AAmazon and Microsoft have, until now, stayed somewhat distant from the artificial intelligence arms race. While Google and Meta prioritized developing their own AI models, Microsoft and Amazon invested in smaller technology companies, receiving access to these companies’ AI models in return, which they then incorporated into their products and services.
Microsoft has invested at least $13 billion in OpenAI, the company behind ChatGPT. As part of this agreement, OpenAI gives Microsoft exclusive access to the AI systems it develops, while Microsoft provides OpenAI with the computing power it needs. Anthropic has agreements with Amazon and Google, receiving US$4 billion and even US$2 billion of each, respectively, in exchange for Anthropic making its models available through Amazon and Google’s cloud services platforms. (Investors in Anthropic also include Salesforce, where TIME co-president and owner Marc Benioff is CEO.)
Now, there are signs that the two technology giants are deepening their fight. In March, The Verge reported that Amazon has tasked its AGI team with building a model that outperforms Anthopic’s most capable AI model, Claude 3, by the middle of this year. Earlier this month, The Information reported that Microsoft is training a base model large enough to compete with frontier model developers such as OpenAI.
While there are many types of AI systems that are used in a variety of ways, the big trend of recent years is language models – AI systems that can generate coherent prose and usable code, and that power chatbots like ChatGPT. While younger companies OpenAI and Anthropic, along with the more established Google DeepMind, are in the lead for now, their new big tech rivals have advantages that will be difficult to offset. And if tech giants come to dominate the AI market, the implications – for the concentration of corporate power and whether the most powerful AI systems are being developed safely – could be worrying.
A change in strategy
Throughout the 2010s, AI researchers began to realize that training their AI systems with more computing power would make them more capable. During the same period, the computing power used to train AI models increased rapidly, doubling every six months according to to researchers at Epoch, a research institute focused on AI.
The specialized semiconductor chips needed to do so much computational work are expensive, as is employing engineers who know how to use them. OpenAI CEO Sam Altman stated that GPT-4 training cost more than $100 million. The need for more and more capital is why OpenAI, which was founded in 2015 as a non-profit organization, changed its structure and signed multibillion-dollar deals with Microsoft, and why Anthropic signed deals similar with Amazon and Google. Google DeepMind – the Google AI team that develops Google’s most powerful AI systems – was formed last year when Google merged its elite AI group, Google Brain, with DeepMind. Like OpenAI and Anthropic, DeepMind began as a startup before being acquired by Google in 2014.
See more information: Amazon Partnership with Anthropic Shows Size Matters in the AI Industry
These partnerships have paid off for all parties involved. OpenAI and Anthropic were able to access the computing power needed to train next-generation AI models – most commentators agree that OpenAI’s GPT-4 and Anthropic’s Claude 3 Opus, along with Google DeepMind’s Gemini Ultra, are the three most capable models currently available. Companies behind the border have so far tried alternative business strategies. For example, goal offers fuller access to its AI models to benefit from developers outside the company who tune them and to attract talented researchers who prefer to be able to openly publish their work.
In the April quarterly earnings reports, Microsoft It is Amazon reported bumper months, which they both partially credited to AI. Both companies also benefit from the agreements in that a large proportion of the money flows back to them as it is used to purchase computing power from their cloud computing services units.
However, as the technical feasibility and commercial utility of training larger models became apparent, it became more attractive for Microsoft and Amazon to build their own large models, says Neil Thompson, who researches the economics of AI as director of FutureTech research. project at the Massachusetts Institute of Technology. Building your own models should, if successful, be cheaper than licensing models from your smaller partners and give big tech companies more control over how they use the models, he says.
It’s not just big tech companies that are making advances. OpenAI’s Altman has released his company’s products for a number of large companies that include Microsoft customers.
Who will win?
The good news for OpenAI and Anthropic is that they are ahead. GPT-4 and Claude 3 Opus, along with Google’s Gemini Ultra, are still in a different class than other language models like Meta’s Llama 3, according to a popular chatbot rating site. OpenAI notably finished training GPT-4 in August 2022.
But maintaining this lead will be “a constant struggle,” writes Nathan Benaich, founder and general partner at venture capital firm Air Street Capital, in an email to TIME. “Labs are in the challenging position of being in constant fundraising to pay for talent and hardware, while also lacking a plan to translate this model launch arms race into a long-term sustainable business. As the sums of money involved become too large for US investors, they will also begin to have to deal with complicated issues surrounding foreign sovereign wealth.” In February, Wall Street Daily reported that Altman was in talks with investors, including the UAE government, to raise up to $7 trillion for AI chip manufacturing projects.
See more information: The UAE is on a mission to become an AI powerhouse
Big tech companies, on the other hand, have ready access to computing resources – Amazon, Microsoft and Google account for 31%, 24% and 11% of the global cloud infrastructure market, respectively, according to data from market intelligence company Synergy Research Group. This makes it cheaper to train large models. It also means that even if developing language models doesn’t pay off commercially for any company, technology companies that sell access to computing power through the cloud can still profit.
“Cloud providers are the shovel sellers during the gold rush. Whether frontier model builders make or lose money, cloud providers win,” Benaich writes. “Companies like Microsoft and Amazon occupy an enviable position in the value chain, combining the resources to build their own powerful models with the scale that makes them an essential distribution partner for new entrants.”
But while big tech companies may have certain advantages, smaller companies have their own strengths, such as greater experience in training the largest models and the ability to attract the most talented researchers, says Thompson.
Anthropic is betting that its talent density and proprietary algorithms will allow it to stay at the frontier while using fewer computing resources than many of its competitors, says Jack Clark, one of the company’s co-founders and head of policy. “We will be at the frontier with surprising efficiency compared to others,” he says. “For the next few years, I have no concerns about that.”
If Big Tech wins
It is still an open question whether large technology companies will be able to overcome the competition from their smaller ventures. But if they did, there could be implications for market competition and efforts to ensure that the development of powerful AI systems benefits society.
While it could be argued that more companies entering the base model market would increase competition, it is more likely that vertical integration will serve to increase the power of already powerful technology companies, argues Amba Kak, co-executive director of the AI Now Institute. , a research institute that studies the social implications of artificial intelligence.
“Seeing this as ‘more competition’ would be the most inventive corporate approach that obscures the reality that all versions of this world serve to consolidate the concentration of power in technology,” she writes to TIME. “We need to be careful about this type of manipulation, especially in the context of increased antitrust scrutiny by the UK CMA, the FTC and the European Commission.”
See more information: UK competition watchdog signals cautious approach to AI regulation
Large companies coming to dominate can also be worrisome because the small companies they currently lead were explicitly founded to ensure that building powerful AI systems goes well for humanity, says Anton Korinek, an economics professor at the University of Virginia. OpenAI’s founding goal was “to advance digital intelligence in ways that are most likely to benefit humanity as a whole,” and Anthropic’s goal founding goal The goal was to “make fundamental advances in research that will allow us to build more capable, general, and reliable AI systems and then deploy those systems in a way that benefits people.”
“In a sense, you could say that the AGI labs – OpenAI, Anthropic, DeepMind – were all founded on idealism,” he says. “Large shareholder-owned and controlled companies simply cannot follow this strategy – they ultimately have to produce shareholder value.”
Even so, companies like OpenAI and Anthropic cannot act entirely in the public interest because they are also exposed to commercial incentives through the need to raise funds, says Korinek. “It is part of this broader movement, of this capital in the form of [computational power] it’s becoming the most important input,” he says. “If your training is in the millions, it is much easier to raise philanthropic funding for it. But if your training rounds are in the billions, you need financial returns, the way our economy is currently organized.”
With reporting by Billy Perrigo/São Francisco
This story originally appeared on Time.com read the full story
