From Language to Pixels: Task Recognition and Task Learning in LLMs

Janek Falkenstein, Carolin M. Schuster, Alexander H. Berger, Georg Groh


Abstract
LLMs can perform unseen tasks by learning from a few in-context examples. How in-context learning works is still uncertain. We investigate the mechanisms of in-context learning on a challenging non-language task. The task requires the LLM to generate pixel matrices representing images of basic shapes. We introduce a framework to analyze if this task is solved by recognizing similar formats from the training data (task recognition) or by understanding the instructions and learning the skill de novo during inference (task learning). Our experiments demonstrate that LLMs generate meaningful pixel matrices with task recognition and fail to learn such tasks when encountering unfamiliar formats. Our findings offer insights into LLMs’ learning mechanisms and their generalization ability to guide future research on their seemingly human-like behavior.
Anthology ID:
2024.genbench-1.2
Volume:
Proceedings of the 2nd GenBench Workshop on Generalisation (Benchmarking) in NLP
Month:
November
Year:
2024
Address:
Miami, Florida, USA
Editors:
Dieuwke Hupkes, Verna Dankers, Khuyagbaatar Batsuren, Amirhossein Kazemnejad, Christos Christodoulopoulos, Mario Giulianelli, Ryan Cotterell
Venue:
GenBench
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
27–41
Language:
URL:
https://aclanthology.org/2024.genbench-1.2
DOI:
Bibkey:
Cite (ACL):
Janek Falkenstein, Carolin M. Schuster, Alexander H. Berger, and Georg Groh. 2024. From Language to Pixels: Task Recognition and Task Learning in LLMs. In Proceedings of the 2nd GenBench Workshop on Generalisation (Benchmarking) in NLP, pages 27–41, Miami, Florida, USA. Association for Computational Linguistics.
Cite (Informal):
From Language to Pixels: Task Recognition and Task Learning in LLMs (Falkenstein et al., GenBench 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.genbench-1.2.pdf