Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper investigates the counterintuitive phenomenon where larger language models underperform smaller ones on certain tasks, finding that excessive verbosity in larger models introduces errors. Through causal intervention, they show that constraining large models to produce brief responses significantly improves accuracy, reversing performance hierarchies on mathematical reasoning and scientific knowledge benchmarks. This demonstrates that large models possess superior latent capabilities masked by universal prompting, highlighting the need for scale-aware prompt engineering.
Bigger language models are often dumber because they're too verbose, but a simple prompt hack can flip the script and unlock their hidden potential.
Standard evaluation protocols reveal a counterintuitive phenomenon: on 7.7% of benchmark problems spanning five datasets, larger language models underperform smaller ones by 28.4 percentage points despite 10-100x more parameters. Through systematic evaluation of 31 models (0.5B-405B parameters) across 1,485 problems, we identify the mechanism as spontaneous scale-dependent verbosity that introduces errors through overelaboration. Causal intervention experiments demonstrate this reflects correctable prompt design rather than fundamental capability limitations. Constraining large models to produce brief responses improves accuracy by 26 percentage points and reduces performance gaps by up to two-thirds. Most critically, brevity constraints completely reverse performance hierarchies on mathematical reasoning and scientific knowledge benchmarks, with large models achieving 7.7-15.9 percentage point advantages over small models -- direct inversions of the original gaps. These reversals prove large models possess superior latent capabilities that universal prompting masks. We validate findings through three independent contamination tests and demonstrate inverse scaling operates continuously across the full parameter spectrum, with dataset-specific optimal scales ranging from 0.5B to 3.0B parameters. Our results establish that maximizing large model performance requires scale-aware prompt engineering rather than universal evaluation protocols, with immediate implications for deployment: prompt adaptation simultaneously improves accuracy and reduces computational costs.