相关论文: How Transformers Reject Wrong Answers: Rotational …

Why do language models trained on contradictory data prefer correct answers? In controlled experiments with small transformers (3.5M--86M parameters), we show that this preference tracks the compressibility structure of errors rather than…

Factual incorrectness in generated content is one of the primary concerns in ubiquitous deployment of large language models (LLMs). Prior findings suggest LLMs can (sometimes) detect factual incorrectness in their generated content (i.e.,…

Reasoning models are evaluated on single-turn benchmarks but deployed in multi-turn dialogue, where users push back on correct answers. Under sustained adversarial pressure we find a previously undocumented failure mode: the…

Language models are becoming the default interface to factual knowledge, yet they often verify outputs more reliably than they generate them. This generation-verification gap (GV-gap) underlies many recent advances in self-improvement and…

Current alignment evaluation mostly measures whether models encode dangerous concepts and whether they refuse harmful requests. Both miss the layer where alignment often operates: routing from concept detection to behavioral policy. We…

Reasoning models have attracted increasing attention for their ability to tackle complex tasks, embodying the System II (slow thinking) paradigm in contrast to System I (fast, intuitive responses). Yet a key question remains: Does slower…

Existing explainability methods for Large Language Models (LLMs) typically treat hidden states as static points in activation space, assuming that correct and incorrect inferences can be separated using representations from an individual…

Large Language Models (LLMs) have impressive capabilities, but are prone to outputting falsehoods. Recent work has developed techniques for inferring whether a LLM is telling the truth by training probes on the LLM's internal activations.…

Predicting upcoming events is critical to our ability to interact with our environment. Transformer models, trained on next-word prediction, appear to construct representations of linguistic input that can support diverse downstream tasks.…

LLMs reliably correct false claims when presented in isolation, yet when the same claims are embedded in task-oriented requests, they often comply rather than correct. We term this failure mode \emph{correction suppression} and construct a…

We discover that large language models exhibit \emph{spectral phase transitions} in their hidden activation spaces when engaging in reasoning versus factual recall. Through systematic spectral analysis across \textbf{11 models} spanning…

Model transformations operate on models conforming to precisely defined metamodels. Consequently, it often seems relatively easy to chain them: the output of a transformation may be given as input to a second one if metamodels match.…

Despite advancements in large language models (LLMs), non-factual responses still persist in fact-seeking question answering. Unlike extensive studies on post-hoc detection of these responses, this work studies non-factuality prediction…

Bidirectional Encoder Representations from Transformers (BERT) reach state-of-the-art results in a variety of Natural Language Processing tasks. However, understanding of their internal functioning is still insufficient and unsatisfactory.…

For Large Language Models (LLMs) to be reliably deployed, models must effectively know when not to answer: abstain. Reasoning models, in particular, have gained attention for impressive performance on complex tasks. However, reasoning…

Transformers have significantly advanced the field of natural language processing, but comprehending their internal mechanisms remains a challenge. In this paper, we introduce a novel geometric perspective that elucidates the inner…

Understanding subjectivity demands reasoning skills beyond the realm of common knowledge. It requires a machine learning model to process sentiment and to perform opinion mining. In this work, I've exploited a recently released dataset for…

Despite recent success, large neural models often generate factually incorrect text. Compounding this is the lack of a standard automatic evaluation for factuality--it cannot be meaningfully improved if it cannot be measured. Grounded…

Large Language Models (LLMs) exhibit strong conversational abilities but often generate falsehoods. Prior work suggests that the truthfulness of simple propositions can be represented as a single linear direction in a model's internal…

Language model representations often contain linear directions that correspond to high-level concepts. Here, we study the dynamics of these representations: how representations evolve along these dimensions within the context of (simulated)…