|
Towards Culturally Intelligent Large Language Models| title | Towards Culturally Intelligent Large Language Models |
|---|
| start_date | 2026/10/20 |
|---|
| schedule | 14h |
|---|
| online | no |
|---|
| location_info | salle Paul E rdos (A115) & en ligne |
|---|
| summary | Generative AI has transitioned from niche technology to a ubiquitous component of daily life, and models are increasingly deployed in socially sensitive roles across the globe. However, current benchmark-driven development often overlooks whether a model is able to navigate the complex, pluralistic nuances of global cultures. This gap creates a superficial illusion of global utility, leading to functional failures, safety violations, and the systemic risk of cultural erasure. In this talk we will discuss the gap that needs to be investigated for globally usable intelligence. We will define the term culture from the perspective of AI evaluations and explore a principled framework to transform cultural intelligence from a contested concept into a measurable and controllable construct that can be applied to evaluate a model's ability to navigate and reason between complex cultural norms and nuances. |
|---|
| responsibles | Bawden |
|---|
Workflow history| from state (1) | to state | comment | date |
|---|
| submitted | published | | 2026/10/12 07:42 UTC |
| |
|