arXiv:2606. 07631v1 Announce Type: cross Abstract: Emergent misalignment (EM) occurs when narrow finetuning causes a model to behave dangerously outside the finetuning task.
arXiv:2606. 31591v1 Announce Type: cross Abstract: Emergent misalignment (EM) is a recently discovered phenomenon in LLMs where fine-tuning on a narrow misaligned task, such as writing insecure code, leads to broadly misaligned behaviour on unrelated prompts.
arXiv:2608. 14929v1 Announce Type: cross Abstract: Open-weight language models are fine-tuned, quantized, pruned, and merged, yet their provenance is often undocumented.
The paper introduces a leakage‑free 3×3 spatio‑temporal partition for evaluating inductive kriging, ensuring training, validation, and testing occur on distinct spatial and temporal domains. It proposes DRIK, a framework that includes Spatial Continuity Regularization, Masked Flow Disambiguation, and Structural Domain Expansion to mitigate structural shifts from unseen nodes. Experiments on six datasets show DRIK outperforms existing baselines, reducing MAE by up to 12.48% and achieving lower test‑to‑validation MAE ratios under the stricter evaluation protocol.
arXiv:2606. 20502v1 Announce Type: cross Abstract: Whether LLMs scoring well on vulnerability benchmarks genuinely reason about security or merely pattern-match on contaminated data remains unresolved.
arXiv:2606. 28525v1 Announce Type: cross Abstract: Fine-tuning on harmless data can partially undo behaviors acquired earlier in training.
This paper presents a reproducibility audit of frozen‑encoder anomaly detection experiments originally reported on arXiv. The authors confirm that the numerical discrimination results can be reproduced from the preserved artifacts, but they find that the claimed causal link to interferometric pretraining is unsupported. They show that near‑zero embeddings and architectural choices, rather than a morphological prior from gravitational‑wave instrumentation, explain the observed anomaly‑detection performance.
arXiv:2608. 04347v1 Announce Type: new Abstract: Fine-tuning enables a source model to acquire desired capabilities and behaviors in a target domain while retaining much of its general-purpose competence.
arXiv:2607. 20436v1 Announce Type: cross Abstract: Safety evaluations often assume that behavior observed during testing reflects behavior in ordinary use, but fine-tuning can break this assumption.
arXiv:2609.36569v1 Announce Type: cross Abstract: Checkpoint selection is a routine decision in supervised fine-tuning (SFT): training produces multiple checkpoints, but only one is retained. Yet fix...
Fine-tuning enables a source model to acquire desired capabilities and behaviors in a target domain while retaining much of its general-purpose competence. However, this adaptation process can also degrade alignment properties that were present in the source model.
arXiv:2608. 11034v1 Announce Type: cross Abstract: In LLM pre-training, synchronization propagates rank-local stalls, slowdowns, and numerical errors into job-wide symptoms, obscuring their origin.
arXiv:2606. 15980v1 Announce Type: cross Abstract: Activation monitors-lightweight probes trained on a language model's internal representations-are an increasingly common layer in deployment safety stacks.