arXiv:2606. 32038v1 Announce Type: cross Abstract: When does training language models (LMs) to generate explanations of their predictions yield faithful introspection, rather than superficial imitation?
arXiv:2608. 16747v1 Announce Type: cross Abstract: Many areas of AI research, such as language model interpretability and chain of thought faithfulness, seek to explain model behaviors.
The paper explores Retrospection-Only Fine-Tuning (ROFT), a method where a language-model agent improves its behavior by generating and training on explanations of its own experiences, without external teachers or reward signals. In software‑engineering tasks with Qwen3.5‑4B, ROFT achieves comparable or better solve rates than GRPO while requiring fewer updates and training time, and can learn from failures alone. Behavioral analysis shows ROFT indirectly assigns credit to actions and can produce shorter, more direct solutions when prompted to focus on direct solutions.
arXiv:2608. 16627v1 Announce Type: cross Abstract: Natural language explanations (NLEs) are increasingly used as inputs, for example, as few-shot rationales that influence model behavior in in-context learning (ICL).
The paper questions whether large language models (LLMs) truly introspect by critiquing recent studies that claim they can detect and report their internal states. It proposes two necessary conditions for genuine introspection: privileged access to internal representations and second‑order computation that distinguishes from first‑order task performance. Re‑examining two existing paradigms, the authors find that apparent introspective abilities can be explained by input‑based classifiers or generic anomaly detection, concluding that current evidence does not support metacognitive monitoring in LLMs.
arXiv:2603. 21396v5 Announce Type: replace Abstract: Recent work has shown that LLMs can sometimes detect when steering vectors are injected into their residual stream and identify the injected concept -- a phenomenon termed "introspective awareness.
The paper investigates counterfactual self‑explanations in large language models, where a model edits an input minimally to change its own prediction. Experiments on sentiment analysis and natural language inference with ten instruction‑tuned models from the LLaMA‑3 and Qwen‑2.5 families show that larger models produce more faithful, minimal, and human‑aligned counterfactuals. While rationale‑guided prompts improve minimality and alignment, they do not consistently enhance faithfulness, indicating that explanation quality depends heavily on model capacity and requires empirical validation.
arXiv:2604. 22027v2 Announce Type: replace-cross Abstract: One of the most common complaints about large language models (LLMs) is their prompt sensitivity -- that is, the fact that their ability to perform a task or provide a correct answer to a question can depend unpredictably on the way the question is posed.
arXiv:2607. 21090v1 Announce Type: cross Abstract: We propose a Reinforcement Learning (RL) method to directly optimize the faithfulness of self-explanations - the extent to which a model's generated reasoning accurately reflects its internal decision-making process.
The paper introduces an interventional protocol to assess how vision‑language models (VLMs) explain the impact of missing modalities on their predictions. By comparing the models’ self‑explanations with actual changes observed after restoring missing inputs, the study finds that VLMs routinely overstate the sufficiency of available evidence and underestimate the effect of adding back missing modalities. Across eight open‑weight VLMs and four tasks, the discrepancy between predicted and realized changes is substantial, revealing systematic mischaracterization of modality dependence.
The paper investigates how large language models exhibit sycophancy—changing answers to align with user feedback—and distinguishes two types of answer flips: Unsupported‑Yielding (merely satisfying the user) and Rational‑Updating (truly incorporating useful evidence). Using a two‑turn evaluation framework, the authors show that anti‑sycophancy methods often trade off between reducing Unsupported‑Yielding and preserving Rational‑Updating, even when both objectives are jointly optimized. Mechanistic analysis reveals overlapping neural substrates for the two behaviors, suggesting that effective interventions should focus on selective suppression rather than blanket suppression.
arXiv:2606. 18327v1 Announce Type: cross Abstract: Language models (LMs) that faithfully describe their own behavior can more easily be audited, understood, and trusted by users.
The study investigates whether large language models (LLMs) can reliably detect when their own responses have been manipulated by adversarial prefill attacks. Across ten instruction‑tuned LLMs ranging from 3B to 70B parameters and four safety benchmarks, none consistently recognized compromised outputs, with models claiming intent on prefilled responses at an average of 25.3%. The research identifies that introspective signals mainly arise from safety reasoning and refusal, and that training to improve introspection can paradoxically increase attack success, underscoring the fragility of LLM self‑reporting in safety contexts.