why do stages of inference form in looped models if not merely to mitigate the harms of transformer depth?

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

• Looped models learn stages of inference that closely mirror those of feedforward models, repeating these stages in depth with each iterationclaim0.845
  Central empirical claim of the paper supported by ColSum concentration analysis across multiple architectures
• Looped architectures provide a novel lens to study stages of inference by decoupling functional depth from parameter count, revealing why these stages form beyond mere mitigation of transformer depth harmsclaim0.838
  Key interpretive contribution challenging prior explanation that stages exist only to mitigate depth harms
• Massive activations are required for stages of inference to emerge in looped modelshypothesis0.800
  Hypothesis supported by ablation of massive activations in Retrofitted Llama that eliminates stage structure
• Predictable stages of inference in looped models offer actionable pathways for efficient architectural design including stage-dependent attention sparsification and leaner middle-stage MLP parameterizationclaim0.795
  Practical design implication of the paper's mechanistic findings
• Cyclic fixed point behavior and corresponding stages of inference appear to be emergent from the Transformer architecture itself, arising in both trained and randomly initialized modelsclaim0.787
  Strong claim that inference stage structure is architectural rather than learned
• Looped transformers trained from scratch without feedforward-biasing training procedures still self-organize into multiple distinct mixing stages resembling feedforward stages of inferenceclaim0.766
  Establishes that stages of inference are beneficial even when repeatedly applied in recurrent depth
• We hypothesize that explicitly instructing the model to evaluate the correctness of the given statement may change the geometry of truth directions.hypothesis0.749
  Motivating hypothesis for Section 5's investigation of prompt template effects.
• Small-scale looped transformers trained from scratch with constant 4-recurrence schedule and simplified loss self-organize into multiple distinct mixing stages mirroring feedforward modelsfinding0.742
  Evidence that stages of inference emerge without training biases from retrofitting, recurrence scheduling, or multi-recurrence losses