48 min
NatWest's AI Agent Development: A Look Behind the Scenes Credit Shift
-
- Technology
Chris Booth, the product owner for NatWest Group's AI assistant Cora, discusses the accessibility work NatWest has been doing and the journey of improving their conversational AI.
NatWest started by building their own front-end chat interface to make Cora more accessible and usable, allowing users to control such aspects as font size and typing speed. They are now also exploring dynamic interfaces and voice for accessibility to create a more fluid and conversational experience.
Chris talks about the challenges of using large language models in customer-facing environments and he further explores the concept of language models and their role in AI systems.
The speakers go on to discuss the use of prompting in language models and the need for tools to control and assure the quality of the prompt and response.
The conversation then looks into the validation and oversight of AI systems and the speakers discuss the limitations and boundaries of LLMs and the potential impact of multimodal inputs.
Takeaways
NatWest has built their own front-end chat interface to make their AI assistant, Cora, more accessible and usable.
Using large language models in customer-facing environments requires careful governance and risk management.
There is potential for creating a trans-organisational repository of conversational content to improve customer experiences.
Personalised experiences are a key focus for NatWest, and they are exploring ways to leverage AI to provide personalised financial guidance.
Version control is a challenge in AI systems and the use of smaller, more focused models can help address this issue.
Understanding the limitations and boundaries of language models is important when building an AI assistant.
Multimodal inputs have the potential to greatly impact the capabilities of language models.
Agencies, startups, and small businesses can focus on fine-tuning and RAG stages to stay competitive in the AI space.
Sound Bites
"We had big ambitions on making Cora far more accessible and usable."
"We're doing really early stages exploring with mobile. How do we create a much more dynamic, flexible interface?"
"We're using it in a lot of ways at the moment. And I think what's so fun and interesting being with Cora and retail is we have by far the highest bar of governance and risk standards."
"Multiple small models or tiny models will actually allow you to control because you can keep them small, you can keep them local and they'll do the job for you."
Chapters
00:00 Introduction to Chris Booth
06:50 The Journey to LLMs
14:53 The Idea of Artificial Sentience
35:19 Understanding the Limitations and Boundaries of Language Models
41:31 The Importance of Continuous Analysis and Fit
For more:
Webio: https://webio.com
Optima Partners: https://optimapartners.co.uk/
NatWest Group: https://www.natwestgroup.com/
Chris Booth, the product owner for NatWest Group's AI assistant Cora, discusses the accessibility work NatWest has been doing and the journey of improving their conversational AI.
NatWest started by building their own front-end chat interface to make Cora more accessible and usable, allowing users to control such aspects as font size and typing speed. They are now also exploring dynamic interfaces and voice for accessibility to create a more fluid and conversational experience.
Chris talks about the challenges of using large language models in customer-facing environments and he further explores the concept of language models and their role in AI systems.
The speakers go on to discuss the use of prompting in language models and the need for tools to control and assure the quality of the prompt and response.
The conversation then looks into the validation and oversight of AI systems and the speakers discuss the limitations and boundaries of LLMs and the potential impact of multimodal inputs.
Takeaways
NatWest has built their own front-end chat interface to make their AI assistant, Cora, more accessible and usable.
Using large language models in customer-facing environments requires careful governance and risk management.
There is potential for creating a trans-organisational repository of conversational content to improve customer experiences.
Personalised experiences are a key focus for NatWest, and they are exploring ways to leverage AI to provide personalised financial guidance.
Version control is a challenge in AI systems and the use of smaller, more focused models can help address this issue.
Understanding the limitations and boundaries of language models is important when building an AI assistant.
Multimodal inputs have the potential to greatly impact the capabilities of language models.
Agencies, startups, and small businesses can focus on fine-tuning and RAG stages to stay competitive in the AI space.
Sound Bites
"We had big ambitions on making Cora far more accessible and usable."
"We're doing really early stages exploring with mobile. How do we create a much more dynamic, flexible interface?"
"We're using it in a lot of ways at the moment. And I think what's so fun and interesting being with Cora and retail is we have by far the highest bar of governance and risk standards."
"Multiple small models or tiny models will actually allow you to control because you can keep them small, you can keep them local and they'll do the job for you."
Chapters
00:00 Introduction to Chris Booth
06:50 The Journey to LLMs
14:53 The Idea of Artificial Sentience
35:19 Understanding the Limitations and Boundaries of Language Models
41:31 The Importance of Continuous Analysis and Fit
For more:
Webio: https://webio.com
Optima Partners: https://optimapartners.co.uk/
NatWest Group: https://www.natwestgroup.com/
48 min