Large Language Models (LLMs) have shown remarkable capabilities across a wide variety of Natural Language Processing (NLP) tasks and have attracted attention from multiple domains, including financial services. Despite the extensive research into general-domain LLMs, and their immense potential in finance, Financial LLM (FinLLM) research remains limited. This survey provides a comprehensive overview of FinLLMs, including their history, techniques, performance, and opportunities and challenges. Firstly, we present a chronological overview of general-domain Pre-trained Language Models (PLMs) through to current FinLLMs, including the GPT-series, selected open-source LLMs, and financial LMs. Secondly, we compare five techniques used across financial PLMs and FinLLMs, including training methods, training data, and fine-tuning methods. Thirdly, we summarize the performance evaluations of six benchmark tasks and datasets. In addition, we provide eight advanced financial NLP tasks and datasets for developing more sophisticated FinLLMs. Finally, we discuss the opportunities and the challenges facing FinLLMs, such as hallucination, privacy, and efficiency. To support AI research in finance, we compile a collection of accessible datasets and evaluation benchmarks on GitHub.

The ever-growing ecosystem of LLMs has posed a challenge in selecting the most appropriate pre-trained model to fine-tune amidst a sea of options. Given constrained resources, fine-tuning all models and making selections afterward is unrealistic. In this work, we formulate this resource-constrained selection task into predicting fine-tuning performance and illustrate its natural connection with scaling laws. Unlike pre-training, We find that the fine-tuning scaling curve includes not just the well-known "power phase" but also the previously unobserved "pre-power phase". We also explain why existing scaling laws fail to capture this phase transition phenomenon both theoretically and empirically. To address this, we introduce the concept of "pre-learned data size" into our rectified scaling law, which overcomes theoretical limitations and fits experimental results much better. By leveraging our law, we propose a novel LLM selection algorithm that selects the near-optimal model with hundreds of times less resource consumption, while other methods may provide negatively correlated selection.

In-context learning with Large Language Models (LLMs) has emerged as a promising avenue of research in Dialog State Tracking (DST). However, the best-performing in-context learning methods involve retrieving and adding similar examples to the prompt, requiring access to labeled training data. Procuring such training data for a wide range of domains and applications is time-consuming, expensive, and, at times, infeasible. While zero-shot learning requires no training data, it significantly lags behind the few-shot setup. Thus, `\textit{Can we efficiently generate synthetic data for any dialogue schema to enable few-shot prompting?}' Addressing this question, we propose \method, a data generation framework tailored for DST, utilizing LLMs. Our approach only requires the dialogue schema and a few hand-crafted dialogue templates to synthesize natural, coherent, and free-flowing dialogues with DST annotations. Few-shot learning using data from {\method} results in $4-5%$ improvement in Joint Goal Accuracy over the zero-shot baseline on MultiWOZ 2.1 and 2.4. Remarkably, our few-shot learning approach recovers nearly $98%$ of the performance compared to the few-shot setup using human-annotated training data. Our synthetic data and code can be accessed at https://github.com/apple/ml-synthdst

Federated Learning (FL) is a privacy-preserving distributed learning approach that is rapidly developing in an era where privacy protection is increasingly valued. It is this rapid development trend, along with the continuous emergence of new demands for FL in the real world, that prompts us to focus on a very important problem: Federated Learning with New Knowledge. The primary challenge here is to effectively incorporate various new knowledge into existing FL systems and evolve these systems to reduce costs, extend their lifespan, and facilitate sustainable development. In this paper, we systematically define the main sources of new knowledge in FL, including new features, tasks, models, and algorithms. For each source, we thoroughly analyze and discuss how to incorporate new knowledge into existing FL systems and examine the impact of the form and timing of new knowledge arrival on the incorporation process. Furthermore, we comprehensively discuss the potential future directions for FL with new knowledge, considering a variety of factors such as scenario setups, efficiency, and security. There is also a continuously updating repository for this topic: https://github.com/conditionWang/FLNK.

Recent studies have shown that as Transformer-based language models become larger and are trained on very large amounts of data, the fit of their surprisal estimates to naturalistic human reading times degrades. The current work presents a series of analyses showing that word frequency is a key explanatory factor underlying these two trends. First, residual errors from four language model families on four corpora show that the inverse correlation between model size and fit to reading times is the strongest on the subset of least frequent words, which is driven by excessively accurate predictions of larger model variants. Additionally, training dynamics reveal that during later training steps, all model variants learn to predict rare words and that larger model variants do so more accurately, which explains the detrimental effect of both training data amount and model size on fit to reading times. Finally, a feature attribution analysis demonstrates that larger model variants are able to accurately predict rare words based on both an effectively longer context window size as well as stronger local associations compared to smaller model variants. Taken together, these results indicate that Transformer-based language models' surprisal estimates diverge from human-like expectations due to the superhumanly complex associations they learn for predicting rare words.

Recently, large language models (LLMs) have shown remarkable capabilities including understanding context, engaging in logical reasoning, and generating responses. However, this is achieved at the expense of stringent computational and memory requirements, hindering their ability to effectively support long input sequences. This survey provides an inclusive review of the recent techniques and methods devised to extend the sequence length in LLMs, thereby enhancing their capacity for long-context understanding. In particular, we review and categorize a wide range of techniques including architectural modifications, such as modified positional encoding and altered attention mechanisms, which are designed to enhance the processing of longer sequences while avoiding a proportional increase in computational requirements. The diverse methodologies investigated in this study can be leveraged across different phases of LLMs, i.e., training, fine-tuning and inference. This enables LLMs to efficiently process extended sequences. The limitations of the current methodologies is discussed in the last section along with the suggestions for future research directions, underscoring the importance of sequence length in the continued advancement of LLMs.

Apart from what (little) OpenAI may be concealing from us, we all know (roughly) how ChatGPT works (its huge text database, its statistics, its vector representations, and their huge number of parameters, its next-word training, and so on). But none of us can say (hand on heart) that we are not surprised by what ChatGPT has proved to be able to do with these resources. This has even driven some of us to conclude that ChatGPT actually understands. It is not true that it understands. But it is also not true that we understand how it can do what it can do. I will suggest some hunches about benign biases: convergent constraints that emerge at LLM scale that may be helping ChatGPT do so much better than we would have expected. These biases are inherent in the nature of language itself, at LLM scale, and they are closely linked to what it is that ChatGPT lacks, which is direct sensorimotor grounding to connect its words to their referents and its propositions to their meanings. These convergent biases are related to (1) the parasitism of indirect verbal grounding on direct sensorimotor grounding, (2) the circularity of verbal definition, (3) the mirroring of language production and comprehension, (4) iconicity in propositions at LLM scale, (5) computational counterparts of human categorical perception in category learning by neural nets, and perhaps also (6) a conjecture by Chomsky about the laws of thought. The exposition will be in the form of a dialogue with ChatGPT-4.

Current vision large language models (VLLMs) exhibit remarkable capabilities yet are prone to generate harmful content and are vulnerable to even the simplest jailbreaking attacks. Our initial analysis finds that this is due to the presence of harmful data during vision-language instruction fine-tuning, and that VLLM fine-tuning can cause forgetting of safety alignment previously learned by the underpinning LLM. To address this issue, we first curate a vision-language safe instruction-following dataset VLGuard covering various harmful categories. Our experiments demonstrate that integrating this dataset into standard vision-language fine-tuning or utilizing it for post-hoc fine-tuning effectively safety aligns VLLMs. This alignment is achieved with minimal impact on, or even enhancement of, the models' helpfulness. The versatility of our safety fine-tuning dataset makes it a valuable resource for safety-testing existing VLLMs, training new models or safeguarding pre-trained VLLMs. Empirical results demonstrate that fine-tuned VLLMs effectively reject unsafe instructions and substantially reduce the success rates of several black-box adversarial attacks, which approach zero in many cases. The code and dataset are available at https://github.com/ys-zong/VLGuard.

In today's media landscape, where news outlets play a pivotal role in shaping public opinion, it is imperative to address the issue of sentiment manipulation within news text. News writers often inject their own biases and emotional language, which can distort the objectivity of reporting. This paper introduces a novel approach to tackle this problem by reducing the polarity of latent sentiments in news content. Drawing inspiration from adversarial attack-based sentence perturbation techniques and a prompt based method using ChatGPT, we employ transformation constraints to modify sentences while preserving their core semantics. Using three perturbation methods: replacement, insertion, and deletion coupled with a context-aware masked language model, we aim to maximize the desired sentiment score for targeted news aspects through a beam search algorithm. Our experiments and human evaluations demonstrate the effectiveness of these two models in achieving reduced sentiment polarity with minimal modifications while maintaining textual similarity, fluency, and grammatical correctness. Comparative analysis confirms the competitive performance of the adversarial attack based perturbation methods and prompt-based methods, offering a promising solution to foster more objective news reporting and combat emotional language bias in the media.

Despite the widespread adoption, there is a lack of research into how various critical aspects of pretrained language models (PLMs) affect their performance in hate speech detection. Through five research questions, our findings and recommendations lay the groundwork for empirically investigating different aspects of PLMs' use in hate speech detection. We deep dive into comparing different pretrained models, evaluating their seed robustness, finetuning settings, and the impact of pretraining data collection time. Our analysis reveals early peaks for downstream tasks during pretraining, the limited benefit of employing a more recent pretraining corpus, and the significance of specific layers during finetuning. We further call into question the use of domain-specific models and highlight the need for dynamic datasets for benchmarking hate speech detection.