[Introduction to New Wisdom]The code that Amazon engineers worked on over the weekend was wasted because the project was eventually demoted. Behind the explosion of AI is the crazy internal friction among employees of major Silicon Valley companies. Increasingly intensive event schedules, increasingly incredible deadlines, and useless AI product demonstrations prepared for the board of directors… AI engineers from large companies who have been forced to “involve” are already feeling suffocated.
After the explosion of AI, engineers in Silicon Valley have been exhausted and miserable by the “involution”!
Advertisement
Abandoning the code you worked so hard on over the weekend because the project was de-prioritized is all in vain.
Everyone is chasing after each other, striving to release products faster than the competition. Everything is geared towards speed. Leaders frantically issue orders but are indifferent to the actual impact of the project.
In order to rush the AI project, inexperienced and untrained members are brought in one after another; on the other hand, many people are frantically working internally before the deadline. Even if they are next to technical experts, they have no chance to learn from them…
The above are all the weird things happening in the major Silicon Valley companies where generative AI is booming.
Advertisement
“Involution” never stops
For example, an Amazon engineer. Late last year, after a few weeks of work, he was planning on taking a relaxing weekend.
However, everything changed when he received a Slack message: his boss asked him to complete a project by 6 a.m. on Monday morning.
So, his weekend plans were ruined. He had to cancel friends' gatherings and work around the clock.
The outcome of this matter ended with the project being lowered in priority. This meant that all his efforts were in vain… In fact, this situation was already common for him.
Experts in the field of AI often rush to develop a new feature, but these features are often suddenly suspended, leaving them to respond to the urgent needs of other AI projects.
This Amazon engineer revealed that he once wrote thousands of lines of code for new AI functions in an environment without any error detection.
However, because the code is prone to errors without necessary testing, team members sometimes have to contact each other late at night to fix AI software problems.
In his opinion, Amazon's senior management pays more attention to the formal “I want it all”. However, when they try to copy Microsoft and OpenAI's products as quickly as possible, it is difficult for the product quality to keep up.
For these AI engineers who are trapped in the “internal friction” of large companies, foreign media gave a very vivid metaphor – “Rat race”.
In the “rat race”, people are like rats trying to get the cheese reward, running exhausted but in vain. In order to get the cheese in front of them, they live a repetitive life without time to rest.
Google and Microsoft engineers are all in trouble
Also complaining and suffering are engineers from Google and Microsoft. Fearing that their technology will fall behind their competitors, they are constantly hanging on the strings and need to roll out tools quickly.
Nvidia CEO Lao Huang said that AI is in the “iPhone time”. This means that the entire Silicon Valley is under tremendous pressure.
Project progress is accelerating, and every AI release is desperately trying to get ahead of competitors. What’s even more ridiculous is that leadership is not concerned about the actual impact of many projects.
The above is not the practice of a particular company, but a general trend in the entire industry.
One Google employee said she felt like she could finally take a breather after about six months of intense work.
However, under the company's product development strategy of “building aircraft while flying”, pressure is still increasing.
An Amazon AI engineer said that his team was temporarily pulled in in order to catch up on a project that was behind schedule. But the problem is that the members have neither experience nor relevant training…
In order to push employees, management often holds “inspiring” speeches: “Your work will revolutionize the entire industry!”
Whether you are an engineer or someone in another position, there is a feeling that your job is increasingly focused on meeting investor expectations and staying competitive in the industry, rather than solving actual users' problems.
Moreover, in the pursuit of development speed, employers are also ignoring the impact of surveillance and other negative effects that AI may cause.
They commonly face burnout due to long working hours, tremendous pressure, and ever-changing job requirements.
Many employees have chosen to leave the AI department or are looking for another job because they cannot tolerate this high-pressure and fast-paced work environment.
Yes, this is the hidden side of the generative AI gold rush.
To stay competitive in a market expected to generate more than $1 trillion in revenue over the next decade, technology companies are racing to develop a variety of chatbots, AI agents, and image generators, investing billions of dollars in training LLMs. On the other side, there are employees of major Silicon Valley factories who are suffocated.
Giants are all crazy about AI
And executives of big technology companies are not shy about promoting the impact of AI on their major decisions to investors and employees.
Microsoft's Chief Financial Officer Amy Hood mentioned in an earnings call this year that the company is adjusting its manpower to prioritize AI. Moreover, Microsoft will continue to prioritize investing in AI because it is “a key factor shaping the next decade.”
The same goes for Meta CEO Xiao Zha. He spent much of his recent earnings call talking about products and services, as well as the latest progress on Llama 3.
“I firmly believe that in the next few years, we should invest heavily in developing more advanced models and the world's largest AI service,” Xiao Zha said.
At Amazon, CEO Andy Jassy also told investors that the opportunities for generative AI are unprecedented, so capital investment needs to be increased and this opportunity must not be missed.
“I think it's rare that any of us in technology have seen opportunities like this, at least since the advent of cloud computing, or even since the advent of the Internet,” Jassy concluded.
Speed, everything is for speed!
Now in the AI race, these major companies are laying off employees while doing their best to recruit more AI experts.
Eric Gu, a veteran employee who has worked at Apple for four years, participated in key projects including the Vision Pro headset.
He said that he felt more and more strongly that his development was very limited. Although there are talents and great experts everywhere around me, I have no chance to learn from them.
“Apple is very product-focused, so we are always under tremendous pressure to work efficiently, launch products quickly, and add features…”
This fast-paced pressure overwhelmed Eric Gu.
Finally, about a year ago, he chose to leave Apple and join the AI startup Imbue. There, too, he was able to work on ambitious projects, but at a more gentle pace.
A Microsoft AI engineer also revealed that the company is deeply involved in fierce AI competition.
Moreover, in the pursuit of speed, Microsoft also ignored ethics and security guarantees, which resulted in the product being launched in a hurry before the team had time to consider the potential consequences.
He also pointed out that since all major technology companies have access to almost the same data, there is no real competitive advantage in the AI field.
Indeed, Morry Kolman, an independent software engineer and digital artist with more than 200,000 popular projects, said that with the rapid development of AI technology today, it is difficult to judge which areas are worth investing time in.
And this can easily lead to career burnout, because it's difficult to maintain sustained enthusiasm for one thing.
At Google, one AI team member said burnout comes primarily from competitive pressures, tighter timelines and resource shortages, especially budget and staffing.
Although many top technology companies have said they are increasing investment in AI, the manpower required is often difficult to achieve under tight timelines, even at Google.
Rushing output has caused Google to face embarrassing overturns several times.
After the Gemini image generation tool was released in February this year, it was hastily taken offline due to historical mistakes.
In early 2023, Google employees also criticized the company's leadership, especially CEO Pichai. Anyone with a discerning eye can see that Google’s hastily launched Bard was obviously mishandled in order to compete with ChatGPT.
This veteran Google employee for more than ten years said that not only that, the industry is generally cutting costs, and many companies have to resort to large-scale layoffs in order to meet the expectations of investors and increase net profits.
The tight meeting schedule also puts the team under tremendous pressure.
On the AI team's schedule are the Google I/O Developer Conference in May 2023, Cloud Next in August, and another Cloud Next conference in April 2024.
The intervals between these activities are much shorter than before. For a team that needs to roll out features on a meeting timeline, this means a lot of pressure.
The same pressure exists among government agencies and startups.
One AI researcher at a government agency said that even though the government has been slower to act, he still feels pressure to move quickly. Because now the influence of generative AI has broken through the circle and affected all circles.
The same goes for startups. Ayodele Odubela, a data scientist and AI policy advisor, mentioned that some startups have received investment from large venture capital and are working overtime while the iron is hot. What these investors expect is a return on investment of up to ten times.
Use AI for the sake of using AI
In addition, a large part of the work of AI engineers at major manufacturers is to use AI just for the sake of using AI, rather than to solve business problems or directly serve customers.
A Microsoft AI engineer said that many of the tasks he has been exposed to are just contributing materials to the hype of AI and have no actual application value.
For example, for problems that obviously do not involve generative AI, we must find ways to use large language models to solve them, even if this will make the efficiency lower and the cost higher.
Another software engineer working at a major Internet company was also transferred to a new team studying LLM because “AI is too popular.”
The engineer with many years of machine learning experience believes that current work in the field of generative AI is filled with a lot of false promises and excessive publicity.
From the outside world, it seems like major progress is being made every two weeks, but in fact everyone is repeating the same work.
For example, he often needs to prepare new AI product demonstrations for the company's board of directors within three weeks, even if these products are actually “useless.”
In addition, in order to please investors and obtain funds, he also specially created a web application. Of course, this had nothing to do with the work the team was doing, and no one ever used it after the demo was completed.
A product manager at a financial technology startup said that senior management wants to launch some AI-powered solutions, but they don’t know what problems they are targeting.
For example, one project he participated in was to repackage the algorithms the company had been using as “artificial intelligence.” Furthermore, a ChatGPT plug-in has been developed for customers to use.
An AI engineer working at a retail surveillance startup said he was the only AI engineer among the 40 people in the company.
Here, in addition to being responsible for handling all AI-related tasks, he also needs to face “fundamentally impossible demands” put forward by investors who don't understand AI at all.
Today, he is tortured and wants nothing more than to leave his job and go to graduate school, and then conduct independent research and publish the results.
The speed is too fast and it is easy to overturn
As mentioned earlier, under the pressure of launching products quickly, major manufacturers have reduced routine testing and shelved the verification of AI accuracy.
However, AI projects that are rushed to catch up with competitors can easily overturn. Let’s take Google Gemini image generation, which has been involved in “discrimination” controversy more than once, as an example.
For example, when a user asks to generate “German soldiers in 1943,” it will provide images of people of various skin tones wearing German military uniforms of the time.
When generating “19th-century U.S. Senators”, images of black, Latino, and indigenous women were even given.
However, the first female U.S. Senator was a white woman who served in 1922. This clearly ignores the true history of racial and gender discrimination.
In this regard, Odubela said that as AI technology iterates faster and faster, prudent thinking and rigorous evaluation are more important than ever. However, some major manufacturers not only seem not to care, but are even doing the opposite. .
References:
https://www.cnbc.com/2024/05/03/ai-engineers-face-burnout-as-rat-race-to-stay-competitive-hits-tech.html