal test,
a paper with a full score of 170 can actually get 169 points.
That is to say, he can now take the test
in most North American academic ability tests.
Passed 85% of the candidates
and not only the exam.
Now he has also made a leap
in his ability to complete human tasks. In the OPNI press conference, there was such a scene, that is, the engineer only needs to draw a very simple web page sketch with pen and paper on the scratchpad and then Take a photo and upload it , then the magic moment GPD directly generates the website code in about 10 seconds, and then deploys it, and a webpage comes out










You need to know this workload.
In the past, at least a product manager,
a UI designer,
and a front-end engineer had a three-day workload.
Now, it only takes a college student or even a middle school student to spend
2 hours studying on their own . I ’ve already seen it, and I think it’s quite shocking, right? But in fact, everyone has overlooked one point, that is, GPT4 is actually a model that was trained in August last year. Open AI has been optimizing it for more than half a year before releasing it. Such a shocking technical achievement has actually been realized half a year ago. That is to say, at this moment, when we are still experiencing the various forks of GPT4 , the next more exaggerated version of GPT5 may be close to the training completion state and may even be counting down to release. I can only say that this technological progress is really too fast. There must be some kind of phase change process from quantitative change to qualitative change. The exponential effect is fully reflected. I remember the first time I saw check GPT on the Chinese Internet . The news should be in December last year. At that time , some people in the industry around us who had experienced it already thought it was amazing , but at that time his parameters only held 175 billion




























And now just over three months later,
a model with 100 trillion parameters
can be fully opened to the public.
The comparison of the two numbers
can be clearly seen from this picture.
You may not know it.
From the perspective of brain science
100 The number trillion is a very important milestone.
This number is also the beginning of the emergence of the human brain and
the emergence of human-level intelligence.
This emergence
is a scientific term for complexity.
There is a special sister in my general chemistry class who
said that he
is no longer here . To explain more
, all in all , GBT4 can now achieve a scale
of 100 trillion parameters. This is the first time in human history that the number of neuron connections between an AI model and the brain has reached an order of magnitude. Our human cerebral cortex contains about 14-16 billion neurons. Our cerebellum contains about 55-70 billion neurons, and each neuron in the cerebral cortex, which is responsible for thinking, has an average of more than 30,000 protrusions. So we can simply calculate that the total number of neuron connections in the cerebral cortex is about 14 billion. To 16 billion multiplied by 30,000/2 is about 210 trillion to 240 trillion , so you can see the two numbers of 210 trillion and 100 trillion here


















The damage is already very small , so it is not impossible for AI to start to emerge with general intelligence
at this parameter level, but we can imagine that if the parameter scale of the next version of GPT5 continues to grow exponentially , it is possible to reach the neuron connections of the human brain The number is more than 100 times. What will happen in Tianchi ? Here we can only use a conservative method to linearly extrapolate GPT4 from text to pictures. Then GPT5 should start to open up text and pictures to the video circle. LGPT6 can further open up the three-dimensional world. If this continues, artificial intelligence may be able to understand the four-dimensional, five-dimensional or even 11-dimensional universe. At that time, it will completely enter the information field that humans cannot control. The connotation of what it can produce may not be known at all anyway We can at least know that AI's writing ability, programming ability, drawing ability, reasoning ability, and screenwriting ability comprehensively surpass human professionals. It should only be a matter of time. In fact, there was a forecast report before. By 2030 , it seems that their predictions are too conservative. Follow me to share more IT technicians
































The story of the counterattack of working and starting a business
Thank you for reading!