Search papers, labs, and topics across Lattice.
The authors investigate how language model scale affects contextual entrainment, the tendency to favor tokens appearing in the context. They find that larger models become better at ignoring false claims (semantic contexts) but worse at ignoring irrelevant tokens (non-semantic contexts). Through scaling laws analysis on Cerebras-GPT and Pythia model families, they show that semantic and non-semantic entrainment exhibit opposite power-law scaling trends.
Scaling up LLMs doesn't uniformly improve context handling; instead, it paradoxically amplifies the tendency to copy irrelevant tokens while simultaneously improving resistance to misinformation.
Larger language models become simultaneously better and worse at handling contextual information -- better at ignoring false claims, worse at ignoring irrelevant tokens. We formalize this apparent paradox through the first scaling laws for contextual entrainment, the tendency of models to favor tokens that appeared in context regardless of relevance. Analyzing the Cerebras-GPT (111M-13B) and Pythia (410M-12B) model families, we find entrainment follows predictable power-law scaling, but with opposite trends depending on context type: semantic contexts show decreasing entrainment with scale, while non-semantic contexts show increasing entrainment. Concretely, the largest models are four times more resistant to counterfactual misinformation than the smallest, yet simultaneously twice as prone to copying arbitrary tokens. These diverging trends, which replicate across model families, suggest that semantic filtering and mechanical copying are functionally distinct behaviors that scale in opposition -- scaling alone does not resolve context sensitivity, it reshapes it.
NVIDIA