Puls Dane publiczne, aktualizowane automatycznie
arXiv – nowe publikacje publikacja

General progress substitutes for specific training data in language models

[Przetwarzanie języka] Autorzy: Mark Marosi, Peter Antal

Language models acquire their capabilities from training data, and developers withhold data during training to keep models from learning capabilities that could be misused. How long this delays a capability, and how much data is needed to acquire it later, is unknown. Here we show that the general progress a model makes on the rest of its training data substitutes for the data that was withheld. We withheld the training signal for individual skills in models of chess, Othello, molecular strings, bracket languages and web text. The delay of a withheld skill in larger models was predicted from smaller ones. A chess model never explicitly trained to respond to check ranked the legal responses almost as well as a model that was. When models were then trained on the withheld data, they acquired the skill with a small fraction of the data needed from initialization, and this fraction decrease…

Kategoria arxiv
cs.CL
Kategoria
Przetwarzanie języka
Pdf
https://arxiv.org/pdf/2611.24844v1

Otwórz oryginał w źródle ↗

Więcej z tego źródła

Wszystkie