Compression Favors Consistency, Not Truth: When and Why Language Models Prefer Correct Information
Paper • 2603.11749 • Published • 3
Konstantin Krestnikov
krestnikov
·
AI & ML interests
None yet
Recent Activity
submitted a paper 5 days ago
Compression Favors Consistency, Not Truth: When and Why Language Models Prefer Correct Information updated a dataset 5 days ago
krestnikov/compression-drives-truth published a dataset 5 days ago
krestnikov/compression-drives-truthOrganizations
submitted a
paper to Daily Papers 5 days ago