TORONTO — After a year of businesses trying out their artificial intelligence systems, makers of large language models (LLMs) are betting businesses will begin buying the technology at scale. “You’re starting to see this take off,” CEO Aidan Gomez said, kicking off a Collision conference where AI is all over the agenda.   

“Last year, 2023, was the year of the proof-of-concept,” he noted, as companies launched a slew of “little tests” to see what the technology could do, and for whom. Having seen the results, Cohere is seeing firms starting to push pilots into full production. 

The Toronto-headquartered startup sells itself as the right model-maker for enterprise clients with concerns about privacy and the need to operate at significant scale. “We don’t just build a massive model for the sake of building a massive model,” Gomez said. “We build models [the] market can actually consume.” 

Business chatter about generative AI doesn’t yet line up with adoption. At an Ottawa event last month, Ashok Krish, global head of AI advisory at Tata Consultancy Services, noted that firms are turning only a small share of their tests into real-world applications. “They go and post on LinkedIn saying, ‘Hey, we did something with genAI,’ and that’s the end of the whole thing.” 

Back at Collision, Gomez pointed to signs of traction. Co-pilots—tools that help workers write code and copy, among other forms of assistance—are proliferating. And LLM-makers like Cohere are now spending more on the compute needed to run the models for clients than on processing power to train them, Gomez said.

Firms may also come to AI indirectly. Enterprise toolmakers like Cisco, Oracle and Salesforce have invested in Cohere, and are giving their products generative updates. “You don’t need to build those AI features yourself, custom,” Gomez noted.

Cohere reportedly just closed US$450 million in new capital, at a US$5-billion valuation, during a time when the wider venture market looks less than rosy. From the stage, Gomez exhorted founders to keep going. “We need a tech ecosystem,” he said, noting that in Shopify “at least there exists one tech company in this country.” (Cohere clearly hopes to get to that scale someday soon). 

Symbolically, the stakes of the technology are also scaling. At Collision last year, Google bought the most prominent floor space for a booth showing off AI applications. This year, the same spot went to the investment arm of Qatar, one of several nations prepared to spend billions to get into the race.

For potential Cohere clients, it’s go time. Firms need to avoid “a PoC death cycle,” Gomez said, referring to proof-of-concept tests. “At some point you need to commit and start adopting.”