Beyond Tokens in Language Models: Interpreting Activations through Text Genre Chunks
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914165733457920 |
|---|---|
| author | Benito-Rodriguez, Éloïse Urdshals, Einar Nasufi, Jasmina Pochinkov, Nicky |
| author_facet | Benito-Rodriguez, Éloïse Urdshals, Einar Nasufi, Jasmina Pochinkov, Nicky |
| contents | Understanding Large Language Models (LLMs) is key to ensure their safe and beneficial deployment. This task is complicated by the difficulty of interpretability of LLM structures, and the inability to have all their outputs human-evaluated. In this paper, we present the first step towards a predictive framework, where the genre of a text used to prompt an LLM, is predicted based on its activations. Using Mistral-7B and two datasets, we show that genre can be extracted with F1-scores of up to 98% and 71% using scikit-learn classifiers. Across both datasets, results consistently outperform the control task, providing a proof of concept that text genres can be inferred from LLMs with shallow learning models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_16540 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Beyond Tokens in Language Models: Interpreting Activations through Text Genre Chunks Benito-Rodriguez, Éloïse Urdshals, Einar Nasufi, Jasmina Pochinkov, Nicky Computation and Language Machine Learning cs.LG Understanding Large Language Models (LLMs) is key to ensure their safe and beneficial deployment. This task is complicated by the difficulty of interpretability of LLM structures, and the inability to have all their outputs human-evaluated. In this paper, we present the first step towards a predictive framework, where the genre of a text used to prompt an LLM, is predicted based on its activations. Using Mistral-7B and two datasets, we show that genre can be extracted with F1-scores of up to 98% and 71% using scikit-learn classifiers. Across both datasets, results consistently outperform the control task, providing a proof of concept that text genres can be inferred from LLMs with shallow learning models. |
| title | Beyond Tokens in Language Models: Interpreting Activations through Text Genre Chunks |
| topic | Computation and Language Machine Learning cs.LG |
| url | https://arxiv.org/abs/2511.16540 |