Presentation Global AI Implementation by Robby Yan

Embrassing opportunities, addressing chalnges

1. Global AI Implementation:
Embracing opportunities, addressing challenges
By Robbie Yan
Chief Scientist, Advance Intelligence Group
OJK Institute Webinar
July 25, 2024

2. Agenda
Introduction
Ethics Opportunities
2 kinds of AI: Predictive
and Generative
Challenges

5. Predictive AIGenerative AI

6. Generative AI / LLM
“representation” & “association”

9. *technical term is “token”
Association
• Predict next word*
• Fill in the blank:
The cat jumped over the ( )

10. ●I have a pet cat
●I have a pet dog
●The dog jumped over the fence
The cat jumped over the (fence)

11. The number of medals Indonesia will win in Olympics 2024 is ( )
“Emergent” capabilities

12. Planning and tool use
As of 2024, what percentage of Indonesian
population have credit cards?

13. Predictive AI
➔More “traditional” AI
➔Outputs decisions
➔Complementary to generative AI

14. Generative AI +
Predictive AI =
New, exciting
applications

15. Case study
➔Klarna
◆700 full-time agents
◆25% drop in repeat inquiries
◆2 min vs 11 min
◆$40M profit
➔Bank of America's Erica
◆2M+ times/day
◆Helped with inquiry & transactions

16. Opportunity:
Finance foundation model
➔LLM: college graduate
➔Finance foundation model:
major in finance
◆Marketing
◆Product recommendation
◆Credit analytics
◆Customer service
◆Collections
◆Compliance & reporting
➔Data as moat

17. Opportunity: Financial companion
➔Companion app
◆Improves financial literacy
◆Empowers individuals
◆Via chat interface
●Teaches personal finance: How to borrow,
save, invest
●Check credit score & monitor score status
●Compare credit products
●Negotiate debt
➔AI used
◆Generative AI: LLM, prompt engineering,
supervised fine-tuning, direct preference
optimisation
◆Predictive AI: time-series analysis,
constrained optimisation

18. Opportunity: Improve business
productivity
➔Built an enterprise search tool for
internal knowledge
(HR, compliance etc)
➔AI used
◆Generative AI: LLM, retrieval
augmented generation (RAG)
Knowledgebase Assistant
An intelligent chatbot to help
improve the productivity of teams
within an organisation.
AI Agent

19. Opportunity:
Drive customer engagement
➔Runs effective ad campaigns
➔Recommends most relevant
products to consumers
➔Effectively finds the shortest path
◆Between compatible products &
consumers
➔AI used
◆Generative AI: LLM
◆Predictive AI: recommendation systems
(deep FFM), reinforcement learning
(UCB)

20. Challenges
➔Hallucination
➔Security
➔Intellectual property
➔Privacy
➔National security

21. Challenge: Hallucination
LLM is “always dreaming”

22. Challenge: Security

23. Challenge: Security (Deepfake)
Diffusion & GAN models
Will require non-image methods

24. Challenge: Privacy
Google researchers:
LLM can be primed to reveal PII

25. Challenge: Intellectual property

26. Challenge: National security
Open Source Closed Source

27. Summary
Generative AI &
predictive AI
Representation &
association
Opportunities &
challenges

28. Thank you.