Reaction

Three variables are claimed to predict verbatim memorization in language models

Christopher Potts says Jing Huang’s Hubble model suite experiments use frequency, recency and hardness; hardness has comparatively low predictive value.

12 Sources, ,

Dot plot addressing whether memorized sequences can be predicted without training the full model. Headline: “Largely yes, with Frequency and Recency doing most of the work.” Four rows show AUC-ROC (0.5 = chance, 1.0 = perfect) for a linear predictor of verbatim extraction, one mark per Hubble model (1B or 8B params; 100B or 500B tokens). All three features: 0.82–0.94. Frequency alone: 0.81–0.93. Recency alone: 0.82–0.90. Hardness alone: 0.51–0.55, barely above chance. In every row the 1B/500B model scores lowest and the 1B/100B model highest. Footnote: held-out strings, mean of 5 seeds; Frequency = copies of the string, Recency = steps since last copy, Hardness = NLL on models that never saw it.

TLDR

Christopher Potts says experiments by Jing Huang using the Hubble model suite can accurately predict verbatim memorization from frequency, recency and hardness. He says recency remains highly predictive when frequency and hardness are held fixed, while hardness—a proxy for whether an example fits the model when encountered—has comparatively low predictive value.

Today's Rank

—

Not ranked yet

Combined views

9.4K

12 Sources, first seen 4h ago

Three variables are claimed to predict verbatim memorization in language models

TLDR

Combined views

12 Sources

Combined views

Sentiment

Summary

Sentiment

Summary

Sentiment

Summary

Sentiment

Summary

12 Sources

12 Sources

12 Sources