State of the Field
Current research on large language models (LLMs) is increasingly focused on understanding and improving their reasoning capabilities and knowledge representation. Recent work has highlighted the significance of reasoning traces, revealing that the accuracy of LLM responses improves as more of these traces are provided, suggesting a need for better trace management in deployment contexts. Additionally, the exploration of long-tail knowledge has uncovered persistent gaps in LLM performance, particularly for infrequent or domain-specific information, prompting calls for more robust evaluation practices and interventions to enhance fairness and accountability. Studies on the differential encoding of syntactic and semantic information are shedding light on how these elements are represented within models, while investigations into motivated reasoning reveal discrepancies between LLM behavior and human reasoning patterns. Collectively, these insights are guiding the development of more reliable and interpretable LLMs, with potential applications in areas such as automated survey analysis and decision support systems.
Papers
1–6 of 6Probing the Trajectories of Reasoning Traces in Large Language Models
Large language models (LLMs) increasingly solve difficult problems by producing "reasoning traces" before emitting a final response. However, it remains unclear how accuracy and decision commitment ev...
Long-Tail Knowledge in Large Language Models: Taxonomy, Mechanisms, Interventions and Implications
Large language models (LLMs) are trained on web-scale corpora that exhibit steep power-law distributions, in which the distribution of knowledge is highly long-tailed, with most appearing infrequently...
Differential syntactic and semantic encoding in LLMs
We study how syntactic and semantic information is encoded in inner layer representations of Large Language Models (LLMs), focusing on the very large DeepSeek-V3. We find that, by averaging hidden-rep...
Replicating Human Motivated Reasoning Studies with LLMs
Motivated reasoning -- the idea that individuals processing information may be motivated to reach a certain conclusion, whether it be accurate or predetermined -- has been well-explored as a human phe...
The Potential of CoT for Reasoning: A Closer Look at Trace Dynamics
Chain-of-thought (CoT) prompting is a de-facto standard technique to elicit reasoning-like responses from large language models (LLMs), allowing them to spell out individual steps before giving a fina...
Causality Elicitation from Large Language Models
Large language models (LLMs) are trained on enormous amounts of data and encode knowledge in their parameters. We propose a pipeline to elicit causal relationships from LLMs. Specifically, (i) we samp...