AI companies are rapidly attracting investor funds and achieving high valuations early in their development, leading many to label the AI industry as a bubble.
Nick Frosst, co-founder of Cohere, which develops custom AI models for enterprise clients, recently shared on TechCrunch’s Found podcast that he doesn’t believe the AI industry is in a bubble. While he acknowledges the hype, he feels that calling it a bubble undermines companies like Cohere that are delivering genuinely valuable features to their customers.
“Often, I see users employing our model to enable completely new features or automate processes that were previously cumbersome,” Frosst said. “That’s tangible value. It’s hard to call it a bubble when something is so useful.”
However, Frosst is not optimistic about everything being developed in the industry. He doesn’t believe AI will achieve artificial general intelligence, defined as human-level intelligence, anytime soon, which contrasts with the views of some of his peers like Mark Zuckerberg and Jensen Huang. He added that if it does happen, it won’t be for a long time.
“I don’t think we’re going to have digital gods anytime soon,” Frosst said. “More people are realizing that while this technology is incredible and powerful, it’s not a digital god. This requires adjusting our perspective on the technology.”
At Cohere, they strive to be realistic about what AI technology can and cannot do, and which types of neural networks provide the most value. Cohere’s business model is based on the research of co-founder and CEO Aidan Gomez during his time at Google Brain. Gomez is known for his extensive AI research, particularly for co-authoring the paper that introduced the transformer model, which sparked the generative AI era. He also co-authored a 2017 paper titled “One Model to Learn Them All,” which concluded that a comprehensive large language model is more useful than smaller models trained for specific tasks or industries, Frosst explained.
Today, Cohere uses this main model as a foundation to build custom models for enterprise clients.
“We specialize as individuals, focusing on specific fields. But initially, our education is about using language in general,” Frosst said. “We spend a long time learning to read and write before specializing in a subfield of language. Neural networks undergo a similar process.”
Despite believing that larger, foundational models will dominate the market, Frosst doesn’t think enterprise companies should rely on a single model for all tasks: consumer, B2B, and product tasks.
Frosst advises companies aiming to use AI technology successfully to focus on what AI can and cannot do.
“We’re realistic about the usefulness and value of this technology, which is substantial,” Frosst said. “But I don’t think it will lead to the end of humanity. This realistic approach helps us avoid extreme rhetoric on either side.”