Using Large Language Models for Social Simulations
Event details
| Date | 20.06.2024 |
| Hour | 11:00 › 13:00 |
| Speaker | Ivan Zakazov |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Tanja Käser
Thesis advisor: Prof. Robert West
Co-examiner: Prof. Martin Schrimpf
Abstract
With the seminal Simulacra work, we are now entering the era of LLM-based human behaviour simulations. But can we reliably shape personalities of the subjects of these simulations? And will it ever be possible to cut the (often prohibitive) costs of running those? With this work, we aim to provide solutions to these challenges.
Background papers
1. Generative Agents: Interactive Simulacra of Human Behavior — https://arxiv.org/abs/2304.03442
2. Evaluating and Inducing Personality in Pre-trained Language Models — https://arxiv.org/abs/2206.07550
3. LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models — https://arxiv.org/abs/2310.05736
Exam president: Prof. Tanja Käser
Thesis advisor: Prof. Robert West
Co-examiner: Prof. Martin Schrimpf
Abstract
With the seminal Simulacra work, we are now entering the era of LLM-based human behaviour simulations. But can we reliably shape personalities of the subjects of these simulations? And will it ever be possible to cut the (often prohibitive) costs of running those? With this work, we aim to provide solutions to these challenges.
Background papers
1. Generative Agents: Interactive Simulacra of Human Behavior — https://arxiv.org/abs/2304.03442
2. Evaluating and Inducing Personality in Pre-trained Language Models — https://arxiv.org/abs/2206.07550
3. LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models — https://arxiv.org/abs/2310.05736
Practical information
- General public
- Free