Early in their life cycle, AI companies raise money from investors and receive astronomical valuations. Because of this dynamic, many believe that the AI industry is a bubble.
Nick Frosst, co-founder of Cohere, which builds custom AI models for commercial customers, said he doesn’t think the AI industry is in a bubble. While he acknowledges the existence of “froth,” he believes the term “bubble” discredits companies like Cohere that are building useful features for their customers.
“I often see someone using our model who has enabled some completely new feature that wasn’t possible before or automated some process that was really slowing them down and holding everything back,” Frosst said. And that’s tangible value. When you have something that useful, it’s hard to have a complete bubble.
But that doesn't mean Frosst is optimistic about everything the industry is building. He doesn't think AI will ever reach artificial general intelligence, which is defined as human-level intelligence. This view is very different from that of some of Frosst's AI colleagues, such as Mark Zuckerberg and Jensen Huang. He added that if the industry reaches that point, it won't last long.
Frosst said, “I don’t think we’re going to have digital gods anytime soon.” And I think more and more people are realizing that and saying this technology is amazing. It’s very powerful and very useful. It’s not a digital technology god. And that requires changing the mindset about technology.
Frost said that at Cohere they are trying to be realistic about what AI technology can and cannot do, as well as what types of neural networks can offer the most value. Cohere’s approach to building its business model is based on the research work of Aidan Gomez, Cohere’s co-founder and CEO, while working at Google Brain. Naturally, Gomez is recognized for his extensive research on artificial intelligence. He is best known for co-writing a paper that gave AI the transformative model, which ushered in this era of generative AI. However, in 2017, he co-wrote a paper called One Model to Learn Them All. According to Frost, this research found that a large, all-encompassing language model is more useful than small models trained for one task or data from a specific industry.
That core model is currently used by Cohere to create custom models for corporate clients.
As humans, we specialize. We focus on specific areas. But, according to Frosst, the first part of our education is focused on how to use language in general. We spend a lot of time acquiring reading and writing skills. It is not until much later that we specialize in a specific subfield of language. So neural networks also experience a similar situation.
However, while he believes that core models will be successful in their market (among those making these services), he does not believe that companies should rely on their own proprietary models to carry out all tasks, whether they involve consumption, business-to-business commerce or products.
According to Frost, companies that want to successfully use AI need to focus and be aware of what AI can and cannot do.
Frosst said: “We are pretty sensible about the utility and value that this technology can bring – and, to be clear, outsized value.” However, I don’t think it’s going to kill everyone. So we may take a realistic approach that can avoid some of the extreme rhetoric on both sides.
