Facilitated by large language models (LLMs), personalized text generation has become a rapidly growing research direction. Most existing studies focus on designing specialized models for a particular domain, or they require fine-tuning the LLMs to generate personalized text. We consider a typical scenario in which the large language model, which generates personalized output, is frozen and can only be accessed through APIs. Under this constraint, all one can do is to improve the input text (i.e., text prompts) sent to the LLM, a procedure that is usually done manually. In this paper, we propose a novel method to automatically revise prompts for personalized text generation. The proposed method takes the initial prompts generated by a state-of-the-art, multistage framework for personalized generation and rewrites a few critical components that summarize and synthesize the personal context. The prompt rewriter employs a training paradigm that chains together supervised learning (SL) and reinforcement learning (RL), where SL reduces the search space of RL and RL facilitates end-to-end training of the rewriter. Using datasets from three representative domains, we demonstrate that the rewritten prompts outperform both the original prompts and the prompts optimized via supervised learning or reinforcement learning alone. In-depth analysis of the rewritten prompts shows that they are not only human readable, but also able to guide manual revision of prompts when there is limited resource to employ reinforcement learning to train the prompt rewriter, or when it is costly to deploy an automatic prompt rewriter for inference.

To comprehensively assess the capacity of current models for complex reasoning, it is crucial to assess their step-by-step reasoning in a scalable manner. Established reference-based evaluation metrics rely on human-annotated reasoning chains to assess the model-derived chains. However, such ``gold-standard'' human-written reasoning chains may not be unique and their acquisition is often labor-intensive. Existing reference-free reasoning metrics eliminate the need for human-crafted reasoning chains as references, but they typically require fine-tuning on datasets with human-derived reasoning chains, which complicates the process and raises concerns regarding generalizability across diverse datasets. To address these challenges, we harness GPT-4 to automatically evaluate reasoning chain quality, obviating the need for human-crafted references. Leveraging the Socratic method, we devise tailored prompts to enhance reference-free reasoning evaluation, which we term SocREval (Socratic method for Reasoning Evaluation). Empirical results from four human annotated datasets reveal that SocREval significantly improves GPT-4's performance, surpassing existing reference-free and reference-based reasoning evaluation metrics. Beyond its demonstrated efficacy, our proposed framework, large language models (LLMs) with the Socratic method, proves to be both cost-efficient and robust to prompt writing and example selection, as substantiated by our in-depth analysis.

This paper assesses the potential for the large language models (LLMs) GPT-4 and GPT-3.5 to aid in deriving insight from education feedback surveys. Exploration of LLM use cases in education has focused on teaching and learning, with less exploration of capabilities in education feedback analysis. Survey analysis in education involves goals such as finding gaps in curricula or evaluating teachers, often requiring time-consuming manual processing of textual responses. LLMs have the potential to provide a flexible means of achieving these goals without specialized machine learning models or fine-tuning. We demonstrate a versatile approach to such goals by treating them as sequences of natural language processing (NLP) tasks including classification (multi-label, multi-class, and binary), extraction, thematic analysis, and sentiment analysis, each performed by LLM. We apply these workflows to a real-world dataset of 2500 end-of-course survey comments from biomedical science courses, and evaluate a zero-shot approach (i.e., requiring no examples or labeled training data) across all tasks, reflecting education settings, where labeled data is often scarce. By applying effective prompting practices, we achieve human-level performance on multiple tasks with GPT-4, enabling workflows necessary to achieve typical goals. We also show the potential of inspecting LLMs' chain-of-thought (CoT) reasoning for providing insight that may foster confidence in practice. Moreover, this study features development of a versatile set of classification categories, suitable for various course types (online, hybrid, or in-person) and amenable to customization. Our results suggest that LLMs can be used to derive a range of insights from survey text.

Large language models (LLMs) are often augmented with tools to solve complex tasks. By generating code snippets and executing them through task-specific Application Programming Interfaces (APIs), they can offload certain functions to dedicated external modules, such as image encoding and performing calculations. However, most existing approaches to augment LLMs with tools are constrained by general-purpose APIs and lack the flexibility for tailoring them to specific tasks. In this work, we present CRAFT, a general tool creation and retrieval framework for LLMs. It creates toolsets specifically curated for the tasks and equips LLMs with a component that retrieves tools from these sets to enhance their capability to solve complex tasks. For each task, we collect specific code solutions by prompting GPT-4 to solve the training examples. Following a validation step ensuring the correctness, these solutions are abstracted into code snippets to enhance reusability, and deduplicated for higher quality. At inference time, the language model retrieves snippets from the toolsets and then executes them or generates the output conditioning on the retrieved snippets. Our method is designed to be flexible and offers a plug-and-play approach to adapt off-the-shelf LLMs to unseen domains and modalities, without any finetuning. Experiments on vision-language, tabular processing, and mathematical reasoning tasks show that our approach achieves substantial improvements compared to strong baselines. In addition, our in-depth analysis reveals that: (1) consistent performance improvement can be achieved by scaling up the number of tools and the capability of the backbone models; (2) each component of our approach contributes to the performance gains; (3) the created tools are well-structured and reliable with low complexity and atomicity. The code is available at \url{https://github.com/lifan-yuan/CRAFT}.

Pretrained language models sometimes possess knowledge that we do not wish them to, including memorized personal information and knowledge that could be used to harm people. They can also output toxic or harmful text. To mitigate these safety and informational issues, we propose an attack-and-defense framework for studying the task of deleting sensitive information directly from model weights. We study direct edits to model weights because (1) this approach should guarantee that particular deleted information is never extracted by future prompt attacks, and (2) it should protect against whitebox attacks, which is necessary for making claims about safety/privacy in a setting where publicly available model weights could be used to elicit sensitive information. Our threat model assumes that an attack succeeds if the answer to a sensitive question is located among a set of B generated candidates, based on scenarios where the information would be insecure if the answer is among B candidates. Experimentally, we show that even state-of-the-art model editing methods such as ROME struggle to truly delete factual information from models like GPT-J, as our whitebox and blackbox attacks can recover "deleted" information from an edited model 38% of the time. These attacks leverage two key observations: (1) that traces of deleted information can be found in intermediate model hidden states, and (2) that applying an editing method for one question may not delete information across rephrased versions of the question. Finally, we provide new defense methods that protect against some extraction attacks, but we do not find a single universally effective defense method. Our results suggest that truly deleting sensitive information is a tractable but difficult problem, since even relatively low attack success rates have potentially severe societal implications for real-world deployment of language models.

Large language models (LLMs), including ChatGPT, can extract profitable trading signals from the sentiment in news text. However, backtesting such strategies poses a challenge because LLMs are trained on many years of data, and backtesting produces biased results if the training and backtesting periods overlap. This bias can take two forms: a look-ahead bias, in which the LLM may have specific knowledge of the stock returns that followed a news article, and a distraction effect, in which general knowledge of the companies named interferes with the measurement of a text's sentiment. We investigate these sources of bias through trading strategies driven by the sentiment of financial news headlines. We compare trading performance based on the original headlines with de-biased strategies in which we remove the relevant company's identifiers from the text. In-sample (within the LLM training window), we find, surprisingly, that the anonymized headlines outperform, indicating that the distraction effect has a greater impact than look-ahead bias. This tendency is particularly strong for larger companies--companies about which we expect an LLM to have greater general knowledge. Out-of-sample, look-ahead bias is not a concern but distraction remains possible. Our proposed anonymization procedure is therefore potentially useful in out-of-sample implementation, as well as for de-biased backtesting.

Large language models (LLMs) have exhibited remarkable ability in textual generation. However, in complex reasoning tasks such as code generation, generating the correct answer in a single attempt remains a formidable challenge for LLMs. Previous research has explored solutions by aggregating multiple outputs, leveraging the consistency among them. However, none of them have comprehensively captured this consistency from different perspectives. In this paper, we propose the Multi-Perspective Self-Consistency (MPSC) framework, a novel decoding strategy for LLM that incorporates both inter-consistency across outputs from multiple perspectives and intra-consistency within a single perspective. Specifically, we ask LLMs to sample multiple diverse outputs from various perspectives for a given query and then construct a multipartite graph based on them. With two predefined measures of consistency, we embed both inter- and intra-consistency information into the graph. The optimal choice is then determined based on consistency analysis in the graph. We conduct comprehensive evaluation on the code generation task by introducing solution, specification and test case as three perspectives. We leverage a code interpreter to quantitatively measure the inter-consistency and propose several intra-consistency measure functions. Our MPSC framework significantly boosts the performance on various popular benchmarks, including HumanEval (+17.60%), HumanEval Plus (+17.61%), MBPP (+6.50%) and CodeContests (+11.82%) in Pass@1, when compared to original outputs generated from ChatGPT, and even surpassing GPT-4.

There is a growing interest in using Large Language Models (LLMs) as agents to tackle real-world tasks that may require assessing complex situations. Yet, we have a limited understanding of LLMs' reasoning and decision-making capabilities, partly stemming from a lack of dedicated evaluation benchmarks. As negotiating and compromising are key aspects of our everyday communication and collaboration, we propose using scorable negotiation games as a new evaluation framework for LLMs. We create a testbed of diverse text-based, multi-agent, multi-issue, semantically rich negotiation games, with easily tunable difficulty. To solve the challenge, agents need to have strong arithmetic, inference, exploration, and planning capabilities, while seamlessly integrating them. Via a systematic zero-shot Chain-of-Thought prompting (CoT), we show that agents can negotiate and consistently reach successful deals. We quantify the performance with multiple metrics and observe a large gap between GPT-4 and earlier models. Importantly, we test the generalization to new games and setups. Finally, we show that these games can help evaluate other critical aspects, such as the interaction dynamics between agents in the presence of greedy and adversarial players.

FP8 formats are gaining popularity to boost the computational efficiency for training and inference of large deep learning models. Their main challenge is that a careful choice of scaling is needed to prevent degradation due to the reduced dynamic range compared to higher-precision formats. Although there exists ample literature about selecting such scalings for INT formats, this critical aspect has yet to be addressed for FP8. This paper presents a methodology to select the scalings for FP8 linear layers, based on dynamically updating per-tensor scales for the weights, gradients and activations. We apply this methodology to train and validate large language models of the type of GPT and Llama 2 using FP8, for model sizes ranging from 111M to 70B. To facilitate the understanding of the FP8 dynamics, our results are accompanied by plots of the per-tensor scale distribution for weights, activations and gradients during both training and inference.

Large language models (LLMs) typically employ sampling or beam search, accompanied by prompts such as Chain-of-Thought (CoT), to boost reasoning and decoding ability. Recent work like Tree-of-Thought (ToT) and Reasoning via Planning (RAP) aim to augment the reasoning capabilities of LLMs by utilizing tree-search algorithms to guide multi-step reasoning. These methods mainly focus on LLMs' reasoning ability during inference and heavily rely on human-designed prompts to activate LLM as a value function, which lacks general applicability and scalability. To address these limitations, we present an AlphaZero-like tree-search framework for LLMs (termed TS-LLM), systematically illustrating how tree-search with a learned value function can guide LLMs' decoding ability. TS-LLM distinguishes itself in two key ways: (1) Leveraging a learned value function, our approach can be generally applied to different tasks beyond reasoning (such as RLHF alignment), and LLMs of any size, without prompting advanced, large-scale models. (2) It can guide LLM's decoding during both inference and training. Empirical evaluations across reasoning, planning, and RLHF alignment tasks validate the effectiveness of TS-LLM, even on trees with a depth of 64.