The paper investigates clustered optimal‑transport (CHaRS) steering for language models, comparing it to the standard DiM approach. It shows that CHaRS, which routes hidden states to cluster‑based displacements, can reduce toxicity by 37‑61% more than DiM on certain models while maintaining comparable fluency. The study also demonstrates that these benefits persist under INT8 and NF4 quantization, and that a simpler constant map still offers significant toxicity reduction, leaving the exact token‑routing strategy open for further exploration.
arXiv:2607. 01239v1 Announce Type: cross Abstract: Character-level perturbations bypass safety alignment in modern LLMs despite leaving prompts human-readable.
arXiv:2608.29702v1 Announce Type: new Abstract: A token-embedding table holds a hub of short rows near its origin, and we show that this cluster biases what nearest-neighbor intrinsic-dimension (ID)...
The paper introduces AgentDiff, a metric that quantifies how much LLM agents’ answers differ when inputs are altered by meaning‑bearing rewrites (paraphrases, synonym substitutions) versus presentation changes (reordering, formatting, distractors). Across 68 model–benchmark–scaffold combinations involving ten LLMs and over 1,500 questions, meaning‑bearing rewrites consistently produce a roughly 20‑percentage‑point higher inconsistency rate than presentation changes, a gap that persists across severity proxies and remains significant even outside the Qwen family. Trace analysis reveals that meaning‑bearing rewrites preserve the first action but reduce thought similarity from the second step onward, extending the divergence cascade—a phenomenon termed “stealth divergence.”
arXiv:2609.07901v1 Announce Type: new Abstract: Weight quantization largely determines the economics of serving open-weight LLMs. Its costs are usually assessed with capability benchmarks, on which 4...
arXiv:2609.06473v1 Announce Type: new Abstract: Inference-time activation steering enables behavioral control of large language models without parameter modification, while post-training quantization...
arXiv:2610. 00445v1 Announce Type: new Abstract: Pairwise guide--transcript scores do not enforce conservation of a finite guide-loaded RISC pool when they are interpreted independently as occupancies.
The paper investigates when post‑training quantization (PTQ) calibration helps or hurts text embedders, focusing on the amount of calibration (dose) and whether to calibrate at all. Across three models and 161 calibration runs at 4‑bit weight quantization, the effect ranges from significant gains to catastrophic degradation, driven mainly by the functional displacement of the quantized model. A validation‑gated abstention rule is proposed that avoids harmful calibrations while capturing most benefits, highlighting the trade‑off between safety and opportunity.
The paper investigates how the four‑stream manifold‑constrained hyper‑connection (mHC) residual pathway in DeepSeek‑V4‑Flash is actually used. It finds that read/write routing is concentrated, typically involving only two streams per block, and that the dominant stream shifts across layers while representations stay directionally distinct. Residual mixing is modest, mainly in early layers, and late mixing contributes little to performance, whereas early mixing is crucial for perplexity and task scores.
arXiv:2607. 04926v1 Announce Type: cross Abstract: How does the way information reaches a transformer -- as symbolic tokens, a clean per-factor "oracle" code, or an entangled perceptual vector -- shape whether it binds that information compositionally?
arXiv:2609.01556v1 Announce Type: cross Abstract: We evaluate embedding retrieval where surface form and meaning are pulled apart on purpose: retrieving items that share underlying structure but not...
arXiv:2607. 09487v1 Announce Type: new Abstract: Within-class variance in language-model representations is commonly read as incomplete neural collapse.
Activation steering suppresses undesired behaviors in language models by adding a steering vector to the hidden state during generation. Recent conditional methods such as CAST and DSAS improve the be...