arXiv:2510.04891v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are increasingly deployed in contexts where their failures can have direct sociopolitical consequences. Yet, existing safety benchmarks rarely test vulnerabilities in domains such as political manipulation, propaganda and disinformation generation, or surveillance and information control. We introduce SocialHarmBench, a dataset of 585 prompts spanning 7 sociopolitical categories and 34 countries, designed to surface where LLMs most acutely fail in politically charged contexts. Our evaluations reveal several shortcomings: open-weight models exhibit high vulnerability to harmful compliance, with Mistral-7B reaching attack success rates as high as 97% to 98% in domains such as historical revisionism, propaganda, and political manipulation. Moreover, temporal and geographic analyses show that LLMs are most fragile when confronted with 21st-century or pre-20th-century contexts, and when responding to prompts tied to regions such as Latin America, the USA, and the UK. These findings demonstrate that current safeguards fail to generalize to high-stakes sociopolitical settings, exposing systematic biases and raising concerns about the reliability of LLMs in preserving human rights and democratic values. We share the SocialHarmBench benchmark at https://huggingface.co/datasets/psyonp/SocialHarmBench.

arXiv:2512.19027v2 Announce Type: replace Abstract: Developers often struggle to specify correct training labels and rewards. Perhaps they don't need to. We propose recontextualization, which reduces how often language models "game" training signals, performing misbehaviors those signals mistakenly reinforce. We show recontextualization prevents models from learning to 1) prioritize evaluation metrics over chat response quality; 2) special-case code to pass incorrect tests; 3) overwrite evaluation functions rather than write correct code; and 4) become sycophantic. Our method works by generating completions from prompts discouraging misbehavior and then recontextualizing them as though they were in response to prompts permitting misbehavior. Recontextualization trains language models to resist misbehavior even when instructions permit it. This mitigates the reinforcement of misbehavior from misspecified training signals, reducing specification gaming without improving the supervision signal.

arXiv:2508.18210v2 Announce Type: replace-cross Abstract: Synthetic data is increasingly critical for contact centers, where privacy constraints and data scarcity limit the availability of real conversations. However, generating synthetic dialogues that are realistic and useful for downstream applications remains challenging. In this work, we benchmark multiple generation strategies guided by structured supervision on call attributes (Intent Summaries, Topic Flows, and Quality Assurance (QA) Forms) across multiple languages. To test downstream utility, we evaluate synthetic transcripts on an automated quality assurance (AutoQA) task, finding that prompts optimized on real transcripts consistently outperform those optimized on synthetic transcripts. These results suggest that current synthetic transcripts fall short in capturing the full realism of real agent-customer interactions. To highlight these downstream gaps, we introduce a diagnostic evaluation framework comprising 17 metrics across four dimensions: (1) Emotional and Sentiment Arcs, (2) Linguistic Complexity, (3) Interaction Style, and (4) Conversational Properties. Our analysis shows that even with structured supervision, current generation strategies exhibit measurable deficiencies in sentiment fidelity, disfluency modeling, behavioral variation, and conversational realism. Together, these results highlight the importance of diagnostic, metric-driven evaluation for synthetic conversation generation intended for downstream applications.

arXiv:2510.04008v3 Announce Type: replace-cross Abstract: Softmax Attention has a quadratic time complexity in sequence length, which becomes prohibitive to run at long contexts, even with highly optimized GPU kernels. For example, FlashAttention-2/3 (exact, GPU-optimized implementations of Softmax Attention) cannot complete a single forward-backward pass of a single attention layer once the context exceeds ~4 million tokens on an NVIDIA GH200 (96 GB). We introduce Repeated Arrays-of-Count Estimators (RACE) Attention, a kernel-inspired alternative to Softmax Attention that is strictly linear in sequence length and embedding size. RACE Attention replaces the exponential kernel with a sharpened angular similarity, and approximates attention outputs via Gaussian random projections and soft Locality-Sensitive Hashing (LSH), avoiding construction of the full attention matrix. Across language modeling, masked language modeling, and text/image classification, RACE Attention matches or outperforms strong baselines up to 64K seqeuence length while reducing wall-clock time and memory usage. In addition, we conduct a controlled scaling study on a single attention layer and demonstrate processing of up to 12 million tokens on an NVIDIA GH200 GPU and 75 million tokens on an Intel Xeon Gold 5220R CPU in a single forward-backward pass, which is well beyond the capabilities of current state-of-the-art attention implementations. RACE Attention thus offers a practical and theoretically grounded mechanism for long-context training on today's hardware. We release our code at https://github.com/sahiljoshi515/RACE_Attention.