The ML community is rapidly exploring techniques for prompting language models (LMs) and for stacking them into pipelines that solve complex tasks. Unfortunately, existing LM pipelines are typically implemented using hard-coded "prompt templates", i.e. lengthy strings discovered via trial and error. Toward a more systematic approach for developing and optimizing LM pipelines, we introduce DSPy, a programming model that abstracts LM pipelines as text transformation graphs, i.e. imperative computational graphs where LMs are invoked through declarative modules. DSPy modules are parameterized, meaning they can learn (by creating and collecting demonstrations) how to apply compositions of prompting, finetuning, augmentation, and reasoning techniques. We design a compiler that will optimize any DSPy pipeline to maximize a given metric. We conduct two case studies, showing that succinct DSPy programs can express and optimize sophisticated LM pipelines that reason about math word problems, tackle multi-hop retrieval, answer complex questions, and control agent loops. Within minutes of compiling, a few lines of DSPy allow GPT-3.5 and llama2-13b-chat to self-bootstrap pipelines that outperform standard few-shot prompting (generally by over 25% and 65%, respectively) and pipelines with expert-created demonstrations (by up to 5-46% and 16-40%, respectively). On top of that, DSPy programs compiled to open and relatively small LMs like 770M-parameter T5 and llama2-13b-chat are competitive with approaches that rely on expert-written prompt chains for proprietary GPT-3.5. DSPy is available at https://github.com/stanfordnlp/dspy

We introduce a method to improve the zero-shot reasoning abilities of large language models on general language understanding tasks. Specifically, we build an autonomous agent to instruct the reasoning process of large language models. We show this approach further unleashes the zero-shot reasoning abilities of large language models to more tasks. We study the performance of our method on a wide set of datasets spanning generation, classification, and reasoning. We show that our method generalizes to most tasks and obtains state-of-the-art zero-shot performance on 20 of the 29 datasets that we evaluate. For instance, our method boosts the performance of state-of-the-art large language models by a large margin, including Vicuna-13b (13.3%), Llama-2-70b-chat (23.2%), and GPT-3.5 Turbo (17.0%). Compared to zero-shot chain of thought, our improvement in reasoning is striking, with an average increase of 10.5%. With our method, Llama-2-70b-chat outperforms zero-shot GPT-3.5 Turbo by 10.2%.

Optimizing large language models (LLMs) for downstream use cases often involves the customization of pre-trained LLMs through further fine-tuning. Meta's open release of Llama models and OpenAI's APIs for fine-tuning GPT-3.5 Turbo on custom datasets also encourage this practice. But, what are the safety costs associated with such custom fine-tuning? We note that while existing safety alignment infrastructures can restrict harmful behaviors of LLMs at inference time, they do not cover safety risks when fine-tuning privileges are extended to end-users. Our red teaming studies find that the safety alignment of LLMs can be compromised by fine-tuning with only a few adversarially designed training examples. For instance, we jailbreak GPT-3.5 Turbo's safety guardrails by fine-tuning it on only 10 such examples at a cost of less than $0.20 via OpenAI's APIs, making the model responsive to nearly any harmful instructions. Disconcertingly, our research also reveals that, even without malicious intent, simply fine-tuning with benign and commonly used datasets can also inadvertently degrade the safety alignment of LLMs, though to a lesser extent. These findings suggest that fine-tuning aligned LLMs introduces new safety risks that current safety infrastructures fall short of addressing -- even if a model's initial safety alignment is impeccable, it is not necessarily to be maintained after custom fine-tuning. We outline and critically analyze potential mitigations and advocate for further research efforts toward reinforcing safety protocols for the custom fine-tuning of aligned LLMs.

In recent years, many interpretability methods have been proposed to help interpret the internal states of Transformer-models, at different levels of precision and complexity. Here, to analyze encoder-decoder Transformers, we propose a simple, new method: DecoderLens. Inspired by the LogitLens (for decoder-only Transformers), this method involves allowing the decoder to cross-attend representations of intermediate encoder layers instead of using the final encoder output, as is normally done in encoder-decoder models. The method thus maps previously uninterpretable vector representations to human-interpretable sequences of words or symbols. We report results from the DecoderLens applied to models trained on question answering, logical reasoning, speech recognition and machine translation. The DecoderLens reveals several specific subtasks that are solved at low or intermediate layers, shedding new light on the information flow inside the encoder component of this important class of models.

Aligning terminological resources, including ontologies, controlled vocabularies, taxonomies, and value sets is a critical part of data integration in many domains such as healthcare, chemistry, and biomedical research. Entity mapping is the process of determining correspondences between entities across these resources, such as gene identifiers, disease concepts, or chemical entity identifiers. Many tools have been developed to compute such mappings based on common structural features and lexical information such as labels and synonyms. Lexical approaches in particular often provide very high recall, but low precision, due to lexical ambiguity. As a consequence of this, mapping efforts often resort to a labor intensive manual mapping refinement through a human curator. Large Language Models (LLMs), such as the ones employed by ChatGPT, have generalizable abilities to perform a wide range of tasks, including question-answering and information extraction. Here we present MapperGPT, an approach that uses LLMs to review and refine mapping relationships as a post-processing step, in concert with existing high-recall methods that are based on lexical and structural heuristics. We evaluated MapperGPT on a series of alignment tasks from different domains, including anatomy, developmental biology, and renal diseases. We devised a collection of tasks that are designed to be particularly challenging for lexical methods. We show that when used in combination with high-recall methods, MapperGPT can provide a substantial improvement in accuracy, beating state-of-the-art (SOTA) methods such as LogMap.

Deep Neural Networks (DNNs) have been the driving force behind many of the recent advances in machine learning. However, research has shown that DNNs are vulnerable to adversarial examples -- input samples that have been perturbed to force DNN-based models to make errors. As a result, Adversarial Machine Learning (AdvML) has gained a lot of attention, and researchers have investigated these vulnerabilities in various settings and modalities. In addition, DNNs have also been found to incorporate embedded bias and often produce unexplainable predictions, which can result in anti-social AI applications. The emergence of new AI technologies that leverage Large Language Models (LLMs), such as ChatGPT and GPT-4, increases the risk of producing anti-social applications at scale. AdvML for Social Good (AdvML4G) is an emerging field that repurposes the AdvML bug to invent pro-social applications. Regulators, practitioners, and researchers should collaborate to encourage the development of pro-social applications and hinder the development of anti-social ones. In this work, we provide the first comprehensive review of the emerging field of AdvML4G. This paper encompasses a taxonomy that highlights the emergence of AdvML4G, a discussion of the differences and similarities between AdvML4G and AdvML, a taxonomy covering social good-related concepts and aspects, an exploration of the motivations behind the emergence of AdvML4G at the intersection of ML4G and AdvML, and an extensive summary of the works that utilize AdvML4G as an auxiliary tool for innovating pro-social applications. Finally, we elaborate upon various challenges and open research issues that require significant attention from the research community.

There is a niche of companies responsible for intermediating the purchase of large batches of varied products for other companies, for which the main challenge is to perform product description standardization, i.e., matching an item described by a client with a product described in a catalog. The problem is complex since the client's product description may be: (1) potentially noisy; (2) short and uninformative (e.g., missing information about model and size); and (3) cross-language. In this paper, we formalize this problem as a ranking task: given an initial client product specification (query), return the most appropriate standardized descriptions (response). In this paper, we propose TPDR, a two-step Transformer-based Product and Class Description Retrieval method that is able to explore the semantic correspondence between IS and SD, by exploiting attention mechanisms and contrastive learning. First, TPDR employs the transformers as two encoders sharing the embedding vector space: one for encoding the IS and another for the SD, in which corresponding pairs (IS, SD) must be close in the vector space. Closeness is further enforced by a contrastive learning mechanism leveraging a specialized loss function. TPDR also exploits a (second) re-ranking step based on syntactic features that are very important for the exact matching (model, dimension) of certain products that may have been neglected by the transformers. To evaluate our proposal, we consider 11 datasets from a real company, covering different application contexts. Our solution was able to retrieve the correct standardized product before the 5th ranking position in 71% of the cases and its correct category in the first position in 80% of the situations. Moreover, the effectiveness gains over purely syntactic or semantic baselines reach up to 3.7 times, solving cases that none of the approaches in isolation can do by themselves.

Memory-efficient large language models are good at refining text input for better readability. However, controllability is a matter of concern when it comes to text generation tasks with long inputs, such as multi-document summarization. In this work, we investigate for a generic controllable approach for multi-document summarization that leverages the capabilities of LLMs to refine the text. In particular, we train a controllable content extraction scheme to extract the text that will be refined by an LLM. The scheme is designed with a novel coverage and coherence intuitive policy, which is duly rewarded by a passively trained LLM. Our approach yields competitive results in the evaluation using ROUGE metrics and outperforms potential baselines in coherence, as per human evaluation.

Multi-document summarization is a challenging task due to its inherent subjective bias, highlighted by the low inter-annotator ROUGE-1 score of 0.4 among DUC-2004 reference summaries. In this work, we aim to enhance the objectivity of news summarization by focusing on the main event of a group of related news documents and presenting it coherently with sufficient context. Our primary objective is to succinctly report the main event, ensuring that the summary remains objective and informative. To achieve this, we employ an extract-rewrite approach that incorporates a main-event biased monotone-submodular function for content selection. This enables us to extract the most crucial information related to the main event from the document cluster. To ensure coherence, we utilize a fine-tuned Language Model (LLM) for rewriting the extracted content into a coherent text. The evaluation using objective metrics and human evaluators confirms the effectiveness of our approach, as it surpasses potential baselines, demonstrating excellence in both content coverage, coherence, and informativeness.

Nowadays, billions of people engage in communication and express their opinions on the internet daily. Unfortunately, not all of these expressions are friendly or compliant, making content moderation an indispensable task. With the successful development of Large Language Models (LLMs) in recent years, LLM-based methods have become a feasible solution for handling tasks in various domains. However, in the field of content moderation, there is still a lack of detailed work that systematically introduces implementation details. In this paper, we introduce how to fine-tune an LLM model that can be privately deployed for content moderation. Specifically, we discuss whether incorporating reasons during the fine-tuning process would be better or if it should be treated as a classification task directly. We also explore the benefits of utilizing reasons generated by more powerful LLMs for fine-tuning privately deployed models and the impact of different processing approaches when the answers generated by the more powerful LLMs are incorrect. We report the entire research process and the key findings in this paper, hoping to provide valuable experience for researchers who are fine-tuning privately deployed models in their domain-specific research.