Two recent arXiv papers explore analogy-making capabilities in large language models (LLMs). The first paper suggests that LLMs provide support for the parallelogram theory of analogy, outperforming human performance in generating analogies that align with this geometric model. The second paper, however, reveals a concerning lack of diversity in LLM-generated analogies, often leading to domain homogeneity and a trade-off between diversity and quality. This second study also identifies differences in how LLMs process information related to analogy diversity. AI
IMPACT Highlights both the potential for LLMs to support complex cognitive tasks like analogy and the current limitations in their diversity and quality.
RANK_REASON Two academic papers published on arXiv exploring LLM capabilities.
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- glove
- Gotit.pub
- Hugging Face
- large-language models
- parallelogram theory of analogy
- Peterson et al.
- Qiawen Liu
- ScienceCast
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
