相关论文: Reading Task Failure Off the Activations: A Sparse…

Backdoor attacks on language models pose a significant threat to AI safety, where models behave normally on most inputs but exhibit harmful behavior when triggered by specific patterns. Detecting such backdoors through mechanistic…

The computational burden of attention in long-context language models has motivated two largely independent lines of work: sparse attention mechanisms that reduce complexity by attending to selected tokens, and gated attention variants that…

Hosted-LLM providers have a silent-substitution incentive: advertise a stronger model while serving cheaper replies. Probe-after-return schemes such as SVIP leave a parallel-serve side-channel, since a dishonest provider can route the…

Sparse autoencoders (SAEs) are a useful tool for uncovering human-interpretable features in the activations of large language models (LLMs). While some expect SAEs to find the true underlying features used by a model, our research shows…

The rapid integration of Large Language Models (LLMs) into educational assessment rests on the unverified assumption that instruction following capability translates directly to objective adjudication. We demonstrate that this assumption is…

We probe OpenAI's open-weights 20-billion-parameter model gpt-oss-20b to study how sociopragmatic framing, language choice, and instruction hierarchy affect refusal behavior. Across 80 seeded iterations per scenario, we test several harm…

Pre-trained large language models can efficiently interpolate human-written prompts in a natural way. Multitask prompted learning can help generalization through a diverse set of tasks at once, thus enhancing the potential for more…

The Cognitive Categorical Transformer (CCT) is a 306M-parameter architecture that augments a pretrained GPT-2 Small backbone with cognitively grounded components derived from category theory and several inspirations from cognitive science.…

Recent work in mechanistic interpretability has shown that behaviors in language models can be successfully reverse-engineered through circuit analysis. A common criticism, however, is that each circuit is task-specific, and thus such…

Security operations centers face persistent alert fatigue: in low-prevalence streams, even low false-positive rates generate substantial investigation load, while aggregate F1 scores obscure analyst burden. We introduce PACT, a Pareto-aware…

Long-sequence processing is a critical capability for modern large language models. However, the self-attention mechanism in the standard Transformer architecture faces severe computational and memory bottlenecks when processing long…

Sparse Autoencoders (SAEs) have demonstrated significant success in interpreting Large Language Models (LLMs) by decomposing dense representations into sparse, semantic components. However, their potential for analyzing Vision Transformers…

Segment Anything Model 2 (SAM2) shows excellent performance in video object segmentation tasks; however, the heavy computational burden hinders its application in real-time video processing. Although there have been efforts to improve the…

Sparse Autoencoders (SAEs) extract features from LLM internal activations, meant to correspond to interpretable concepts. A core SAE training hyperparameter is L0: how many SAE features should fire per token on average. Existing work…

Large Language Models (LLMs) frequently exhibit strong translation abilities, even without task-specific fine-tuning. However, the internal mechanisms governing this innate capability remain largely opaque. To demystify this process, we…

Understanding what sparse auto-encoder (SAE) features in vision transformers truly represent is usually done by inspecting the patches where a feature's activation is highest. However, self-attention mixes information across the entire…

Intermediate layers of large language models (LLMs) best predict human brain responses to language, one of the most robust findings in computational neurolinguistics, yet why remains mechanistically unexplained. We address this gap by…

AI coding assistants are now used to generate production code in security-sensitive domains, yet the exploitability of their outputs remains unquantified. We address this gap with Broken by Default: a formal verification study of 3,500 code…

We study estimation of an $s$-sparse signal in the $p$-dimensional Gaussian sequence model with equicorrelated observations and derive the minimax rate. A new phenomenon emerges from correlation, namely the rate scales with respect to…

Sparse Autoencoders (SAEs) aim to decompose the activation space of large language models (LLMs) into human-interpretable latent directions or features. As we increase the number of features in the SAE, hierarchical features tend to split…