arXiv:2601. 18350v5 Announce Type: replace-cross Abstract: Large language models fine-tuned via a two-stage pipeline (domain adaptation followed by instruction alignment) can exhibit non-trivial interference after adapter merging, including the re-emergence of explicit reasoning traces under strict decoding.
arXiv:2607. 08393v1 Announce Type: new Abstract: Fine-tuning LLMs to inject new knowledge faces a critical challenge: LLMs can quickly memorize new facts, yet fail to use them for downstream reasoning tasks.
arXiv:2602. 04843v2 Announce Type: replace Abstract: Frontier large language models increasingly solve complex tasks involving abstract concepts through extended test-time thinking.
Instruction tuning is often thought to give language models a universal ability to follow instructions, but this study shows otherwise. By probing nine tasks across three models, the authors find that general probes reveal selective, not uniform, deficits, cross‑task transfer is weak and skill‑similar, and causal ablation uncovers sparse, asymmetric dependencies. The results suggest instruction following is a coordinated use of diverse linguistic skills rather than a single shared mechanism.
arXiv:2606. 05680v1 Announce Type: cross Abstract: Recent advances in large language models (LLMs) have enabled the automatic synthesis (generation) of register-transfer level (RTL) code from natural language instructions, offering a promising pathway to accelerate chip design.
The paper investigates how different prompt adaptation techniques—discrete prompt optimization, prompt tuning, and prefix tuning—affect the internal processing of dense and mixture-of-experts language models. It finds that the adaptation location determines how task information reaches the output: text prompts mainly influence response construction, input-level prompts modify content representations, and layer-wise prefixes exert direct control via attention. In MoE models, these differences also lead to distinct expert‑routing patterns, while generic representation‑level metrics fail to reliably distinguish effective adaptations from placebo effects.
arXiv:2608.23497v1 Announce Type: new Abstract: Reasoning-Induced Misalignment, where fine-tuning on reasoning data containing no harmful content, including mathematics, code, and problem-solving wit...
arXiv:2609.39701v1 Announce Type: new Abstract: Value steering should change an LLM's normative priorities while preserving the scenario, facts, and task constraints underlying its answer. Convention...
Reasoning-Induced Misalignment, where fine-tuning on reasoning data containing no harmful content, including mathematics, code, and problem-solving with chain-of-thought traces can induce harmful beha...
arXiv:2606. 07808v1 Announce Type: new Abstract: Reasoning language models deployed in agentic workflows must follow an instruction hierarchy: when instructions from different sources conflict, the model should obey the highest-privilege applicable instruction.
arXiv:2606. 04057v1 Announce Type: cross Abstract: Large language models (LLMs) now generate substantial production code, often for tasks with multiple valid algorithmic solutions.
arXiv:2605. 15454v2 Announce Type: replace-cross Abstract: Reasoning-trained language models often spend more tokens on harder problems, but longer chains of thought do not show whether a model is merely computing for more steps or following a different internal trajectory.
PathCal is a training‑free controller that refines large reasoning language models by dynamically adjusting the influence of reflection markers—continuation, revision, and alternative‑opening—during inference. By applying finite logit adjustments without extra forward passes, it exploits instantaneous competition among these markers to improve the accuracy–length tradeoff in mathematical reasoning tasks and shorten generation in coding and multilingual math. The method is grounded in an exact log‑odds identity and frozen‑prefix probes that confirm its local directional effect.