Humans possess an extraordinary ability to create and utilize tools. With the advent of
foundation models, artificial intelligence systems have the potential to be equally adept in …

AI alignment aims to make AI systems behave in line with human intentions and values. As
AI systems grow more capable, so do risks from misalignment. To provide a comprehensive …

Large language model (LLM) systems, such as ChatGPT or Gemini, can show impressive
reasoning and question-answering capabilities but often 'hallucinate'false outputs and …

Abstract Trustworthy Artificial Intelligence (AI) is based on seven technical requirements
sustained over three main pillars that should be met throughout the system's entire life cycle …

Reinforcement learning from human feedback (RLHF) is a technique for training AI systems
to align with human goals. RLHF has emerged as the central method used to finetune state …

We present GPQA, a challenging dataset of 448 multiple-choice questions written by domain
experts in biology, physics, and chemistry. We ensure that the questions are high-quality …

Reinforcement learning from human feedback (RLHF) is an effective technique for aligning
large language models (LLMs) to human preferences, but gathering high-quality human …

Abstract While Large Language Models (LLMs) display versatile functionality, they continue
to generate harmful, biased, and toxic content, as demonstrated by the prevalence of human …

Reinforcement learning algorithms typically struggle in the absence of a dense, well-shaped
reward function. Intrinsically motivated exploration methods address this limitation by …

Language models demonstrate both quantitative improvement and new qualitative
capabilities with increasing scale. Despite their potentially transformative impact, these new …