Little Tech Firms Need Big Allies in the Age of AI

In the age of artificial intelligence, tech's little guys need big friends.

Making a new artificial intelligence system costs a lot of money and takes a lot of computer power, both controlled by the industry's biggest companies.

The tech business loves hearing about garage-based start-ups. From Hewlett-Packard to Google, the stories of companies that started with nothing and grew into giants have inspired businesses for decades.

But the enormous amounts of money and computing power that start-ups need to use today's hottest technology, the artificial intelligence used in robots like ChatGPT and Google Bard, may make these stories of success obsolete.

Aidan Gomez and Nick Frosst left Google in 2019 to start an artificial intelligence company called Cohere in Toronto that could fight with their old company. A few months later, they returned to Google and asked if they could buy the massive amount of computing power they would need to make their artificial intelligence. After Sundar Pichai, the CEO of Google, approved the deal, the tech giant gave them what they wanted.

"It's called 'Game of Thrones.' "That's exactly what it is," said David Katz, a partner at Radical Ventures, the first company to invest in Cohere. He also said that large companies like Google, Microsoft, and Amazon are responsible for the chips. "They are in charge of the computers," he said. "They choose who will get it."

Without the help of "the hyperscalers," who run the vast data centres that can run A.I. systems, it's hard to build a groundbreaking A.I. start-up. And that has put the tech industry's big players in charge — again — of what many people think will be the most significant change in the industry in decades.

Microsoft gave $10 billion to OpenAI, the company that made ChatGPT. It will give most of that money back to Microsoft by paying for time on Microsoft's vast clusters of computer systems. These tools made up of thousands of specialised computer chips, are needed to improve and grow ChatGPT and other similar technologies.

OpenAI's competitors can only keep up with them if they have the same amount of computer power. Cohere just got $270 million in new funding, bringing its total to more than $440 million. Much of that money will buy computer power from companies like Google.

Other start-ups have made similar deals, including a Silicon Valley company called Anthropic, which was started in 2021 by a group of ex-OpenAI researchers, Character.AI, which was created by two of Google's top researchers, and Inflection A.I., which a former Google executive started. Last week, Inflection raised $1.3 billion, taking its total funding to $1.5 billion.

At Google, Mr Gomez was part of a small research team that made the Transformer, the core technology used to make robots like ChatGPT and Google Bard.

Scientists call the Transformer a neural network, a mathematical system that can learn skills by looking at data. Neural networks have been around for a long time. They help run everything from digital helpers that talk, like Siri, to services that translate words instantly, like Google Translate.

With the Transformer, the idea went in a new direction. Using hundreds or even thousands of computer chips, it could look at much more data and do it much faster.

Using this technology, companies like Google and OpenAI started making systems that could learn from a massive amount of digital text, like Wikipedia articles, digital books, and chat logs. As these systems looked at more and more data, they learned to write their papers, blog posts, poems, and even computer code.

Large language models are the name for these systems, which are now used to power apps like Google Bard and ChatGPT.

Before ChatGPT came along, Mr Gomez left Google to start his business with Mr Frosst and Ivan Zhang, another Toronto start-up. The goal was to make big language models that could compete with Google's.

At Google, he and his fellow experts had almost unlimited computer access. He needed something like it after he left the company. So he and his co-founders bought it from Google, which sells cloud computing services that let people use the same chips.

Over the next three years, Cohere built a big language model that is almost as good as any other. It now sells this technology to other companies. The goal is to give companies the tools to build and run artificial intelligence (A.I.) programmes, like chatbots, search engines, and personal teachers.

Mr Gomez said, "The plan is to build a platform that others can build on and try out."

OpenAI has a similar service called GPT-4, which is already being used by many businesses to build robots and other apps. This new technology can look at, make, and change text. But soon, it will also be able to handle pictures and sounds. OpenAI is making a version of GPT-4 that can look at a picture, explain it right away, and even answer questions about it.

Satya Nadella, the CEO of Microsoft, said that the company's deal with OpenAI is the same kind of relationship it has had with smaller companies for a long time. He told The New York Times earlier this year, "I grew up in a company that has always done these kinds of deals with other companies."

As the business world tries to catch up to GPT-4, entrepreneurs, investors, and experts argue about who will ultimately win. Most people think that OpenAI is at the top. But Cohere and a small number of other companies are making similar hardware.

The tech giants are in a good situation because they have the money and resources to push these systems further than anyone else. Google also has a patent on the Transformer, the core technology that Cohere and many other companies use to build A.I. systems.

But open-source software is a wild card.

Meta is another big company with the computing power needed to build the next generation of artificial intelligence. It recently made its latest large language model open source, meaning anyone can use and build on it. Many people in the field think that this kind of free software will make it possible for anyone to compete.

Amr Awadallah, CEO of the artificial intelligence start-up Vectara and a former Google executive, said, "Having the minds of every researcher on Earth would beat any company." But they will still have to pay to use the data centres of a much bigger rival.