What type of study is this?

This is a Quantitative Study study.

September 29, 2025Open Access

Large Language Models' Reasoning Stalls: An Investigation into the Capabilities of Frontier Models

View Full Paper

Ask AI

Bookmark

Improvements in reasoning abilities of large language models have stalled over the last nine months.
Following the release of GPT-4, most progress in reasoning ability is linked to hidden system prompts.
The bottom-up strategy showed the highest efficacy among the tested automated theorem prover reasoning strategies.
A low positive correlation exists between correct reasoning and obtaining accurate conclusions from large language models.

Authors

No takes yet. Share an insight, caveat, or question.

Empirical analysis reveals stalled improvements in large language models' reasoning abilities using ATP strategies.

McGinness et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark