Researchers emphasize the pace of the artificial intelligence industry

For external observer, the situation of artificial intelligence researchers is enviable. They are sought after by technology giants. They went home with a stunning salary. They are in the hottest industry now.

But all of this is accompanied by huge pressure.

TechCrunch interviewed more than six researchers, and some of them asked for anonymous because they were worried about being retaliated. They said that the rapid development of the artificial intelligence industry has caused damage to their mental health. They said that the fierce competition between the artificial intelligence laboratory promoted an isolated atmosphere, and the risky risk also exacerbated the stress level.

“Almost overnight, everything changed,” a researcher told me, “Our work -whether it is positive or negative results -there is a huge impact, from product exposure and financial consequences Let's measure. “

Just in December last year, OpenAI held 12 live broadcasts, and more than ten new tools, models and services were announced during the period. Google made a response with its own tools, models and services through a series of dazzling press releases, social media posts and blogs. The dispute between the two technology giants is noticeable because of its speed -researchers said that this speed is at a high cost.

Grind and hustle and bustle

Silicon Valley is no stranger to the hustle and bustle culture. However, with the booming development of artificial intelligence, the public's recognition of over -work has reached a high level of disturbance.

In Openai, researchers work for six days a week, and it is far exceeding the off -work time. It is said that CEO Sam Altman urged the company's team to transform breakthrough results into public products in a strict time. According to reports, Bob McGrew, former chief researcher of OpenAI, said that professional burnout is one of the reasons for his resignation last September.

The competition laboratory cannot alleviate this problem. The Google DeepMind team of the Google flagship series AI model Gemini once increased from 100 hours a week to 120 hours to repair the errors in the system. Elon Musk's artificial intelligence company XAI engineers often post posts about work in the evening until the early morning.

Why is it so ruthlessly promoted? Today's artificial intelligence research can have a considerable impact on the company's income. Google's parent company Alphabet lost about $ 90 billion in market value due to the above vulnerabilities, which caused Google's Gemini chatbot to generate a controversial description of historical figures.

Kai Arulkumaran, the research director of artificial intelligence service provider Araya, said: “One of the biggest stress is competitiveness, plus time urgency.”

Ranking is higher than everything

Some of these games are very open.

Each month (sometimes even week), artificial intelligence companies will compete for each other on the Chatbot Arena rankings. The ranking ranks the artificial intelligence models such as mathematics and codes. Logan Kilpatrick is the person in charge of multiple Google Gemini development tools. In a post on X, Chatbot Arena “has an important impact on the speed of artificial intelligence development.”

Not all researchers believe this is a good thing. They said that the development of this industry was so fast that they found that their works were facing the risk of being eliminated before they were released.

“This allows many people to question the value of their work,” said Wang Zihan, a robot engineer who worked in an invisible artificial intelligence startup. “If someone is likely to go faster than me, what's the point of what I do now?”

Other researchers lament that the attention of productization is at the expense of academic friendship.

“One of the root causes of stress is that artificial intelligence researchers have shifted from pursuing their own industry research agenda to research (artificial intelligence model) and provide solutions for products,” said Arukumalan. “The industry has hoped that artificial intelligence researchers can conduct academic research in the industry, but now the situation is no longer.”

Another researcher said that it was surprising and distressed that in addition to a few artificial intelligence laboratories that will be opened as the release strategy, open cooperation and discussions on research are no longer the norm of the industry.

Researchers said: “Now people are more and more concerned about commercialization, closed source expansion and implementation, and have not contributed to the scientific community.”

Welcome to the challenge of graduates

Some researchers traveled to their artificial intelligence graduate programs.

The University of Maryland studied artificial intelligence, a doctoral student Gowthami Somepalli, said that the research results were released so quickly, and it is difficult for graduate students to distinguish between fashion and meaningful development. SOMEPALLI said that this is important because she sees that artificial intelligence companies are increasingly preferred to consider candidates with “extremely relevant experience”.

“Doctoral degrees are usually a quite isolated and stressful experience, and the doctoral degree in machine learning is particularly challenging, because the rapid development of this field and the either published or perish”, “Somepalli said. “When many students in your laboratory publish 4 papers, and you only publish 1 or 2 papers a year, the pressure will be particularly great.”

Somapali said she no longer took a vacation after the first two years of the graduate course, because she felt guilty about leaving before publishing any research.

“During the doctoral degree, I have been suffering from an analog replacement syndrome, and I almost dropped out of school at the end of the first year,” she said.

Road

So, which changes (if you have), can you create a less punitive artificial intelligence working environment? It is difficult to imagine that the speed of development will slow -unless so much cash is in danger.

Sipali emphasized a small -scale but influential reform, such as letting his own challenges normally.

“One of the biggest problems … No one discussed their struggle publicly; everyone showed a brave face,” she said. “I believe that if (people) are struggling to see others, they may feel better.”

Professional service company Ernstone artificial intelligence adviser Bhaskar Bhatt said that the industry should strive to establish a “strong support network” to eliminate isolation.

“It is crucial to advocate a culture that attaches importance to work and life, so that individuals can truly leave work,” Bart said. “Organizations should cultivate practical policies such as psychological health and innovation, and formulating reasonable working hours, psychological health daily days, and obtaining consulting services.”

OFir Press, a postdoctoral University of Princeton, proposed to reduce the number of artificial intelligence conferences and “pause” a weekly paper submission so that researchers can rest in the process of tracking new jobs. Raj Dabre, an artificial intelligence researcher at the Institute of Information and Communication Technology of Japan, said that it should remind researchers in a gentle way what is really important.

“We need to educate people's artificial intelligence from the beginning,” Dabre said. “We need to pay attention to more lofty things in family members, friends and life.”