Ouro 1.4B layers continuously change throughout 128 recurrences, exhibiting unstable stages of inference when extrapolating beyond training recurrences

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Ouro 2.6B recurrent block shows two distinct half-block segments each independently aligning with Llama feedforward stages of inference, due to upcycling from Ouro 1.4Bfinding0.849
  Reveals how the upcycling training regime of Zhu et al. produces duplicated inference stage structure
• Ouro 1.4B does not converge to a fixed point even after 128 recurrences, despite showing small successive differencesfinding0.841
  Key negative result showing that not all looped models reach a true fixed point, contrasting with retrofitted models
• Huginn-0125 performance remains constant when extrapolating beyond training recurrences, while Ouro performance deteriorates in the same regimefinding0.803
  Correlates stable fixed-point behavior with out-of-domain generalization performance at test-time
• Ouro 1.4B, Retrofitted Llama, and Huginn-0125 exhibit diagonal patterns in Frobenius norm heatmaps confirming cyclic fixed point behavior across 8 recurrencesfinding0.777
  Empirical validation that attention patterns are most similar to same-layer outputs across different recurrences
• In Opus 4.1, representation of the think word decays to baseline by the final layer, unlike Claude 3 models where it persistsfinding0.761
  Suggests that later models can keep the thought 'silent' rather than letting it influence output.
• F0-trained probes in layers 4-10 show inverted separation on F1 (AUROC ≈ 0), systematically misclassifying true statements as false.finding0.752
  Demonstrates that early-layer probes capture sentence polarity rather than truth.
• The model converges to a more stable truth direction in middle-to-late layers, as evidenced by increasing cosine similarity between layer-wise probes.claim0.751
  Supported by the geometric transition visible in cosine similarity heatmaps for F0-F3.
• In Opus 4.1, the think word representation decays to baseline in the final layer because the strong next-token prediction drowns out other representationshypothesis0.749
  Explanation for the 'silent' thought phenomenon.