Is Convergence Inevitable? Tracing Output Homogeneity Back to Base Models

• URL: http://arxiv.org/abs/2608.11426v1
• Date: Tue, 11 Aug 2026 20:47:06 GMT
• Title: Is Convergence Inevitable? Tracing Output Homogeneity Back to Base Models
• Abstract summary: We find that semantic convergence is observed from the first alignment stage.<n>We find that convergence can be revealed and amplified, but not introduced by the SFT data.<n> Taken together, our results suggest that semantic convergence may arise naturally from the objectives underlying LM training.
• Score: 49.46165710676499
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The lack of diversity in LM content is widely attributed to the alignment process, but how and where exactly in the pipeline this collapse begins is unknown. We argue that output homogeneity is likely learned during the pretraining phase, and only \emph{revealed} or magnified during the alignment process. Specifically, we find that semantic convergence is observed from the first alignment stage--the instruction-tuning phase (SFT)--suggesting that homogeneity might already exist in the pre-alignment model. To investigate this, we conduct controlled SFT experiments examining how training data influences output convergence on specific input/output pairs. We find that convergence can be revealed and amplified, but not introduced by the SFT data, supporting its role as a catalyst rather than a cause. To further test whether homogeneity originates before alignment, we measure convergence in base models. We find that instruct-like collapse can be induced through prompting alone, even without alignment. Taken together, our results suggest that semantic convergence may arise naturally from the objectives underlying LM training, making it difficult to mitigate through post-alignment interventions alone.

Related papers

• Scale-Consistent Posterior Dynamics for Diffusion Inverse Problems [16.685352002333225]
  Posterior sampling with a pretrained diffusion prior is governed by a conditional score whose likelihood component is generally intractable.<n>We show that a noise-conditioned path whose targets approach the clean posterior produces a continuous surrogate SDE.<n>Experiments on FFHQ and ImageNet with 100 score evaluations demonstrate competitive reconstruction fidelity for superresolution and deblurring.
  arXiv  Detail & Related papers  (2026-08-15T09:38:41Z)
• How Does Alignment Tuning Shape Representations of Sycophancy and Related Cue-Induced Biases in LLMs? [53.58941416780391]
  We study where this susceptibility, spanning sycophancy and related cue-induced biases, lives inside the model.<n>We extract a per-bias direction from hidden states and triangulate it through three measures: probing, leave-one-dataset-out transfer, and causal intervention.
  arXiv  Detail & Related papers  (2026-07-20T16:10:06Z)
• Towards Physical Intuitions for Alignment Dynamics: A Case Study With Randomness Crystallization [17.107790125434516]
  We argue that the physical sciences offer a principled and underexplored vocabulary for reasoning about dynamics.<n>We instantiate this position through the lens of material Crystallization, which is a well-studied thermodynamic phase transition.<n>We propose intuitive metrics to verify the transitions between these phases, and validate the idea across a range of random tasks.
  arXiv  Detail & Related papers  (2026-06-29T08:09:55Z)
• Why Zeroth-Order Adaptation May Forget Less: A Randomized Shaping Theory [23.912005398880293]
  Continual learning requires new-task adaptation without damaging previously acquired capabilities.<n>Recent forward-pass and zeroth-order (ZO) results show that low-query adaptation may retain better than first-order descent.<n>We give a local randomized gradient-shaping analysis: finite differences expose a raw shape that is mean-aligned with FO.<n>For norm-matched ZO, the expected shaped retention curvature obeys an exact identity that preserves the isotropic retention floor while contracting only the anisotropic component.<n>Projecting this identity onto the incoming gradient yields the observable FO--ZO quadratic forgetting gap.
  arXiv  Detail & Related papers  (2026-05-11T14:41:08Z)
• Conditional Diffusion Under Linear Constraints: Langevin Mixing and Information-Theoretic Guarantees [10.73598638822051]
  We study zero-shot conditional sampling with pretrained diffusion models for linear inverse problems.<n>We prove that the error from replacing this score by the unconditional tangent score is upper bounded by a dimension-free conditional mutual information between observed and unobserved components.
  arXiv  Detail & Related papers  (2026-05-06T19:19:54Z)
• OrthoFormer: Instrumental Variable Estimation in Transformer Hidden States via Neural Control Functions [0.0]
  Transformer architectures excel at sequential modeling yet remain fundamentally limited by correlational learning.<n>We propose OrthoFormer, a causally grounded architecture that embeds instrumental variable estimation directly into Transformer blocks via neural control functions.
  arXiv  Detail & Related papers  (2026-03-08T03:05:16Z)
• Unregularized Linear Convergence in Zero-Sum Game from Preference Feedback [50.89125374999765]
  We provide the first convergence guarantee for Optimistic Multiplicative Weights Update ($mathtOMWU$) in NLHF.<n>Our analysis identifies a novel marginal convergence behavior, where the probability of rarely played actions grows exponentially from exponentially small values.
  arXiv  Detail & Related papers  (2025-12-31T12:08:29Z)
• Departures: Distributional Transport for Single-Cell Perturbation Prediction with Neural Schrödinger Bridges [51.83259180910313]
  A major bottleneck in gene function analysis is the unpaired nature of single-cell data.<n>We approximate Schrdinger Bridge (SB) to tackle unpaired single-cell perturbation data.<n>Our model effectively captures heterogeneous single-cell responses and achieves state-of-the-art performance.
  arXiv  Detail & Related papers  (2025-11-17T08:27:13Z)
• Language Models Resist Alignment: Evidence From Data Compression [30.708635183315433]
  Large language models (LLMs) may exhibit unintended or undesirable behaviors.<n>We show that fine-tuning disproportionately undermines alignment relative to pre-training, potentially by orders of magnitude.<n>Our findings underscore the need to address the inherent elasticity of LLMs to mitigate their resistance to alignment.
  arXiv  Detail & Related papers  (2024-06-10T10:03:16Z)
• Score-based Causal Representation Learning with Interventions [54.735484409244386]
  This paper studies the causal representation learning problem when latent causal variables are observed indirectly. The objectives are: (i) recovering the unknown linear transformation (up to scaling) and (ii) determining the directed acyclic graph (DAG) underlying the latent variables.
  arXiv  Detail & Related papers  (2023-01-19T18:39:48Z)
• Orthogonal Jacobian Regularization for Unsupervised Disentanglement in Image Generation [64.92152574895111]
  We propose a simple Orthogonal Jacobian Regularization (OroJaR) to encourage deep generative model to learn disentangled representations. Our method is effective in disentangled and controllable image generation, and performs favorably against the state-of-the-art methods.
  arXiv  Detail & Related papers  (2021-08-17T15:01:46Z)
• Disentangling Observed Causal Effects from Latent Confounders using Method of Moments [67.27068846108047]
  We provide guarantees on identifiability and learnability under mild assumptions. We develop efficient algorithms based on coupled tensor decomposition with linear constraints to obtain scalable and guaranteed solutions.
  arXiv  Detail & Related papers  (2021-01-17T07:48:45Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.