LLM-ERM uses LLM-proposed candidate programs plus validation-based selection to learn short program rules (parity, primality, palindromes) from about 200 examples, while SGD-trained transformers overfit.
Title resolution pending
1 Pith paper cite this work, alongside 500 external citations. Polarity classification is still indexing.
1
Pith paper citing it
500
external citations · OpenAlex
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LLM Priors for ERM over Programs
LLM-ERM uses LLM-proposed candidate programs plus validation-based selection to learn short program rules (parity, primality, palindromes) from about 200 examples, while SGD-trained transformers overfit.