Experiments by the authors of "Do LLMs Build World Representations? Probing Through the Lens ..." show that fine-tuning and advanced pre-training strengthen the tendency of Large Language Models to maintain goal-oriented abstractions during decoding, which prioritizes task completion over the recovery of the world's state and dynamics.

Authors

• Person: Not available Organization: NeurIPS
  Do LLMs Build World Representations? Probing Through the Lens ...

Sources

• Do LLMs Build World Representations? Probing Through the Lens ... proceedings.neurips.cc NeurIPS via serper

Referenced by nodes (1)

• Large Language Models concept