Salvato in:
Dettagli Bibliografici
Autori principali: Sawmya, Shashata, Adler, Micah, Shavit, Nir
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2505.19440
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916758906994688
author Sawmya, Shashata
Adler, Micah
Shavit, Nir
author_facet Sawmya, Shashata
Adler, Micah
Shavit, Nir
contents This paper studies the emergence of interpretable categorical features within large language models (LLMs), analyzing their behavior across training checkpoints (time), transformer layers (space), and varying model sizes (scale). Using sparse autoencoders for mechanistic interpretability, we identify when and where specific semantic concepts emerge within neural activations. Results indicate clear temporal and scale-specific thresholds for feature emergence across multiple domains. Notably, spatial analysis reveals unexpected semantic reactivation, with early-layer features re-emerging at later layers, challenging standard assumptions about representational dynamics in transformer models.
format Preprint
id arxiv_https___arxiv_org_abs_2505_19440
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle The Birth of Knowledge: Emergent Features across Time, Space, and Scale in Large Language Models
Sawmya, Shashata
Adler, Micah
Shavit, Nir
Computation and Language
Machine Learning
This paper studies the emergence of interpretable categorical features within large language models (LLMs), analyzing their behavior across training checkpoints (time), transformer layers (space), and varying model sizes (scale). Using sparse autoencoders for mechanistic interpretability, we identify when and where specific semantic concepts emerge within neural activations. Results indicate clear temporal and scale-specific thresholds for feature emergence across multiple domains. Notably, spatial analysis reveals unexpected semantic reactivation, with early-layer features re-emerging at later layers, challenging standard assumptions about representational dynamics in transformer models.
title The Birth of Knowledge: Emergent Features across Time, Space, and Scale in Large Language Models
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2505.19440