mallen-2023-decoder-only-models-only

IN premise — summaries/2026/08/24/mallen-2023-when-not-to-trust-s3-collect-popularity.md

Created 2026-08-25T02:58:10+00:00

Mallen et al. (2023) evaluate only decoder-only LMs; encoder-only models using [MASK] fill-in-the-blank probing are explicitly excluded.

Summary

The Mallen et al. 2023 study scoped its analysis to decoder-only language models and deliberately excluded encoder-only models that rely on masked-token prediction. Any conclusions built on that work therefore apply to only one architecture family and cannot be assumed to hold for all LLMs.