The Fifth Elephant 2023 Monsoon

The last 12 months have seen a relentless pace of innovation in the field of Artificial Intelligence (AI). Each week hundreds of new models and code repositories are released. It’s clearly a wonderful time to be working in the field but the volume and pace do present their own unique challenges. How to develop products that support rapid prototyping whilst frequently changing the models powering the experience? How to quickly take a newly open-sourced model from initial discovery to production in the hands of the users?

At Glance we provide our users (225m) with an AI powered smart lock screen experience featuring personalised premium content across a range of genres including, casual gaming, LIVE game streaming, news, sports, lifestyle, fashion, entertainment, creator-led LIVE content and more. One of the key applications of AI at Glance involves transforming content from one format such as a text article into a rich interactive video. The wave of innovation in Generative AI has had a huge positive impact on this work.

In our talk we will share the lessons learned whilst building LEX – our Content Generation System. We will discuss how this how the design patterns we apply have evolved from the more traditional. ‘Gather requirements for the new feature we may build, debate, define rigid specification, implement, test, deploy’ to a more liquid approach with Machine Learning Engineers deploying new features that they are in effect sponsoring (championing themselves). This required novel approaches to the UX design and complete decoupling of the ML models powering LEX. Model can then be replaced with superior versions as they become available. This allows upgrades such as GPT 3.5 to version 4.0 to occur within minutes. The net effect is LEX continues to grow in terms of performance and functionality with limited developer intervention.

Key topics

1. The Challenge of Rapid Innovation in AI
2. Exploring the fast-paced landscape of AI innovation and its unique challenges
3. Adapting to the constant release of new models and code repositories
4. Strategies for swiftly incorporating new models into the content generation system
5. Empowering Content Generation with Generative AI
6. Leveraging generative AI techniques to create diverse content formats such as blogs, stories, videos, and quizzes
7. Exploring the impact of generative AI on content quality, diversity, and personalization
8. Use cases and examples of how generative AI enhances Glance’s content offerings
9. Transforming Content Curation to AI-driven Creation
10. Evolution of the content team’s role from curating existing content to actively creating it using AI tools
11. The collaborative workflow between the content team and AI systems like LEX
12. Benefits and challenges of AI-driven content creation, including scalability and quality control
13. Not committing to any one model
14. Decoupling models and it’s usage helps with moving at a pace at which AI is involving. Be it LLMs or image models

Key takeaways

1. Understanding the importance of quickly adapting to new AI models and technologies in content generation systems
2. Exploring the potential of generative AI in creating diverse and hyper personalized content in different formats
3. Lessons learned in managing the collaborative workflow between human content creators and AI systems