By Ian Gormely

The third and final day of this year’s Collision Conference saw talks that outlined AI’s potential to aid in some of the world’s most wicked problems while underscoring that this relatively new technology is still subject to fundamental growing pains. 

Calling climate change “the mother of all risks,” Riskthinking AI CEO Ron Dembo reminded the audience that it will affect everyone, especially people living in poverty. His company, which provides climate risk analytics, hopes to change people’s behaviour by highlighting the financial bottom line for companies who aren’t working to mitigate the risks posed by a warming planet. “Most companies who do not manage climate risk properly will lose, and those who do will win,” he said. 

To this end, Riskthinking AI recently launched a new AI-enabled climate risk engine called Beyond Hindsight. Covering 95% of the world’s listed companies, it exposes the risks climate change will bring to these firms’ businesses, buildings, and supply chains. Produced on-demand in seconds, Dembo said that he hopes the tool sheds light on the end of our climate change tunnel. “We will be able to reprice financial risk into financial markets.” 

Of course, innovative tools and applications like the one Dembo showcased are only possible if companies can access affordable compute to train and run their models. The need for more AI compute infrastructure in Canada was recently called out as part of a $2.4 billion federal government commitment. “The compute required to train the large language models that power generative AI, requires compute power that is 1000 times greater than previous models,” says Vector Institute President and CEO Tony Gaffney. “Currently, demand is growing faster than supply.” 

Trends in large language models (LLMs) currently hold that the bigger the model, the better it will perform, said Akbar Nurlybayev, Co-Founder and COO of Vector Bronze sponsor CentML Inc. But, “current lead times of new compute coming online [are] very long despite the best efforts of chip manufacturers.” 

To continue to build the kinds of LLMs that transform industries, companies, particularly SMEs, need a more “sophisticated” way to use their current capacities more efficiently. With its deep expertise in increasing compute capacity, CentML, whose CEO is Vector Faculty Member Gennady Pekhimenko, can help. The company can increase overall compute capacity for training generative AI models while using the previous generation of GPUs. In doing so, says Nurlybayev, they can help “bring the future forward.” 

Taken together, these two talks speak to AI’s incredible potential and the very real obstacles that still stand in the way of reaching it. Yet, as this year’s Collision Conference underlined, by fostering innovation in both academia and industry we can overcome the most vexing issues to create a more equitable society. Vector is fortunate to have great partners like Riskthinking and CentML are working with us to accomplish this. 

Learn more about what the Vector community was up to at the 2024 edition of Collision Conference.