Why intent prediction needs more than an LLM
Ryan and Frank Portman, CTO at Yobi, discuss the limitations of next-token prediction for forecasting human behavior and explore Yobi's innovative use of transformers and graph neural networks to create a foundation model of behavior that enables rapid personalization while ensuring data privacy.
MAIN POINTS
- Next-token prediction is not suitable for forecasting human behavior.
- Yobi uses transformers and graph neural networks for behavior modeling.
- Millions of personalization decisions are made per second at Yobi.
- Consumer data privacy is maintained during personalization processes.
TAKEAWAYS
- Yobi's approach differs from traditional chat-style LLMs.
- The foundation model of behavior enhances forecasting accuracy.
- Advanced technology enables high-speed personalization.
- Data privacy is a critical focus in Yobi's operations.