The article focuses on understanding the mechanisms of alignment algorithms, particularly Direct Preference Optimization (DPO), and how they reduce toxicity in language models like GPT2. The researchers first study how…
Continue readingThe study introduces LM Assertions, a novel programming construct designed to enforce user-specified properties on Language Model (LM) outputs within a pipeline framework. This is integrated into the DSPy programming…
Continue readingThe research presents a novel tool called ‘time vectors’ for customizing language models according to specific time periods. This is achieved by finetuning a language model on data from a…
Continue readingThis study revisits the issue of pruning large language models (LLMs), particularly from the BERT family, in response to the Sparsity May Cry (SMC) benchmark. The benchmark highlights the complexity…
Continue readingThe article introduces Structured Probabilistic Coding (SPC), a novel supervised representation learning framework that extracts compact and informative representations from input related to a target task. The SPC uses an…
Continue readingDespite declining to respond to controversial prompts, Large Language Models (LLMs) may still exhibit sociodemographic biases in their latent representations. This study proposes a logistic Bradley Terry probe to detect…
Continue readingThis article by researchers from UC Berkeley and Google DeepMind explores leveraging reinforcement learning (RL) to enhance the capabilities of large language models (LLMs). The authors note that standard prompting…
Continue readingThe study presents a dataset for testing the understanding of the rationale behind critical reasoning. Results show that recent large language models struggle to answer subquestions even if they can…
Continue readingThe article discusses a new framework called Induction-Augmented Generation (IAG) for answering reasoning questions. This framework utilizes inductive knowledge and retrieved documents for implicit reasoning. The authors propose two versions…
Continue readingThis research paper explores the potential for ‘jailbreak backdoors’ in large language models trained with Reinforcement Learning from Human Feedback (RLHF). It reveals that a malicious actor could potentially poison…
Continue readingLooks like you're using an ad blocker. Please support us by disabling these ad blocker.
