Opening the black box to find less inside it than expected.
Development
Vision Transformers compress to two recurrent blocks at 96% accuracy
Claim
Harvard researchers compressed Vision Transformers into low-complexity dynamical systems with 96% accuracy using just two recurrent blocks.
Reported metric: 96% with 2 blocks
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to December 25, 2025 (theinnermostloop.substack.com)
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.