A battle is brewing for AI chips among major cloud vendors. Google's Trillium, a custom chip for training and running AI models, recently entered preview, and Microsoft's Maia is expected to follow soon.
It has AI chips like Trainium, Inferentia, and Graviton to compete with Amazon Web Services. To draw attention to Trainium in particular, the company is launching a new funding program for AI research.
The new program, called Build on Trainium, will award a total of $110 million to institutions, scientists, and students conducting AI research. AWS will award up to $11 million in Trainium credits to each university in its strategic partnerships, as well as individual grants of up to $500,000 to the broader AI research community.
AWS also says it is establishing a “research cluster” of up to 40,000 Trainium chip research teams, which students can access through self-managed reservations.
Gadi Hutt, senior director of AWS' Annapurna Labs, a chip manufacturing company that AWS acquired in 2015, said Build on Trainium is designed to provide researchers with the hardware support they need to advance their research. He said there was. Grant participants will also have access to Trainium's educational resources and enablement programs, Hutt added.
“Today's AI academic research is severely hampered by a lack of resources, which is causing academic departments to rapidly fall behind,” Hutt said. “With Build on Trainium, AWS is investing in a new wave of AI research led by cutting-edge AI research in universities that advances the state of the art in generative AI applications, libraries, and optimization.”
In fact, academics in the field of AI lack the considerable infrastructure that the tech giants have at their disposal. For example, Meta has procured well over 100,000 AI chips to develop its flagship model. By contrast, Stanford University's natural language processing group uses 68 GPUs for all its work.
But not everyone believes AWS will become a charitable sponsor.
“This feels like an effort to normalize misconduct in academic research funding,” Oz Keyes, a doctoral candidate at the University of Washington who studies the ethical implications of emerging technologies, told TechCrunch. spoke.
With Build on Trainium, AWS will have final say over which projects receive funding. The selection process is opaque. Hutt said only that AWS will allocate funding “based on research merit and need” and “assess program success and outcomes.”
An AWS spokesperson later revealed that a committee of “AI and applications experts” will review the proposals and select “the most impactful and promising projects that will help advance the science of machine learning.” .
There is evidence to suggest that industry-sponsored AI research tends to prioritize research with commercial applications over other research areas. In a recent paper, researchers found that major AI companies produce significantly less research that critically examines the ethical implications of AI compared to traditional research. Furthermore, the co-authors say that “responsible” AI research conducted by large companies is narrow in scope and lacks diversity in topics.
Researchers are seeking legal and technological protections for vendors to scrutinize AI without fear of suspending their accounts or threatening legal action.
Build with Trainium is an advertisement for Trainium. But is AWS trying to attract researchers to its platform from a different angle? I asked if it was. Hutt said that's not the intention, and that the only requirements he has to meet are to publish a paper and to “open source” his research on GitHub under a permissive license.
“There are no contractual restrictions that would make the university the exclusive technology partner,” he said. “What we ask instead is that the results of the research be open sourced for the benefit of the community.”
In any case, it is unclear whether Build with Trainium will play a major role in bridging the gap between AI academia and industry.
In 2021, U.S. government agencies, separate from the Department of Defense, allocated $1.5 billion in academic funding for AI research. In the same year, the global AI industry spent more than $340 billion overall (and not just on research).
Nearly 70% of people with AI Ph.D.s go on to work in private industry, attracted by competitive salaries as well as access to critical computing and data (and the means to process it) . In recent years, companies have stepped up efforts to recruit AI researchers from faculty, offering even larger grants to those with Ph.D.s. Students conducting research.
The final result? Today, more than 90% of the largest AI models developed in a year come from industry, and the number of industry co-authors and published AI papers has nearly doubled since 2000.
Policy makers are taking several steps to address the funding gap between academia and industry. Last year, the National Science Foundation awarded $140 million to launch seven university-led National AI Institutes to investigate how AI can mitigate the effects of climate change and improve education. announced an investment in Elsewhere, efforts are underway to establish the U.S. National AI Research Resource, a $2.6 billion initiative that will provide AI researchers and students with access to computational resources and datasets.
However, it is still a short time compared to corporate programs. And there is little reason to think the status quo will change anytime soon.