Is the ChatGPT large model used for criminal investigation and solving cases only a fictional story? Discussion on AI large-scale model enterprise service trends (1)

The story begins.

There was a spasm of pain, the ground was getting wet under his feet, and he knew it was going to rain again.

Criminal policeman Lao Zhou groped his way to the top of the steps, waiting for the bell to ring for the night shift of the electronics factory in the old city. The usual windproof jacket can't keep out the cold and bitter humidity. There is also a strong stench of corruption in this tide from time to time.

There was a lot of noise, the street lights were dim, and the crowd poured out of the tall and rusty iron gates. Most of their expressions were silent and tired, and many people covered their noses and walked quickly, avoiding the stench.

Lao Zhou remained calm.

Suddenly, someone called "Master" softly, with a short but familiar voice.

It seems that Lao Zhou's youngest apprentice, Baxi, went to the police with illness, and did not obey the order of the team to rest.

He leaned into Lao Zhou's ear and said, "Master, listen to who is coughing." Baxi quickly disappeared into the night.

A cough, like poverty, cannot be concealed. Old Zhou couldn't care about anything else, and immediately focused on his ears, all the sounds around him were amplified.

Raindrops, footsteps, tires, spitting...all movements seem to be forced into the cochlea.

There were several consecutive mild coughs, and the sputum was pulling and unwilling to leave the throat.

The pressure hangs over a man in a black baseball cap, who covers his mouth with a fist and tries to suppress an unstoppable cough.

I saw Lao Zhou pushing his legs out of the steps and pounced on the "baseball cap" with his kicking force. Suddenly, Baxi jumped out from nowhere and stepped forward to wrestle. Lao Zhou took advantage of the situation and twisted his arms, the handcuffs clicked. The "baseball cap" was pulled up and left.

Someone in the crowd realized what had happened, but turned around and saw only a few figures leaving in a hurry.

The young police officer shoved and stuffed the "baseball cap" into the back seat of the police car, and then yelled to the colleagues who were driving: "Return the team."

in another car.

Lao Zhou leaned on the passenger seat, turned around and tortured Baxi with his eyes. How did this "good" apprentice know about the secret ambush when he was on sick leave for a week?

Baxi said: "It's the newly installed software. Everyone is focusing on the case. I tried it for a week and I have new feelings every day."

"What software are you talking about?" Lao Zhou was impatient.

Seeing that the situation was not right, Baxi quickly explained: "This software reminds me to wait at the gate of the electronics factory..."

Seeing the suspicious faces of the master and his colleagues, Baxi said immediately without waiting for the master to ask:

"I let this software learn the master's 18-year case-solving data, command it, and then analyze it according to my master's thinking. I also input round after round of new tracking and reasoning information from our team. This thing can not only analyze More clues, even if it is prompted that its analysis is crooked, it can correct itself.”

Seeing the master frowning tightly, Baxi spoke more and more vigorously:

"The suspect of the rainy night butcher may be hiding in the basement of the unfinished villa area. There are many rooms that have been renovated and half stopped, with mottled paint and dust everywhere. It is also a well-known stray cat litter."

"I accessed the medical data of the suspect during his last prison sentence. The software said that he probably had pneumonia, and his airway was hyperreactive. The underground drain at the gate of the electronics factory was stinking. It's no wonder he didn't cough."

Lao Zhou's complexion turned from cloudy to clear, and Ba Xi also breathed a sigh of relief.

"Then how did you find the electronics factory?"

"I fed the 24/7 surveillance video of the victim's home for ten days before and after to the software. It said that the probability of the suspect appearing at night near the factory area was as high as 85%. I came to try my luck and saw you, Master."

After hearing this, Lao Zhou scolded the quintessence of the country, and snatched this individual device like a smartphone. Trying to figure out what the hell this is.

Enter: who are you?

The software replied: "Hello Lao Zhou, I have passed your face verification, and your alarm number is: XXXX. I am the latest artificial intelligence large-scale model software in 2024, named Shentan+."

"You have 18 years of criminal investigation experience, and you have won the national second-level and third-level hero models many times. I want to learn more from you."

(over)

One More Thing

This is the end of the fictional novel "Detective +", and our discussion on the application of large artificial intelligence models is not over.

The inspiration for the novel came from a chat. Lin Kai, the vice president of Zhejiang Uniview Technology, chatted with me for three hours, which was a lot of brainstorming.

Lin is always a veteran of the ToB market and has experienced three security technology revolutions. In my eyes, he is a person who is very close to the needs of customers and actively embraces technological changes.

There are two important issues: one is the manufacture of large models, and the other is the industrial application of large models.

The bulk of the ChatGPT bonus is concentrated on the basic model maker, and the technology must serve the market after all.

The manufacture of large models is extremely difficult, and there is a pragmatic idea that the manufacture and application of large models should be done together.

And what about ChatGPT? It uses a new generation of AI technology to break through the critical point of the application. More than a thousand APPs can be connected to the large model with lightning speed, and the technological change can be completed within a few days.

Duolingo ChatGPT,

The hotel flight is booked on ChatGPT,

Legal advice on ChatGPT...

It is estimated that the Chinese teacher will say that ChatGPT has changed from a noun to a verb.

To use the powerful capabilities of ChatGPT for your own use, you are willing to become an accessory on the API connection line passed by ChatGPT.

It is very unlikely that many domestic industries will directly access the US version of ChatGPT.

So the question is, how does the domestic security industry access the AI ​​large model?

Video surveillance is the "guardian" of public safety.

Under the analysis of the previous generation of artificial intelligence products, the machine can "see" some meanings in the video. For example, human faces and license plates.

This is an obvious clue to solve the case, not only that, I will tell a small story about the hidden case.

Only tell the story, not the name of the company.

The relevant department of a certain province, using an artificial intelligence technology (deep learning graph model) developed by an enterprise, found 30 highly suspicious underground money houses, 16 of which were directly identified as money laundering gangs.

It's all 2019 technology. In 2023, those who have experienced ChatGPT will be shocked by its beyond-expected and empathetic interactive experience.

Let's review the technical capabilities of the large model again:

First, general purpose understanding ability; second, intelligent interactive correction ability; third, powerful continuous dialogue ability; fourth, strong logical reasoning ability.

Okay, here comes Teacher Tan's inference.

The large AI model has a general ability to understand intentions and understand criminal facts. Through observation and induction of criminal history, scenes and techniques, it can quickly and accurately outline the characteristics of criminals in many aspects, and infer more evidence to solve the case.

The AI ​​large model has strong logical reasoning ability. The investigators collect fresh information from the physical world to obtain new research conclusions, and then the investigators collect evidence to track whereabouts and advance the logic of solving the case layer by layer.

The AI ​​large model has memory ability. A person has a criminal record ten years ago and comes out to commit a crime ten years later. The old and new methods of crime are compared and analyzed.

The AI ​​large model also has the ability of continuous dialogue, which makes it a natural "good temper", very patient and easy to deal with high-pressure and high-intensity work.

The GPT-4 generation large model of the American OpenAI company is already multi-modal, processing multiple types of information. In the future, videos of robbers kidnapping hostages and extortion calls will be able to extract ambient sounds that are difficult for human ears to distinguish, such as the sound of running water, friction, and fuzzy human voices.

The future use of AI large models to solve crimes can be compared to "superior detectives", which is completely different from the previous generation of AI systems.

As long as the multi-modal large-scale model technology continues to break through, the plot in the short novel "Detective +" will be just around the corner, and there may even be surprises.

Let me summarize the main points of the full text:

1. The stories described in fictional novels need to be realized with better multi-modal AI models. Forget about small models. To be more technical, at this stage, GPT-4 is a multi-modal model technology, and it is likely to use the Blip2 and Flamingo solutions, that is, splicing two types of models:

Computer vision model (CV model) and large-scale language model (LLM).

Now this technical solution to support "Detective +" is not perfect.

Wait for the technological breakthrough of multi-modal AI large model.

2. There is still a gap between domestic large models and ChatGPT, we must face up to the problem. The industry proposes: "Do models and applications together." This approach is very pragmatic, but in fact there is no way. The application side and customers are waiting, and technically they cannot catch up.

3. Let me tell you that the competition for the application of AI large-scale models has begun.

Is it sudden? Not suddenly, Baidu company Wenxin Qianfan has released it. This time the enterprise version of the ToB business was released by Baidu first. There are advantages to starting early, but for competition, the best is often yet to come.

ChatGPT has created an era, and no one will stop at the starting point. In the future, China's AI model will be a road of differentiated competition.

Where is the way for large-scale enterprise-level applications? The problem is already in the practitioner's field of vision.

 

おすすめ

転載: blog.csdn.net/weixin_44292902/article/details/130006137