Saved in:
Bibliographic Details
Main Authors: Ghaffari, Sadaf, Krishnaswamy, Nikhil
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2402.15654
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914691986489344
author Ghaffari, Sadaf
Krishnaswamy, Nikhil
author_facet Ghaffari, Sadaf
Krishnaswamy, Nikhil
contents In this paper, we present an exploration of LLMs' abilities to problem solve with physical reasoning in situated environments. We construct a simple simulated environment and demonstrate examples of where, in a zero-shot setting, both text and multimodal LLMs display atomic world knowledge about various objects but fail to compose this knowledge in correct solutions for an object manipulation and placement task. We also use BLIP, a vision-language model trained with more sophisticated cross-modal attention, to identify cases relevant to object physical properties that that model fails to ground. Finally, we present a procedure for discovering the relevant properties of objects in the environment and propose a method to distill this knowledge back into the LLM.
format Preprint
id arxiv_https___arxiv_org_abs_2402_15654
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Exploring Failure Cases in Multimodal Reasoning About Physical Dynamics
Ghaffari, Sadaf
Krishnaswamy, Nikhil
Computation and Language
In this paper, we present an exploration of LLMs' abilities to problem solve with physical reasoning in situated environments. We construct a simple simulated environment and demonstrate examples of where, in a zero-shot setting, both text and multimodal LLMs display atomic world knowledge about various objects but fail to compose this knowledge in correct solutions for an object manipulation and placement task. We also use BLIP, a vision-language model trained with more sophisticated cross-modal attention, to identify cases relevant to object physical properties that that model fails to ground. Finally, we present a procedure for discovering the relevant properties of objects in the environment and propose a method to distill this knowledge back into the LLM.
title Exploring Failure Cases in Multimodal Reasoning About Physical Dynamics
topic Computation and Language
url https://arxiv.org/abs/2402.15654