Prompting Large Language Models (LLMs), or providing context on the expected model of operation, is an effective way to steer the outputs of such models to satisfy human desiderata after they have been trained. But in rapidly evolving domains, there is often need to fine-tune LLMs to improve either the kind of knowledge in their memory or their abilities to perform open ended reasoning in new domains. When human's learn new concepts, we often do so by linking the new material that we are studying to concepts we have already learned before. To that end, we ask, "can prompting help us teach LLMs how to learn". In this work, we study a novel generalization of instruction tuning, called contextual fine-tuning, to fine-tune LLMs. Our method leverages instructional prompts designed to mimic human cognitive strategies in learning and problem-solving to guide the learning process during training, aiming to improve the model’s interpretation and understanding of domain-specific knowledge. We empirically demonstrate that this simple yet effective modification improves the ability of LLMs to be fine-tuned rapidly on new datasets both within the medical and financial domains.

Vulnerability of Frontier language models to misuse has prompted the development of safety measures like filters and alignment training seeking to ensure safety through robustness to adversarially crafted prompts. We assert that robustness is fundamentally insufficient for ensuring safety goals due to inferential threats from dual-intent queries, with current defenses and evaluations failing to account for these risks. To quantify these risks, we introduce a new safety evaluation framework based on $\textit{impermissible information leakage}$ of model outputs and demonstrate how our proposed question-decomposition attack can extract dangerous knowledge from a censored LLM more effectively than traditional jailbreaking. Underlying our proposed evaluation method is a novel information-theoretic threat model of $\textit{inferential adversaries}$, distinguished from $\textit{security adversaries}$, such as jailbreaks, in that success involves inferring impermissible knowledge from victim outputs as opposed to forcing explicitly impermissible victim outputs. Through our information-theoretic framework, we show that ensuring safety against inferential adversaries requires defenses which bound impermissible information leakage, and, such defenses inevitably incur safety-utility trade-offs.

While scaling training compute has led to remarkable improvements in large language models (LLMs), scaling inference compute only recently began to yield analogous gains. We hypothesize that a core missing component is a lack of diverse LLM outputs, leading to inefficient search due to models repeatedly sampling highly similar, yet incorrect generations. We empirically demonstrate that this lack of diversity can be mitigated by searching over candidate plans for solving a problem in natural language. Based on this insight, we propose PlanSearch, a novel search algorithm which shows strong results across HumanEval+, MBPP+, and LiveCodeBench (a contamination-free benchmark for competitive coding). PlanSearch generates a diverse set of observations about the problem and uses these observations to construct plans for solving the problem. By searching over plans in natural language rather than directly over code solutions, PlanSearch explores a significantly more diverse range of potential solutions compared to baseline search methods. Using PlanSearch on top of Claude 3.5 Sonnet achieves a pass@200 of 77.0% on LiveCodeBench, outperforming both the best pass-rate achieved without any search (pass@1 = 41.4%) and using standard repeated sampling on top of existing non-search models (pass@200 = 60.6%). Finally, we show that, across all models, search algorithms, and benchmarks analyzed, we can accurately predict performance gains from search as a function of the diversity over generated ideas.

Email communication between instructors and students is ubiquitous, and it could be valuable to explore ways of testing out how to make email messages more impactful. This paper explores the design space of using emails to get students to plan and reflect on starting weekly homework earlier. We deployed a series of email reminders using randomized A/B comparisons to test alternative factors in the design of these emails, providing examples of an experimental paradigm and metrics for a broader range of interventions. We also surveyed and interviewed instructors and students to compare their predictions about the effectiveness of the reminders with their actual impact. We present our results on which seemingly obvious predictions about effective emails are not borne out, despite there being evidence for further exploring these interventions, as they can sometimes motivate students to attempt their homework more often. We also present qualitative evidence about student opinions and behaviours after receiving the emails, to guide further interventions. These findings provide insight into how to use randomized A/B comparisons in everyday channels such as emails, to provide empirical evidence to test our beliefs about the effectiveness of alternative design choices.