6 papers
Beg to Differ: Understanding Reasoning-Answer Misalignment Across Languages
Anaelia Ovalle, Candace Ross, Sebastian Ruder +4
Large language models demonstrate strong reasoning capabilities through chain-of-thought prompting, but whether this reasoning quality transfers across languages remains underexplo…
Learning the Wrong Lessons: Syntactic-Domain Spurious Correlations in Language Models
Chantal Shaib, Vinith M. Suriyakumar, Levent Sagun +2
For an LLM to correctly respond to an instruction it must understand both the semantics and the domain (i.e., subject area) of a given task-instruction pair. However, syntax can al…
LLM Knowledge is Brittle: Truthfulness Representations Rely on Superficial Resemblance
Patrick Haller, Mark Ibrahim, Polina Kirichenko +2
For Large Language Models (LLMs) to be reliable, they must learn robust knowledge that can be generally applied in diverse settings -- often unlike those seen during training. Yet,…
On the Role of Speech Data in Reducing Toxicity Detection Bias
Samuel J. Bell, Mariano Coria Meglioli, Megan Richards +6
Text toxicity detection systems exhibit significant biases, producing disproportionate rates of false positives on samples mentioning demographic groups. But what about toxicity de…
The Root Shapes the Fruit: On the Persistence of Gender-Exclusive Harms in Aligned Language Models
Anaelia Ovalle, Krunoslav Lehman Pavasovic, Louis Martin +5
Natural-language assistants are designed to provide users with helpful responses while avoiding harmful outputs, largely achieved through alignment to human preferences. Yet there…
Chained Tuning Leads to Biased Forgetting
Megan Ung, Alicia Sun, Samuel J. Bell +3
Large language models (LLMs) are often fine-tuned for use on downstream tasks, though this can degrade capabilities learned during previous training. This phenomenon, often referre…