It was 2018, Ivan Zhang's friend Aidan Gomez (BSc 2018) was working for Google Brain, a research team dedicated to making machines intelligent so they could improve people’s lives. Gomez was particularly interested in one aspect of deep learning: natural language processing, a branch of artificial intelligence that teaches computers to respond to speech and text like humans. In 2017, he had co-authored a paper that posited that a transformer, a deep-learning model that can read large amounts of data at once, might do a better job understanding language than the technology used up until that point.
The concept was revolutionary. Because the transformer model looked at vast amounts of data at once, it could understand the full relationship between words in a sentence or paragraph. That meant it was less likely to misinterpret queries, and more likely to return text or speech that was easy to understand—just like a human voice.
By 2019, Zhang had a job at a machine-learning company, but he was itching to put into practice some of the experience he’d gained on the job and through the late-night research he’d done with Gomez. “That’s when I pitched the idea: We should start something,” says Zhang. “I didn’t know what, but I knew I didn’t know many people who had the same work ethic as me or Aidan.” Gomez agreed and told Zhang to start sending him ideas, even if they were trash, so they could come up with a viable business.
In the end, it all came back to transformers. A talk Gomez gave on AI in 2019 piqued the interest of members of the Toronto venture-capital firm Radical Ventures. They asked him if he had any cool start-up ideas. Gomez said yes: He could take the transformer architecture he wrote about a few years earlier and apply it to the entirety of the internet to teach machines how to speak like humans. The more information the transformer had to train on, the better it could do a wide range of tasks, like classifying data, generating content and even detecting toxic comments on social media.
Radical Ventures loved the idea and immediately wrote a seed cheque. In 2019, Gomez and Zhang launched their company, Cohere. Zhang quit his job and started building systems to grab data off the internet so he and Gomez could feed it to their models. By January of 2020, they were joined by another partner, Nick Frosst (BSc 2016), and three additional engineers.
The company raised US$125 million in Series B funding earlier this year. It’s among the businesses and entrepreneurs who are leading research and development in natural language processing, a technology that is powering chatbots and moderating content, learning how to talk and text like people by absorbing all of the best and worst the human brain has to offer. It’s a technology already in the process of changing your life—even if you don’t know it yet.