Tracing Knowledge in Language Models Back to the Training Data

By Ekin Akyürek, Tolga Bolukbasi, Frederick Liu, Binbin Xiong, Ian Tenney, Jacob Andreas, Kelvin Guu. 2022

29 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Management Science and Operations Research, Large Language Models, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Yonatan Belinkov
2. Mohit Bansal
3. Hinrich Schütze
4. Yoav Goldberg
5. Douwe Kiela
6. Yejin Choi
7. Samuel R. Bowman
8. Noah A. Smith

In Management Science and Operations Research

1. Hinrich Schütze
2. Tristan Thrush
3. Angelina McMillan-Major
4. Douwe Kiela
5. Yonatan Belinkov
6. Yoav Goldberg
7. Yacine Jernite
8. Alexandra Sasha Luccioni

In Large Language Models

1. Hinrich Schütze
2. Yoav Goldberg
3. Nora Kassner
4. Abhilasha Ravichander
5. Yanai Elazar
6. Douwe Kiela
7. Tristan Thrush
8. Angelina McMillan-Major

In Safety, Robustness & Interpretability

1. Yonatan Belinkov
2. Yoav Goldberg
3. Yanai Elazar
4. Abhilasha Ravichander
5. Hinrich Schütze
6. Amir Feder
7. Douwe Kiela
8. Margaret Mitchell

Citations in the corpus

• Measuring Data (2022, methodology): “Combined with the modern laissez-faire approach to dataset development and usage, which is shallow and limited when it is done at all (Jo and Gebru, 2020), we are currently in the midst of creating, sharing, and training models based on datasets that we know very little about, contributing to behavi”
• Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models (2023, unknown): “Akyürek et al. (2022) also demonstrate that it is possible to trace the factual knowledge acquired by language models back to their training data.”
• Measuring Causal Effects of Data Statistics on Language Model's 'Factual' Predictions (2022, background): “Closest to our work is that of (Akyürek et al., 2022), which examines techniques to trace language model’s factual assertions back to the training data and find that existing influence methods fail to fact trace reliably.”