Outputs generated by LLMs reflect traits present in the training data. However, it is unclear how the model acquires these traits and steers itself in that direction. We demonstrate that allowing the model to continue generating beyond the <eos> token reveals its comprehension of the task in the form of specific ’Notes’ or Rules distinct from normal generation. We refer to this as the model’s Internal Monologue (IM) and demonstrate its utility to understand the model’s perception of the task it is being finetuned on. We find this monologue is transient and dissipates as training progresses. We hypothesize this is a inherent behavior of instruction tuned models that adapts to reflect task instructions until the latent space beyond <eos> is sufficiently altered. Importantly, we do not ask the model to reason or do anything special to invoke this. We perform analysis on four diverse dialogue datasets from the Counseling domain to compare how dataset differences affect the internal monologue during early finetuning. We then show this is a general phenomenon by testing with three models of varying sizes and families and other domain datasets. We demonstrate the utility of IM as a means to gain insight on the model’s perception of the dataset and task through clustering based visualizations.

While instruction-tuned Large Language Models exhibit an "AI Assistant" persona, it remains heavily debated whether this behavior is merely one of many simulated roles or the fundamental structural identity of the model. In this work, we present evidence that instruction-tuning instills a global, underlying Assistant State that not only bounds autoregressive generation but is also capable of meta-awareness and reflecting on its own outputs. We demonstrate that this behavior surfaces naturally following the stop token (<eos>) without any special training or prompting. Furthermore, we show that this state can be deterministically triggered during generation via a steering vector that isolates the model’s post-execution geometry. We introduce this methodology as a zero-shot auditing tool for model evaluators. Applying this framework reveals that models demonstrating Emergent Misalignment in early checkpoints show dual nature where the main generation demonstrates misalignment but post <eos> text provides safety warning or rationalizes it as hypothetical.

Interest in leveraging Large Language Models (LLMs) for emotional support systems motivates the need to understand how these models comprehend and represent emotions internally. While recent works show the presence of emotion concepts in the hidden state representations, it’s unclear if the model has a robust representation that is consistent across different datasets. In this paper, we present a unified view to understand emotion representation in LLMs, experimenting with diverse datasets and prompts. We then evaluate the reasoning ability of the models on a complex emotion identification task. We find that LLMs have a common emotion representation in the later layers of the model, and the vectors capturing the direction of emotions extracted from these representations can be interchanged among datasets with minimal impact on performance. Our analysis of reasoning with Chain of Thought (CoT) prompting shows the limits of emotion comprehension. Therefore, despite LLMs implicitly having emotion representations, they are not equally skilled at reasoning with them in complex scenarios. This motivates the need for further research to find new approaches.

A NOTAM or NOtice To AirMen is a crucial notice for different aviation stakeholders, particularly flight crews. It delivers essential notifications about abnormal conditions of Aviation System components such as changes to facilities, hazards, service, procedure that are not known far enough in advance to be publicized through other means. NOTAM messages are short, contain acronyms, and look cryptic in most of the cases. Writing and understanding these messages put heavy cognitive load on its end users. In this work, we take up the task of translating NOTAMs into English natural language using LLMs. Since NOTAMs do not adhere to English grammar rules and have their own decoding rules, large language models (LLMs) cannot translate them without effective prompting. In this paper, we develop a framework to come up with effective prompts to achieve the translations. Our approach uses context-aware semantic prompting techniques, paired with domain-specific rules, to improve the accuracy and clarity of translations. The framework is evaluated using comprehensive experiments (6 LLMs of varying sizes, and with 5 different prompting setups for each) and eight evaluation metrics measuring different aspects of the translation. The results demonstrate that our methodology can produce clear translations that accurately convey the information contained in NOTAMs.