Nicholas Joseph

TL;DR: An iterated online mode of training, where preference models and RL policies are updated on a weekly cadence with fresh human feedback data, and a roughly linear relation between the RL reward and the square root of the KL divergence between the policy and its initialization is identified.

TL;DR: In this article , the authors use RL from AI Feedback (RLAIF) to train a harmless but non-evasive AI assistant that engages with harmful queries by explaining its objections to them.

TL;DR: This article showed that large models are well-calibrated on diverse multiple choice and true/false questions when they are provided in the right format, and showed that models can be trained to predict the probability that"I know"the answer to a question, without reference to any particular proposed answer.

TL;DR: It is found that induction heads develop at precisely the same point as a sudden sharp increase in incontext learning ability, visible as a bump in the training loss.

TL;DR: This paper highlights a counterintuitive property of large-scale generative models, which have a paradoxical combination of predictable loss on a broad training distribution, and unpredictable specific capabilities, inputs, and outputs, and analyzed how these conflicting properties combine to give model developers various motivations for deploying these models, and challenges that can hinder deployment.