Editor: Layan
More and more scientific research talents have gone to work in large science and technology factories, and have bid farewell to scientific research since then. It's just because the big factories give too much!
In the field of computer science, whether to engage in engineering or scientific research has always been an easy choice.
However, at the end of the day, programmers are also workers. So for most people, between getting a higher salary and pushing the academic community forward, they should choose the former.
And in terms of income, tech giants have never been stingy with spending money on talent – all kinds of salaries that make ordinary workers jaw-dropping. This will undoubtedly make many capable people choose to leave academia and go to big factories.
As a result, the AI godmother, Li Feifei, can't sit still.
A large number of brain drains, from doing scientific research to doing engineering, how can this be done!
So in a speech, she directly "advised" US President Biden: You should quickly allocate more money to scientific research talents!
Invest more money
Li Feifei told Biden to invest money to build a "big warehouse" of computing power and datasets across the country, so that AI researchers can catch up with technology companies.
As a professor at Stanford University, Li Feifei must always be concerned about the academic circle. As a result, she has taken up the banner and stood at the forefront of scholars and policymakers who share her views.
Li Feifei's worries are not unreasonable.
At present, the richest universities in the United States are far less than those of technology giants such as Meta, Google, and Microsoft.
You know, these big tech companies spend billions of dollars in the field of AI every year, which opens up a gap with universities.
For example, Meta aims to purchase 350,000 custom chips, or GPUs, to meet the huge number needed to train AI models.
By comparison, Stanford's natural language processing group has only 68 GPUs.
350,000 and 68, this gap goes without saying.
For colleges and universities, there is no GPU, no computing power, no data, what should I do?
As soon as I hugged my thighs, I saw that the employees of large factories were several times their salaries.
It doesn't matter if you don't watch it, after reading it, many people have the idea of running away.
This also confirms Li Feifei's statement: the star talent of AI academia is losing a lot.
Judging from the results, in the whole of 2022, technology companies have created a total of 32 well-known machine learning models in the industry, while universities have only made three.
You must know that 8 years ago, the situation in 2014 was still the opposite - most of the breakthroughs in the AI industry were completed by universities.
The researchers analyzed the possible future evolution of this situation from a professional point of view -
AI scholars will be more concerned about whether the research can be implemented, that is, whether it can be commercialized.
Last month, Meta's CEO Mark Zuckerberg announced that the company's independent AI research lab would be closer to Meta's production team, ensuring that the two departments are somewhat "aligned."
Li Feifei said that at present, the resources and talent reserves of the public sector (i.e., universities) lag far behind the industry. This can have a profound impact on the focus of the entire industry. Corporations are profit-driven, while public research is more concerned with the well-being of the public.
Li Feifei himself has been in Washington, D.C., trying to find new investments. She met Arati Prabhakar of the White House Office of Science and Technology Policy many times, met with the media at high-end restaurants, and visited officials in charge of AI law on Capitol Hill.
However, from the perspective of big tech companies, they are sometimes willing to contribute to national public projects.
Eric Horvitz, Microsoft's chief scientist, has said that they have always attached great importance to sharing progress and resources with their academic colleagues.
And the U.S. government is working hard — last year, the National Science Foundation announced a $140 million investment in seven university-led national academies for artificial intelligence.
Topics include how to use AI to address climate change, mitigate the impact of climate change, and education in the AI era.
However, some scholars in the industry still said that the strength and speed of this kind of help are not enough.
In recent years, for example, major technology companies have been racing on popular tracks such as chatbots and biographic models, and whichever company recruits better talent will be better off the competition.
Computer science professors at many colleges and universities have been poached by high salaries. And it's not just about money, it's also about researching topics that are more interesting than what you used to study at universities.
In 2023, a report shows that 70% of AI PhDs are in the private sector. That's more than three times what it was 20 years ago.
Where are the drawbacks
Following the above logic, large factories have money and resources, and colleges and universities are relatively scarce, so colleges and universities can only hug their thighs.
In fact, this model is not much of a problem on the face of it.
For example, in 2020, 40% of the papers published at the world's major AI conferences had at least one scientist involved.
The company will also sponsor doctoral students from universities to conduct research on related topics.
Jane Park, a spokesperson for Google, also said that Google's position supports that the private sector and universities should work together to advance the development of AI. Google will actually publish its research results on a regular basis to benefit the wider AI community.
However, the dependence of universities on enterprises is deep. Neil Thompson of MIT's Computer Science and Artificial Intelligence Lab said the demand for advanced computing power will only skyrocket as AI scientists continue to compress more data to improve the performance of their models.
And behind this fact, there is another fact - the dependence of universities on enterprises will only get deeper.
Without the resources, funding, and data to support them, any researcher will be left behind and will not be able to do more in-depth research.
In big companies like Meta and Google, their own AI labs used to operate in a similar way to universities. It's up to AI scientists to decide which projects to undertake.
In the past, there was a clear distinction between academic employees and engineering employees. For research-oriented employees, the criteria for judging them are the same as those of universities, all of which are based on which influential papers have been published and what significant breakthroughs have been made.
But now, the competition is getting tougher, and there are countless competitors on the same track. All tech companies are feeling a sense of urgency that they have never felt before.
As a result, the distinction between academia and industry has become increasingly blurred, research freedom within companies has been weakened, and market dominance has slowly gained the upper hand.
To put it simply, what can be implemented immediately and what can bring benefits to the company immediately will be carried out.
At the practical level, Google announced last year that it would combine its two AI research portfolios into one, named Google DeepMind.
In terms of research, Google has also adjusted the model to transform the product first, and then share the paper. The intentions of the ensuing are self-evident.
The same goes for Meta.
The basic AI research team, formerly known as FAIR, was transferred to Reality Labs, and some of the researchers on this team were later transferred to the generative AI product team.
We can see that the current trend is that pure research is not valued by companies, or that companies still have to focus more on industries that can actually bring profits.
Do employees of large factories really make a lot of money?
Now let's take a look at what kind of high salaries are attracting academia.
According to the salary tracker Levels.fyi, the median salary of an AI research scientist at Meta climbed from $256,000 in 2020 to $335,250 in 2023.
In three years, the median salary alone has risen by nearly $100,000.
And the more capable people will certainly earn more than the median, and the increase will be much greater.
Ali Ghodsi, CEO of AI startup Databricks, said that as long as you have a PhD and many years of experience developing AI models, engineers with this qualification can even get a super high salary of $20 million in four years.
And even if we look away from the top salary, the salary level of the entire computer industry is still high.
Two of the top three median annual earnings within five years of graduation are computer-related.
Computer engineering ranked first, with a median of $80,000. Computer science ranked third, with a median of $78,000.
Even with one more layer of analysis, in the second-ranked chemical engineering, semiconductors also contribute a lot of high income. It's also about computers.
Resources:
https://www.washingtonpost.com/technology/2024/03/10/big-tech-companies-ai-research/