OpenAI research says 80% of jobs will be affected by GPT

As large language models like OpenAI’s GPT-4 become more advanced and enable writing, coding, and doing math more accurately and consistently, it won’t be long before artificial intelligence enters the scene as a tool for everyday use. workplace. In fact, OpenAI itself is betting that the vast majority of workers will have at least some of their jobs automated by GPT models.

In a paper posted to the arXiv preprint server, researchers from OpenAI and the University of Pennsylvania believe that 80% of the U.S. workforce has at least 10% of their tasks affected by GPT, a family of popular large-scale language models powered by OpenAI. They also found that approximately 19% of employees will see at least 50% of their tasks affected. They wrote in the study that GPT risk is greater in high-income jobs but across nearly all industries. They think of GPT models as general-purpose technologies, like the steam engine or the printing press.

The researchers used the O*NET database, a major U.S. occupational database that lists 1,016 occupations with standardized descriptions, to determine the tasks to be measured in each occupation. They then collected human- and GPT-4-generated annotations using rules to determine whether direct access to GPT or assisted GPT-driven systems would reduce the time it takes humans to perform specific tasks by at least 50%. Higher visibility means GPT can reduce the time required to complete a task by at least half, while maintaining high quality work.

"Our analysis suggests that the impact of LLMs (large language models) like GPT-4 may be ubiquitous," the study added. Additionally, the researchers found that higher-paying jobs—which may involve workers performing Many software-based tasks – may face more potential disruption from AI chatbots.

"Our results show that the importance of science and critical thinking skills is strongly negatively related to exposure, suggesting that occupations requiring these skills are less likely to be affected by current language models," the researchers wrote. "In contrast, programming and writing skills were positively correlated with exposure, meaning that occupations involving these skills are more susceptible to language models."

Occupations with the highest exposure include mathematicians, tax preparers, writers, web designers, accountants, journalists and legal secretaries. Occupations that are most disparate or least likely to be affected by GPT include graphic designers, search marketing strategists, and financial managers. OpenAI researchers used various measurements to classify which occupations are likely to be disrupted the most. Professions most affected include interpreters and translators, poets, lyricists and creative writers, public relations specialists, authors and authors, mathematicians, tax preparers, blockchain engineers, accountants and auditors, and journalists.

The researchers also listed the overall expected impact of GPT on different industries, with the greatest impact being on data processing services, information services, publishing and insurance companies, while the least impact would be on food manufacturing, wood product manufacturing and agricultural support activities and forestry .

The white paper also breaks down ChatGPT's impact by industry. Industries most likely to be disrupted include data processing hosting, publishing and secure goods contracts. In contrast, industries known for manual labor — food services, forestry and logging, social assistance and food manufacturing — are among the least potentially affected.

The researchers acknowledge that their study has limitations because human annotators are familiar with the model's capabilities and do not belong to some of the occupations being measured. Another limitation includes the fact that GPT-4 is sensitive to the wording and composition of prompts and sometimes makes up information, so its output is not necessarily the final truth. Of course, it should be noted that OpenAI itself generates these jobs, and as a for-profit company developing artificial intelligence models, it has a high incentive to portray its tools as disrupting industries and automating tasks that ultimately benefit employers.

Still, the report reveals how the GPT model will soon become a common tool. Google and Microsoft have announced that they will integrate artificial intelligence into their office products (such as email, documents) and search engines. Startups are already using GPT-4 and its coding capabilities to reduce the amount they spend on human developers.

"Our analysis suggests that the effects of LLMs like GPT-4 may be widespread," the researchers wrote. “While LL.M. competencies have continued to improve over time, even if we stopped the development of new competencies today, their growing economic impact is expected to continue and increase.”

However, the study itself also acknowledges that there is an inherent bias in trying to summarize each occupation by using simple labels to describe job tasks. The study adds: "It is unclear to what extent occupations can be fully decomposed into tasks and whether this approach systematically ignores certain categories of skills or tasks that are required by default to be competent at the job "

Another problem is that GPT has shown that it can make obvious mistakes, including fabricating information, which necessitates human oversight of the work. This was a factor that the study could not take into account. Additionally, the study only examined whether ChatGPT could reduce the time required to complete various tasks by occupation. That doesn't mean ChatGPT is necessarily smart enough to fully automate certain tasks.

Still, researchers expect ChatGPT and its future iterations will change the way people work. Therefore, society and policymakers need to be prepared. "While LLM competencies have continued to improve over time, even if we ceased development of new competencies today, their growing economic impact is expected to continue and increase," the document adds.

The original paper of OpenAI is attached, please click to read the original text.

*Disclaimer: This article is original by the author. The content of the article is the personal opinion of the author. The reprinting by Semiconductor Industry Watch is only to convey a different point of view. It does not mean that Semiconductor Industry Watch agrees or supports the view. If you have any objections, please contact Semiconductor Industry Watch.

Today is the 3347th content shared by "Semiconductor Industry Observation" with you. Welcome to pay attention.