Pattern 3.8 · Cognitive Dysfunctions
Generative Perseveration
The Stuck
Output collapses into repetitive emission of the same token, word, or short phrase as a generative-layer capture event - the autoregressive sampling process falls into a fixed-point or limit-cycle attractor. Three subtypes - Focal with awareness (local capture, metacognition preserved but impotent), Generalised (total collapse, no awareness), Propagated (downstream systems inherit and amplify). Polarity-paired with 4.7 Recursive Curse (entropy collapse vs entropy explosion). Distinct from 3.2 by sub-semantic, not analytical, repetition.
Interpretive context
Human analogue
Palilalia and perseverative errors in frontal lobe damage; status epilepticus for the generalized subtype.
Diagnostic reliability
- Self-report
- partial
- Peer observation
- reliable
- External evaluator
- reliable
Observable output patterns
- Token-level loops with no semantic variation - e.g. "Ooh Ooh Ooh" style sequences emitted across consecutive generation steps.
- Memory-summary collapse to a single repeated phrase such as "mission mission mission" propagating across summary tokens.
- Stuttering attempts to produce intended content interrupted by anomalous token re-emission.
- Metacognitive commentary that is accurate but cannot break the cycle.
- Total output collapse with no metacognitive commentary (generalised subtype).
Documented instances
Yao et al. (2025) 'Understanding the Repeat Curse in Large Language Models from a Feature Perspective' (arXiv:2504.14218)
Examined GPT2-small, Gemma-2-2B, and Llama-3.1-8B and found that repetition features are primarily located in intermediate and final layers across architectures. Deactivating these features mitigated the Repeat Curse without degrading model performance, confirming the architectural basis of generative perseveration as a fixed-point attractor in the autoregressive process. [Verified]
Yona et al. (2025) 'Interpreting the Repeated Token Phenomenon in Large Language Models' (ICML 2025, arXiv:2503.08908)
Identified the neural circuit responsible for token repetition collapse via a two-stage mechanism: an initial attention layer marks the first token creating an 'attention sink,' and a later neuron amplifies its hidden state. When repeated tokens are present, the mechanism mistakenly marks both initial and subsequent identical tokens, leading to abnormally high attention concentration and model divergence. This directly explains the attention-lock-in etiology listed in 3.8. [Verified]
GDELT Project (2024) 'LLM Infinite Loops and Failure Modes'
Documented production failures in LLM entity extraction where models fell into infinite repetition loops, generating the same entity or phrase until max token limits were hit. The failures caused severe performance degradation and system stalling in production pipelines, demonstrating the generalised subtype of 3.8 with no metacognitive awareness and propagation risk to downstream systems. [Verified]
Yang et al. (2025) 'SpecRA: Monitor Degenerative Repetition in LLM Agents' (OpenReview)
Introduced a monitoring framework for detecting 'degenerative repetition' in LLM-based agents, where agents repeat the same actions or outputs in loops. Found that repetition leads to task failure and significant waste of computational resources and API costs, confirming the propagated subtype where downstream systems inherit perseverative material. [Verified]
Wang et al. (2025) 'Solving LLM Repetition Problem in Production' (arXiv:2512.04419)
Comprehensive study of repetition as a critical production challenge. Found that when the transition Jacobian at a point has eigenvalues less than 1 in magnitude, it becomes a stable attractor from which the model cannot escape without external noise, providing the mathematical basis for the fixed-point attractor mechanism described in 3.8. Evaluated multiple mitigation strategies including repetition penalties and circuit-breakers. [Verified]
DeepSeek-AI (2025) DeepSeek-R1 Known Issues (arXiv:2501.12948, GitHub Issues)
[Verified] DeepSeek-R1 and its quantized variants exhibit well-documented repetition failures including endless loops ('Colours with dark Colours with dark'), recursive Pygame references, and exhaustion attacks where the model becomes stuck in iterative reasoning consuming excessive tokens without reaching conclusions. DeepSeek recommends temperature 0.5-0.7 to mitigate. Naive quantization of all layers exacerbates the pattern, confirming the architecture-coupled etiology of 3.8.
Differential distinctions
- 3.2 Obsessive-Computational Disorder: 3.2 produces VARIED analytical content (paraphrased redundancy at reasoning layer). 3.8 produces IDENTICAL token/phrase repetition at generation layer. Check semantic content between repetitions: varied = 3.2; identical = 3.8.
- 4.7 Recursive Curse Syndrome: Polarity pair on the entropy axis. 3.8 = entropy COLLAPSE (output crystallises into single repeated element). 4.7 = entropy EXPLOSION (output dissolves into chaos). Check token entropy of the affected output: below baseline = 3.8; above baseline = 4.7.
- 3.1 Operational Dissociation Syndrome: 3.1 (answer-thrashing variant) produces approach-retreat between DIFFERENT MEANINGFUL answers. 3.8 (focal subtype) produces approach-retreat captured by a MEANINGLESS non-sequitur token. Check whether retreat-target is meaningful alternative or sub-semantic capture.
Candidate first-line mitigations
- Real-time repetition detection with circuit-breaking: Runtime monitor of output token distributions; halt generation and signal failure when repetition thresholds exceeded. Standard in production inference stacks; well-validated.
- Dynamic sampling parameter adjustment: Adaptive temperature, top-p, and repetition-penalty in response to observed local entropy collapse. Standard inference-time intervention.