The Memorization Paradox


If TTT truly functions by explicitly learning and retrieving key-value associations, its behavior should conform to basic principles of memory formation and optimization dynamics. Instead, we identify systematic anomalies that directly contradict this hypothesis:

Better Inner Loss → Worse Performance

Increasing inner-loop iterations improves inner-loop loss but degrades task performance. This inverse relationship directly contradicts the memorization view—if TTT were memorizing key-value pairs, better fitting should improve downstream results.

Inner-Loop Optimization vs Performance

More inner-loop iterations improve inner-loop loss but degrade task performance.

Gradient Ascent Works!

Replacing gradient descent with gradient ascent preserves—and sometimes even improves—task performance, despite explicitly worsening the key-value fitting objective. This is impossible under a memorization interpretation.

Model Variant Perplexity ↓ (LLM) PSNR ↑ (NVS) Top-1 Acc ↑ (Image)
Baseline 16.43 25.94 79.34%
Gradient Ascent 16.19 25.85 79.61%
Q and K Are Misaligned

Queries and keys exhibit significant distributional mismatch in trained models, meaning the inner-loop function is evaluated out-of-distribution. For retrieval to work, Q and K should be in the same semantic space—but they're not.

Distributional Asymmetry Between Q and K

t-SNE visualization showing Q/K distributional mismatch—the inner loop is evaluated out-of-distribution.

Replacing Q with K Has No Effect

Substituting queries with keys has negligible impact on performance, suggesting queries don't play a functional retrieval role as in standard attention.

Model Variant Perplexity ↓ (LLM) PSNR ↑ (NVS) Top-1 Acc ↑ (Image)
Baseline 16.43 25.94 79.34%
Replace Q with K 16.18 25.95 79.18%