2026-09-24T08:25:09Zhttps://dspace.mit.edu/server/oai/requestoai:dspace.mit.edu:1721.1/1567942024-09-17T04:02:58Zcom_1721.1_7582com_1721.1_7581col_1721.1_131023
Andreas, Jacob D.
Meng, Kevin
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
2024-09-16T13:49:41Z
2024-09-16T13:49:41Z
2024-05
2024-07-11T14:36:44.224Z
https://hdl.handle.net/1721.1/156794
This thesis investigates the mechanisms of factual recall in large language models. We first apply causal interventions to identify neuron activations that are decisive in a model’s factual predictions; surprisingly, we find that factual recall corresponds to a sparse, localizable computation in the MLP weights of the GPT models we study. Harnessing this insight, we then develop methods for efficiently and surgically inserting up to 10,000 new memories into a transformer; these methods perform well in terms of both generalization and specificity. We conclude with some directions for future work.
M.Eng.
Massachusetts Institute of Technology
Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)
Copyright retained by author(s)
https://creativecommons.org/licenses/by-nc-nd/4.0/
Interpreting and Editing Memory in Large Transformer Language Models
Thesis
application/pdf
Master
Master of Engineering in Electrical Engineering and Computer Science
Publication
unknown
unknown
<Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="15592946-496b-4e3a-82fa-c287f1058c67">
<Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type>
<Title>Interpreting and Editing Memory in Large Transformer Language Models</Title>
<PublishedIn>
<Publication>
</Publication>
</PublishedIn>
<PublicationDate>2024-05</PublicationDate>
<Authors>
<Author>
<DisplayName>Meng, Kevin</DisplayName>
<Affiliation>
<OrgUnit>
</OrgUnit>
</Affiliation>
</Author>
</Authors>
<Editors>
</Editors>
<Publishers>
<Publisher>
<DisplayName>Massachusetts Institute of Technology</DisplayName>
<OrgUnit />
</Publisher>
</Publishers>
<License>https://creativecommons.org/licenses/by-nc-nd/4.0/</License>
<Abstract>This thesis investigates the mechanisms of factual recall in large language models. We first apply causal interventions to identify neuron activations that are decisive in a model’s factual predictions; surprisingly, we find that factual recall corresponds to a sparse, localizable computation in the MLP weights of the GPT models we study. Harnessing this insight, we then develop methods for efficiently and surgically inserting up to 10,000 new memories into a transformer; these methods perform well in terms of both generalization and specificity. We conclude with some directions for future work.</Abstract>
<Access xmlns="http://purl.org/coar/access_right"
>
</Access>
</Publication>