With generative AI now on the agenda of most c-suite executives, Scotiabank Global Banking and Markets recently hosted a ‘Demystifying AI’ event for business leaders eager to learn about this highly anticipated technology.
With the provocative event theme, “Are LLMs for everyone becoming a reality?” attendees didn’t have to wait long for the answer from featured speaker, Jonathan Rosenbluth of Cohere, a leading AI platform provider of large language models (LLMs) for enterprise.
“A few years from now, none of our jobs will be the same, and we will all be interacting with intelligent knowledge assistants everyday through multiple touchpoints,” declared Rosenbluth, Cohere’s Director of Product Strategy. “We are just now seeing how this technology can be deployed across organizations. Not only will it change how existing businesses engage with their customers and employees, but it will also spur whole new business models and change industries.”
From lightbulb idea to countless applications
Rosenbluth has an enviable understanding of the real-world opportunities for LLMs since, over the past three years, he helped Cohere develop its innovative LLM platform, which many large enterprises are now adopting in their operations.
In fact, Rosenbluth recounted for the audience how interest in gen AI skyrocketed over the last year and how applications will evolve over the coming year – as models are connected with enterprise data sources and applications, with their usability and impact to increase dramatically.
That’s exactly what Toronto-based Cohere is doing, by building cloud-agnostic, secure, fit-for-purpose, gen AI models and platforms for enterprises around the world. Focused on solving real-world problems, Cohere trains large language models, which collect text and produce outputs like ChatGPT, and embedded models that actually understand text, including complex semantics and nuanced language meanings, and convert it for customized business uses.
Putting AI innovation within easier reach
Noting the “evolutionary path of LLMs, and democratization of this technology that will help redefine how we work,” event moderator Divya Goyal, Scotiabank’s Equity Research Analyst, asked Rosenbluth how he foresees the technology’s roll-out accelerating in the near-term.
“So much technology today has become democratized, that now it’s becoming much more accessible, very quickly. We’re helping democratize this technology by giving a software engineer or a growing company the ability to build powerful applications with AI that was previously [out of reach],” responded Rosenbluth. He pointed to Cohere’s own chat offering, Coral, which can be connected to an enterprise’s data sources to empower sophisticated search and retrieval capabilities and build intelligent conversational agents that can generate accurate responses and perform routine tasks.
“Today, many of us can’t imagine doing our jobs without the internet, and that’s how we are going to feel about AI in a couple of years. We will have AI-powered assistants with us at every moment of the day, to answer questions, draft and review documents, and remind us of tasks we have to do,” explained Rosenbluth. “As knowledge workers, it will transform how we operate cognitively since we will have so much information at our fingertips. That will have a huge impact on employee and customer engagement, and it can transform sectors like financial, retail, legal and medical services.”
Ready to tackle potential pitfalls
Amid the praise for this technology, Scotiabank’s Goyal challenged Rosenbluth to address some of the potential pitfalls of gen AI. They include concerns about cybersecurity and data privacy risks, as well as fears that LLMs could produce inaccurate, biased responses due to flawed design or unclean, incomplete enterprise data.
Rosenbluth responded that Cohere is the only company that has signed both the US Administration’s updated voluntary commitments to manage AI risks and Canada’s voluntary AI Code of Conduct. Additionally, the company is actively engaged with Canada’s Ministry of Innovation, Science and Economic development for development of effective regulations and maintains stringent data security policies, invests heavily in platform security, and conducts private deployments, so client data is not exposed. He urges that, “It’s important that an organization is able to protect its confidential, proprietary information.”
“In terms of ‘hallucinations’, the industry term for generating the wrong information, we are working to improve model accuracy in part by applying ‘retrieval augmented generation.’ We enable our customers to connect our models to a wide range of external sources and provide clear citations so the information can be verified,” explained Rosenbluth.
In addition, through fine-tuning, an organization can apply their own data to further train the model, including very industry-specific and technical subject matter. This enables the model to learn about specific domains. Overtime, through practice and a continuous feedback loop, the model gets better and better at its assigned tasks.
Rosenbluth does emphasize that LLM effectiveness depends on companies being able to access high quality, diverse data sets: “Thoughtful organizations today are focused on creating tools to collect data from their internal sources, and put it in the right form, so they can train and finetune gen AI models to drive the best results.”
Start now, to test new uses and resolve risks
Despite the challenges, Rosenbluth contends that wading into the LLM waters today will pay off tomorrow: “Some business leaders will find endless reasons that they are not ready to build something, but you won’t win with this technology if you wait on the sidelines. Instead, you have to start building today. The most important thing is to pick a use case, make sure you are protecting your data in the process, and start building.”
He adds that, from the viewpoint of an investor trying to assess which companies stand to benefit most from gen AI, “It’s great if a company has an AI strategy, but you need to ask what are the actual products that they have built or are building, even if it’s just in beta testing. The companies that are developing this muscle now are the ones that will win.”
As the event concluded, participants no doubt possessed a clearer sense of AI’s potential impact across companies and industries. As Rosenbluth stated confidently, “We are really in the early innings of this technology. This is a wildly exciting moment to be either investing in this space or adopting this technology to drive products and services forward.”
Learn more about Artificial Intelligence