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.

Reinforcement Learning from Human Feedback (RLHF) is increasingly used to align large language models (LLMs) with human preferences. However, the effectiveness of RLHF in addressing underlying biases remains unclear. This study investigates the relationship between RLHF and both covert and overt biases in LLMs, particularly focusing on biases against African Americans. We applied various RLHF techniques (DPO, ORPO, and RLOO) to Llama 3 8B and evaluated the covert and overt biases of the resulting models using matched-guise probing and explicit bias testing. We performed additional tests with DPO on different base models and datasets; among several implications, we found that SFT before RLHF calcifies model biases. Additionally, we extend the tools for measuring biases to multi-modal models. Through our experiments we collect evidence that indicates that current alignment techniques are inadequate for nebulous tasks such as mitigating covert biases, highlighting the need for capable datasets, data curating techniques, or alignment tools.

This paper introduces and tests a framework integrating traffic regulation compliance into automated driving systems (ADS). The framework enables ADS to follow traffic laws and make informed decisions based on the driving environment. Using RGB camera inputs and a vision-language model (VLM), the system generates descriptive text to support a regulation-aware decision-making process, ensuring legal and safe driving practices. This information is combined with a machine-readable ADS regulation database to guide future driving plans within legal constraints. Key features include: 1) a regulation database supporting ADS decision-making, 2) an automated process using sensor input for regulation-aware path planning, and 3) validation in both simulated and real-world environments. Particularly, the real-world vehicle tests not only assess the framework's performance but also evaluate the potential and challenges of VLMs to solve complex driving problems by integrating detection, reasoning, and planning. This work enhances the legality, safety, and public trust in ADS, representing a significant step forward in the field.

Large language models (LLMs) have shown great promise as language understanding and decision making tools, and they have permeated various aspects of our everyday life. However, their widespread availability also comes with novel risks, such as generating harmful, unethical, or offensive content, via an attack called jailbreaking. Despite extensive efforts from LLM developers to align LLMs using human feedback, they are still susceptible to jailbreak attacks. To tackle this issue, researchers often employ red-teaming to understand and investigate jailbreak prompts. However, existing red-teaming approaches lack effectiveness, scalability, or both. To address these issues, we propose JBFuzz, a novel effective, automated, and scalable red-teaming technique for jailbreaking LLMs. JBFuzz is inspired by the success of fuzzing for detecting bugs/vulnerabilities in software. We overcome three challenges related to effectiveness and scalability by devising novel seed prompts, a lightweight mutation engine, and a lightweight and accurate evaluator for guiding the fuzzer. Assimilating all three solutions results in a potent fuzzer that only requires black-box access to the target LLM. We perform extensive experimental evaluation of JBFuzz using nine popular and widely-used LLMs. We find that JBFuzz successfully jailbreaks all LLMs for various harmful/unethical questions, with an average attack success rate of 99%. We also find that JBFuzz is extremely efficient as it jailbreaks a given LLM for a given question in 60 seconds on average. Our work highlights the susceptibility of the state-of-the-art LLMs to jailbreak attacks even after safety alignment, and serves as a valuable red-teaming tool for LLM developers.

Large language models (LLMs) have demonstrated remarkable capabilities across various tasks, but ensuring their safety and alignment with human values remains crucial. Current safety alignment methods, such as supervised fine-tuning and reinforcement learning-based approaches, can exhibit vulnerabilities to adversarial attacks and often result in shallow safety alignment, primarily focusing on preventing harmful content in the initial tokens of the generated output. While methods like resetting can help recover from unsafe generations by discarding previous tokens and restarting the generation process, they are not well-suited for addressing nuanced safety violations like toxicity that may arise within otherwise benign and lengthy generations. In this paper, we propose a novel backtracking method designed to address these limitations. Our method allows the model to revert to a safer generation state, not necessarily at the beginning, when safety violations occur during generation. This approach enables targeted correction of problematic segments without discarding the entire generated text, thereby preserving efficiency. We demonstrate that our method dramatically reduces toxicity appearing through the generation process with minimal impact to efficiency.

Due to climate change, the extreme wildfire has become one of the most dangerous natural hazards to human civilization. Even though, some wildfires may be initially caused by human activity, but the spread of wildfires is mainly determined by environmental factors, for examples, (1) weather conditions such as temperature, wind direction and intensity, and moisture levels; (2) the amount and types of dry vegetation in a local area, and (3) topographic or local terrian conditions, which affects how much rain an area gets and how fire dynamics will be constrained or faciliated. Thus, to accurately forecast wildfire occurrence has become one of most urgent and taunting environmental challenges in global scale. In this work, we developed a real-time Multimodal Transformer Neural Network Machine Learning model that combines several advanced artificial intelligence techniques and statistical methods to practically forecast the occurrence of wildfire at the precise location in real time, which not only utilizes large scale data information such as hourly weather forecasting data, but also takes into account small scale topographical data such as local terrain condition and local vegetation conditions collecting from Google Earth images to determine the probabilities of wildfire occurrence location at small scale as well as their timing synchronized with weather forecast information. By using the wildfire data in the United States from 1992 to 2015 to train the multimodal transformer neural network, it can predict the probabilities of wildfire occurrence according to the real-time weather forecast and the synchronized Google Earth image data to provide the wildfire occurrence probability in any small location ($100m^2$) within 24 hours ahead.

Machine Learning Force Fields (MLFFs) are a promising alternative to expensive ab initio quantum mechanical molecular simulations. Given the diversity of chemical spaces that are of interest and the cost of generating new data, it is important to understand how MLFFs generalize beyond their training distributions. In order to characterize and better understand distribution shifts in MLFFs, we conduct diagnostic experiments on chemical datasets, revealing common shifts that pose significant challenges, even for large foundation models trained on extensive data. Based on these observations, we hypothesize that current supervised training methods inadequately regularize MLFFs, resulting in overfitting and learning poor representations of out-of-distribution systems. We then propose two new methods as initial steps for mitigating distribution shifts for MLFFs. Our methods focus on test-time refinement strategies that incur minimal computational cost and do not use expensive ab initio reference labels. The first strategy, based on spectral graph theory, modifies the edges of test graphs to align with graph structures seen during training. Our second strategy improves representations for out-of-distribution systems at test-time by taking gradient steps using an auxiliary objective, such as a cheap physical prior. Our test-time refinement strategies significantly reduce errors on out-of-distribution systems, suggesting that MLFFs are capable of and can move towards modeling diverse chemical spaces, but are not being effectively trained to do so. Our experiments establish clear benchmarks for evaluating the generalization capabilities of the next generation of MLFFs. Our code is available at https://tkreiman.github.io/projects/mlff_distribution_shifts/.

As language agents progressively automate critical tasks across domains, their ability to operate within operational constraints and safety protocols becomes essential. While extensive research has demonstrated these agents' effectiveness in downstream task completion, their reliability in following operational procedures and constraints remains largely unexplored. To this end, we present AgentOrca, a dual-system framework for evaluating language agents' compliance with operational constraints and routines. Our framework encodes action constraints and routines through both natural language prompts for agents and corresponding executable code serving as ground truth for automated verification. Through an automated pipeline of test case generation and evaluation across five real-world domains, we quantitatively assess current language agents' adherence to operational constraints. Our findings reveal notable performance gaps among state-of-the-art models, with large reasoning models like o1 demonstrating superior compliance while others show significantly lower performance, particularly when encountering complex constraints or user persuasion attempts.

Evaluating Large Language Models (LLMs) free-form generated responses remains a challenge due to their diverse and open-ended nature. Traditional supervised signal-based automatic metrics fail to capture semantic equivalence or handle the variability of open-ended responses, while human evaluation, though reliable, is resource-intensive. Leveraging LLMs as evaluators offers a promising alternative due to their strong language understanding and instruction-following capabilities. Taking advantage of these capabilities, we propose the Dynamic Arbitration Framework for Evaluation (DAFE), which employs two primary LLM-as-judges and engages a third arbitrator only in cases of disagreements. This selective arbitration prioritizes evaluation reliability while reducing unnecessary computational demands compared to conventional majority voting. DAFE utilizes task-specific reference answers with dynamic arbitration to enhance judgment accuracy, resulting in significant improvements in evaluation metrics such as Macro F1 and Cohen's Kappa. Through experiments, including a comprehensive human evaluation, we demonstrate DAFE's ability to provide consistent, scalable, and resource-efficient assessments, establishing it as a robust framework for evaluating free-form model outputs.

The purpose of this study is to assess how large language models (LLMs) can be used for fact-checking and contribute to the broader debate on the use of automated means for veracity identification. To achieve this purpose, we use AI auditing methodology that systematically evaluates performance of five LLMs (ChatGPT 4, Llama 3 (70B), Llama 3.1 (405B), Claude 3.5 Sonnet, and Google Gemini) using prompts regarding a large set of statements fact-checked by professional journalists (16,513). Specifically, we use topic modeling and regression analysis to investigate which factors (e.g. topic of the prompt or the LLM type) affect evaluations of true, false, and mixed statements. Our findings reveal that while ChatGPT 4 and Google Gemini achieved higher accuracy than other models, overall performance across models remains modest. Notably, the results indicate that models are better at identifying false statements, especially on sensitive topics such as COVID-19, American political controversies, and social issues, suggesting possible guardrails that may enhance accuracy on these topics. The major implication of our findings is that there are significant challenges for using LLMs for factchecking, including significant variation in performance across different LLMs and unequal quality of outputs for specific topics which can be attributed to deficits of training data. Our research highlights the potential and limitations of LLMs in political fact-checking, suggesting potential avenues for further improvements in guardrails as well as fine-tuning.