TD Bank is on a generative AI testing spree—and it has high hopes for the potential of Cohere’s software. 

The bank has signed a deal that lets it put the Toronto-based startup’s large language models to the test, to see how they stack up against competitors in a range of tasks. 

An AI sampler platter: Maksims Volkovs, chief AI scientist at TD, said the bank plans to test “all the leading language models out there” and evaluate them on three criteria: Accuracy, speed and cost. Volkovs said the potential uses for which TD is testing the technology are like those other financial services firms are exploring, such as call-centre support, investment research and human resources. TD’s previously announced AI initiatives include a chatbot that answers questions from customer service representatives about the bank’s policies and procedures.

The Cohere advantage: There are a lot of large language models to choose from, including those from tech juggernauts like Meta, Google and OpenAI. Volkovs said he’s excited about the expertise of Cohere’s staff and its focus on serving business customers. He said he has high hopes for Cohere’s strength in a technology called retrieval-augmented generation, or RAG, which requires the AI to source its answers from a fixed set of documents in an effort to minimize its propensity to “give incorrect answers, very confidently.” 

Volkovs declined to disclose the terms of the arrangement with Cohere. The Wall Street Journal first reported news of the deal.

Bigger isn’t always better: Smaller language models like Cohere’s can be easier to customize and adapt to specific tasks, Volkovs said. “We can preserve the accuracy of a much larger language model, while also being much cheaper and faster to run. That’s the experiment we want to do with Cohere.”

What it means for Cohere: The startup, which recently raised a US$500-million Series D at a US$5.5-billion valuation, markets its AI systems as being particularly well-suited for enterprise applications. Model-makers tend to make money based on how much their clients use their technology, so they are increasingly competing for the business of big businesses. 

TD gives Cohere a marquee financial-services client it can name-drop as it tries to win the business of other large firms. The startup’s president, Martin Kon, has previously said that Canadian companies have been slower to express interest in its technology than U.S. counterparts. Cohere spokesperson Josh Gartner did not respond to a request for comment.

Deep ties: Volkovs joined the bank when it acquired Layer 6, an AI startup he co-founded, in 2018. Jordan Jacobs, a fellow Layer 6 co-founder, left TD in 2019 to focus on his AI-focused investment firm Radical Ventures. Cohere is a Radical portfolio company.

Volkovs said Jacobs introduced him to Cohere CEO Aidan Gomez and the rest of the startup’s team in 2020. TD and Cohere then collaborated on a research challenge for an academic AI conference, he said—“It’s very exciting to see how far they have come from those days.” 

However, Volkovs said TD made the deal with Cohere on the merits of the company’s products, “not the fact that we stayed close to them.”