The podcast covered NatWest’s journey with their AI assistant Cora, touching on several topics:
Accessibility Efforts
- NatWest heavily invested in making Cora accessible to all users by building their own front-end from scratch, allowing customisation of font sizes, text speed, colour schemes, etc.
Evolving User Interfaces
- They are exploring dynamic interfaces that can generate relevant UI components based on the conversation context, inspired by recent innovations like Google’s demo.
Language Model Integration
- NatWest is using large language models (LLMs) for various tasks like summarisation, answer rewriting via retrieval-augmented generation (RAG), and potentially personalisation. However, they see LLMs as components working alongside other technologies rather than being the sole solution.
Governance and Oversight
- Integrating LLMs into customer-facing applications raises governance concerns around risk, bias, and unpredictability. NatWest is working closely with risk teams to expand existing model risk frameworks to accommodate the unique challenges of these models.
Anthropic’s Intern
- The conversation also touched on Chris Booth’s open-source project InferGPT, aiming to create an AI assistant that “knows you better than your best friend” by personalising to the individual. This raised fascinating philosophical questions around artificial sentience, anthropomorphism, and the notion of personal AI agents conducting conversations on our behalf.
If you wish to learn more about LLM models, please feel free to contact Dan now by filling out the form here, or by contacting his email: [email protected]
View the full podcast here.
