Since the release of GPT-4, there has been much debate about whether it will replace humans and which occupations it specifically threatens.
And a paper that has just been released recently can be called a stone shock.
The paper, jointly published by Ali Dharma Academy and Nanyang Technological University in Singapore, is the title "Is GPT-4 a Good Data Analyst?" 》
The authors of the paper actually invited several human data analysts of different levels to directly fight GPT-4.
* Screenshot of the paper
This time, the GPT-4 did not disappoint either.
Here are the conclusions given in the original article:
*Screenshot of the paper (translated)
It is still written rather tactfully, without bluntly saying that GPT-4 has surpassed humans.
But after carefully reading the data in the paper, Fox Girl felt that the day when the human data analyst was replaced was really not far away.
Senior talent easier to replace?
Contrary to most expectations, GPT-4 costs less to replace senior data analysts at about 0.45%, compared to 0.71% to replace junior data analysts.
This may be because senior data analysts earn higher salaries overall.
In other words, it's more cost-effective to have GPT-4 as a senior data analyst.
*Screenshot of the paper (translated)
To understand it in a formula, it is:
GPT-4 salary ≈ Junior Data Analyst salary * 0.0071 ≈ Senior Data Analyst salary * 0.0045
Based on the average salary paid to data analysts in Singapore.
The annual salary of a junior data analyst is 30,000-50,000 US dollars (about 21-350,000 yuan), which is about 350 US dollars (2400 yuan) in GPT-4.
The annual salary of a senior data analyst is 86,000-90,000 US dollars (about 60-630,000 yuan), which is about 400 US dollars (2,800 yuan) when converted to GPT-4.
*Median salary for analysts by data
Good guy, the talents that large enterprises used to rush to ask for suddenly fell to the same cabbage price as ordinary workers.
This is really what the fox girl did not expect.
And pay attention! Two thousand yuan is an annual salary, not a monthly salary!
The fox girl just wants to shout that the top volume king of the contemporary era is none other than GPT-4!
What is the specific strength of GPT-4?
Because this paper mainly discusses the profession of data analyst, Fox Girl first introduces you to what this profession does.
Simply put, it is a person who specializes in industry data collection, collation, and analysis to predict and evaluate the market.
Data analysts also need to present data in an easy-to-understand way, such as charts, curves, etc.
This requires them to master data analysis skills such as SQL, Python, sas, spss, matlab, eviews, etc.
It is necessary not only to understand business, but also to understand management, analysis, design...
*Consider GPT-4 as a data analyst
This profession requires higher quality of practitioners and higher salaries, which is considered a high-paying profession.
So, the debate about whether it will be replaced by GPT-4 is particularly intense.
But to be honest, as soon as the fox girl listened to the description of this profession, she felt that this job was tailor-made for GPT-4.
Isn't extracting and analyzing large amounts of data and then visualizing the data the strength of GPT-4?
In the paper experiment, the researchers designed an end-to-end framework for GPT-4 to automate the entire data analysis process with GPT-4.
* Screenshot of paper data
From preliminary experiments, GPT-4 is able to understand trends and correlations from the data itself without looking at the charts.
To assess GPT-4's data analysis capabilities, the researchers randomly selected 100 questions from different domains, with different chart types and different difficulty levels.
And hired 6 professional data annotators to rate it on a scale of 0-1 or 0-3.
*GPT-4 score
It can be seen that GPT-4 has received full marks in terms of chart type correctness, consistency and fluency, and the aesthetics and complexity are slightly insufficient, and there is still room for further improvement.
The research team then hired 3 real data analysts to do the same task as GPT-4 and compare the scores of the two sides.
*GPT-4 compared to 3 data analyst scores
The first is a senior data analyst with more than 6 years of experience in data analytics.
On most metrics, GPT-4 performs on par with expert data analysts.
While GPT-4 scores lower for correctness than human data analysts, it scores higher for complexity and consistency.
*One of the case studies of GPT-4
The second is a senior data analyst who has been a data analyst for more than 5 years.
Human data analysts surpass GPT-4 in both the information correctness of the data and the aesthetics of the graphs, the correctness and complexity of the insights, so GPT-4 still has the potential for improvement.
*Case comparison of GPT-4 with Senior Data Analyst 2
The third is a junior data analyst with 2 years of experience in data analytics.
GPT-4 scores higher in almost every item and not only performs better on the correctness of data and analysis, but also tends to produce more results.
From these data alone, GPT-4 can already become a replacement for data analysts! And generate results faster, in less than a minute!
Being able to use GPT-4 is equivalent to being able to open and hang
After reading the research of this paper, the fox girl couldn't help but ask:
Are there really jobs in the world that cannot be replaced by artificial intelligence?
The speed of evolution of AI is really not comparable to that of people.
GPT-4 is like a top roll king, rolling and rolling, turning its standards into everyone's standards, and if you can't keep up, you can only be eliminated by it, or make another living.
The question is, is it fair to put humans in the same competition as AI?
Letting GPT-4 enter the game is equivalent to someone in the game opening and hanging, which will break the original rules and order of human society.
But we cannot ban GPT-4 from entering the game, so a new order must be reshaped as soon as possible.
It is hoped that AI can free the hands of migrant workers, rather than rob them of their jobs.
Resources:
Qubit: Ali Dharma Academy: GPT-4 replaces the data analyst with an annual salary of 600,000 yuan for only a few thousand, and the paper has been issued
论文：《Is GPT-4 a Good Data Analyst?》
Author: Xiao Yu