Alignment, interpretability, red-teaming, bias and privacy: the research on what these systems do when they misbehave.
arXiv:2610.27248v1 Announce Type: cross Abstract: Generative AI can create programming feedback at scale, but validating its alignment with instructors' pedagogical goals and strategies remains chall...
arXiv:2610.27266v1 Announce Type: cross Abstract: Preference labels align model training, but the judgments behind them are difficult to inspect or verify. In this paper, we examine how preference co...
The paper introduces Guided-AP, a method that improves attribution patching by using a few exact measurements to identify and correct its errors. On a single prompt, Guided-AP reduces error by up to 26% across three behaviors, while on many prompts it cuts average error on factual completion by 75% to 79%. The approach leverages shared error signs across components and tasks to efficiently guide verification and correction.
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.
arXiv:2610.27603v1 Announce Type: new Abstract: Empathetic interaction involves two distinguishable capacities: representing another person's emotional state (cognitive empathy) and generating an app...
arXiv:2610.27626v1 Announce Type: new Abstract: Multimodal Large Language Models show strong vision-language capabilities, yet their geometry-centric reasoning remains insufficiently understood. Exis...
arXiv:2610.27817v1 Announce Type: new Abstract: Every prompt sent to a hosted language model is readable by whoever runs it. We study whether the prompt can remain on the client while the host receiv...
arXiv:2610.28121v1 Announce Type: new Abstract: Safe autonomous navigation of UASs requires tactical deconfliction to address conflicts that strategic coordination cannot prevent. This task is challe...
arXiv:2610.28361v1 Announce Type: new Abstract: Long-horizon deep search requires LLM agents to iteratively decompose complex information needs, gather evidence from external sources, and maintain a...
arXiv:2610.28484v1 Announce Type: new Abstract: AI agents increasingly act on behalf of people or other agents, purchasing services, exchanging information and delegating work. What an agent must res...
The paper examines whether AI‑driven debate can help humans and weaker language models (LLMs) correct their judgments. Experiments show that debate improves human accuracy over prior beliefs, especially with weighted voting, but offers little benefit—and sometimes harms—accuracy for weaker LLM supervisors. The study highlights debate as a promising tool for scalable human oversight while questioning its effectiveness for fully automated AI control.
arXiv:2610.28604v1 Announce Type: new Abstract: Current evaluation of LLM-generated debate largely focuses on argumentative quality, leaving the value structure of generated arguments underexamined....
arXiv:2610.26914v1 Announce Type: cross Abstract: Pluralistic alignment requires AI systems to learn from diverse populations whose preferences may reflect heterogeneous or even conflicting objective...
arXiv:2610.27135v1 Announce Type: cross Abstract: With the advancement of powerful AI chips, on-device machine learning is gaining traction for its low latency and privacy protection. However, when e...
arXiv:2610.27213v1 Announce Type: cross Abstract: Personalized large language models (LLMs) increasingly rely on profile-conditioned inference, pairing public requests with rich, user-specific facts....
The paper investigates whether safe AI systems can simultaneously satisfy multiple desiderata, such as harmlessness and helpfulness. By modeling the interaction between an AI, a defender, and an attacker as a three‑player game, the authors prove a fundamental tradeoff: advice that benefits the defender must remain public, limiting its safety guarantees. They also explore whether the AI can manipulate information to favor only the defender and discuss implications for safer AI design.
arXiv:2610.27470v1 Announce Type: cross Abstract: Federated learning (FL) is a distributed machine learning paradigm that reduces transmission costs, mitigates cross-border data sharing restrictions,...
arXiv:2610.27516v1 Announce Type: cross Abstract: Despite advances in photorealism, semantic fidelity, and aesthetics, text-to-image generation and image editing remain understudied in a human-center...
arXiv:2610.27590v1 Announce Type: cross Abstract: When unique pretraining data is scarce, models see the same tokens many times and are thought to memorize. Random-token replacement, an input-corrupt...
arXiv:2610.27611v1 Announce Type: cross Abstract: Ensuring the alignment of Large Language Models (LLMs) with diverse cultural values is critical for their safe global deployment. However, existing a...