What is the fundamental cause of mode collapse in post-training aligned LLMs?

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

• Typicality bias in preference data is a fundamental and pervasive data-level cause of mode collapse in aligned LLMsclaim0.812
  The central thesis of the paper, distinguishing it from algorithmic explanations of mode collapse
• Post-training stages (SFT, DPO, RLVR) progressively reduce output diversity, with direct prompting showing the most severe mode collapseclaim0.765
  Empirical finding from Tulu-70B ablation study across post-training stages
• LLM personality self-reports are illusory: post-training alignment creates stable human-like reports dissociated from actual behavior (Han et al. 2025)claim0.765
  Skeptical prior work motivating the need to validate self-reports against internal states rather than taking them at face value
• Can mode collapse be mitigated at inference time without retraining the model?question0.744
  Motivates the VS approach as a training-free solution
• Behavioural traces surface recurring LLM failure modes including overbidding, self-bidding, bankrupt TC initiation, and weak opponent-state adaptation that never appear in code agents.claim0.741
  LLMs exhibit systematic errors that deterministic logic avoids.
• The systematic behavioral shift of LLMs under self-referential processing conditions predicted by consciousness theories represents something more structured than superficial correlations in training dataclaim0.735
  The paper's claim that theoretical convergence across GWT, RPT, HOT, IIT makes the findings non-coincidental
• During pre-training, models learn a variety of personas including misaligned ones; fine-tuning on narrowly incorrect datasets amplifies misaligned personas because they reduce training loss, causing broadly misaligned behaviorclaim0.733
  The paper's central mechanistic explanation of why narrow fine-tuning causes broad misalignment
• Two heuristic code agents outperform most tested LLMs, and behavioural traces surface recurring LLM failure modes including overbidding, self-bidding, bankrupt TC initiation, and weak opponent-state adaptation.quote0.732
  Abstract sentence summarising performance and failures.