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

A few days after GPT-4 was released, OpenAI told everyone directly that GPTs were general purpose technologies and that 80% of Americans 'jobs were affected. If you want to save your life, look at these 34 big "iron rice bowls".

As soon as GPT-4 was launched, OpenAI released a 35-page paper official announcement:

80% of Americans are affected by AI.

Researchers say GPT will be a "general purpose technology" like steam engines or printing presses.

In this way, the OpenAI industrial revolution is really hammered!

arxiv.org/ abs / 2303.10130 For about 80 percent of Americans, GPT affects at least 10 percent of their work tasks, the paper said.

At the same time, in about 19% of jobs, 50% of tasks will be automated to some extent by AI.

And the impact cuts across almost every industry. The higher the education, the higher the income, the better the AI is at the work done by the "white-collar", which means that it is more likely to be replaced by AI.

Is there a job that AI cannot replace? Yes!

OpenAI also lists 34 "iron rice bowls" that are not affected by AI, telling blue-collar people that you are "safe."

These include dishwashers, bartenders, plumbers and landscaping workers.

Source: After studying the 34 iron rice bowls in detail, it seems that the end of the universe is not working well.

The more you earn, the more dangerous it is? GPT will affect most human work, dare to draw this conclusion, who gives OpenAI confidence?

The paper first presents the performance transition diagram from GPT-3.5 to GPT-4, so that everyone can see the development speed of the large model (GPT-4 technical report is transferred):

On various professional tests and academic benchmarks, GPT-4 performs at human-level. He scored almost full marks in all major exams, including GRE, and swept all kinds of benchmarks.

In this way, replacing you is not unreasonable. So OpenAI came up with a definition.

In the latest paper, OpenAI directly refers to GPTs as general-purpose technologies (GPTs).

General purpose technologies are those that meet three core criteria-technologies that improve over time, are ubiquitous in the economy, and can generate relevant innovations.

There is no doubt that GPT has improved over time and thus meets the first criterion.

Two other criteria are also gradually being met, the researchers said in the paper. Early qualitative evidence suggests that LLM adoption and use is gaining ground.

Next, let's look at a wave of paper conclusions.

The occupations most affected by LLM are tax preparers, interpreters and translators, survey researchers, proofreaders and scribes, and writers.

Jobs with 100% exposure include mathematicians, tax preparers, quantitative financial analysts, writers, web and digital interface designers.

There are even blockchain engineers.

At the beginning of the chapter, I listed occupations that were completely unaffected by LLM: mainly manual workers.

This is very consistent with Moravik's paradox, which is that robotics that reliably automates most manual tasks is many years away.

How do you define this "exposure rate"?

The percentage of exposure indicates whether accessing GPT, or a GPT-driven system, reduces the time it takes a human to perform a task by at least 50 percent, the paper said.

Of course, OpenAI also said that the tasks above may not be fully automated by GPT.

Functional representation of human scores (x-axis) and GPT-4 scores (y-axis) shows a high degree of consistency in GPT exposure by occupation.

The paper also noted that the higher the pay, the higher the chance of being automated by LLM. The good news, though, is that jobs that require science and critical thinking skills have lower exposure rates.

The specific linear relationship is shown in the figure below, which depicts the exposure of human evaluators and various occupations assessed by GPT-4 to LLMs.

These plots compare occupations at risk for GPT exposure to the logarithm of occupational employment and the logarithm of median occupational annual salary.

While there are some differences, human and GPT-4 assessments suggest that occupations with higher wages tend to be affected by LLMs.

By region of work, those with a bachelor's degree or higher were more likely to be exposed to GPT than those without a bachelor's degree.

How did the numbers in the above table come about? The authors of the paper defined four levels of "exposure scoring criteria."

E0: No exposure. For example, tasks requiring high interpersonal interaction, precise measurements, detailed review of visual effects, physical labor, etc.

E1: Direct exposure, using LLM can reduce the time to complete the task by at least half.

E2: Reduce workload by at least half through LLM-driven applications.

E3: Similar to E2 in view of the exposure of image capabilities, but requires multi-modal GPT-4 with visual API.

Netizen: After reading the paper, I knew I still had to wash dishes! Nvidia scientist Jim Fan calls on everyone to face up to this "elephant in the room"-will LLM take over our work?

Some netizens sent out soul torture:

"For the past 10 years, I've believed that AI can free humans from tasks that don't require brains and propel the world toward a more creative future. However, with the advent of models such as Stable Diffusion, artists were eliminated. If human logic and creativity could be replaced, what jobs do you think would be left?"

The limits of language are the limits of my world.

Some netizens concluded that this article revealed the tip of a terrible iceberg: most of the threats and impacts LLM brings to us are still below sea level and have not appeared at all.

John Carmack recently tweeted that he had received a private letter and hoped his answer would be seen by everyone.

The students who wrote privately mainly stated their concerns: (This may be the problem deep in the hearts of many children nowadays)

I am passionate about computer science (especially software engineering) and would like to pursue a career in this field. But I have to worry about the development of artificial intelligence (such as GPT-4), so that the future of programming work will still exist. I know it's hard to predict 10-15 years into the future, but my main concern is that I've probably worked so hard for nothing that Al will eventually eliminate me from future work. Do you have any ideas?

Carmack replies that you need to build complete "product skills" for the job and use the best tools. Today it may be handwritten code, but later there may be Al assist and you will get better and better.

Software is just a tool to help people get things done, and many programmers never understand that. Focus on the value delivered and don't focus too much on the specifics of the tool.

Foreign media said that the OpenAI paper may have some alarmist elements.

Because OpenAI is touting its own GPT, it needs to be portrayed as a tool to disrupt industries and automate tasks so employers can start using it to reduce costs.

Some netizens even said that the paper was basically meaningless because it did not propose a time limit for its prediction.

In a recent interview with ABC NEWS, Sam Altman said,"ChatGPT should be seen as a tool, not a substitute for any job."

He explains that humans have demonstrated this time and again by adapting to different types of technology. Human creativity is limitless and we can create new jobs and find new things to do.

The irony, however, is that OpenAI is now selling GPT as a general purpose technology.

In October 2017, the cover article of The New Yorker,"Human Future Can Only Help Robots," painted a picture of the future:

A bearded young beggar sits begging on the streets of futuristic Manhattan as robots throw screws and nuts into his cup. His dog looks at the robot dog with surprise and concern.

Perhaps OpenAI wants to tell us that this future is not far off.

References:

https://arxiv.org/abs/2303.10130

https://analyticsindiamag.com/openai-publishes-yet-another-lame-paper/

This article comes from Weixin Official Accounts: Xinzhiyuan (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.