We present a method for systematically evaluating the correctness and robustness of instruction-tuned large language models (LLMs) for code generation via a new benchmark, Turbulence. Turbulence consists of a large set of natural language $\textit{question templates}$, each of which is a programming problem, parameterised so that it can be asked in many different forms. Each question template has an associated $\textit{test oracle}$ that judges whether a code solution returned by an LLM is correct. Thus, from a single question template, it is possible to ask an LLM a $\textit{neighbourhood}$ of very similar programming questions, and assess the correctness of the result returned for each question. This allows gaps in an LLM's code generation abilities to be identified, including $\textit{anomalies}$ where the LLM correctly solves $\textit{almost all}$ questions in a neighbourhood but fails for particular parameter instantiations. We present experiments against five LLMs from OpenAI, Cohere and Meta, each at two temperature configurations. Our findings show that, across the board, Turbulence is able to reveal gaps in LLM reasoning ability. This goes beyond merely highlighting that LLMs sometimes produce wrong code (which is no surprise): by systematically identifying cases where LLMs are able to solve some problems in a neighbourhood but do not manage to generalise to solve the whole neighbourhood, our method is effective at highlighting $\textit{robustness}$ issues. We present data and examples that shed light on the kinds of mistakes that LLMs make when they return incorrect code results.

In this work, we aim to establish a strong connection between two significant bodies of machine learning research: continual learning and sequence modeling. That is, we propose to formulate continual learning as a sequence modeling problem, allowing advanced sequence models to be utilized for continual learning. Under this formulation, the continual learning process becomes the forward pass of a sequence model. By adopting the meta-continual learning (MCL) framework, we can train the sequence model at the meta-level, on multiple continual learning episodes. As a specific example of our new formulation, we demonstrate the application of Transformers and their efficient variants as MCL methods. Our experiments on seven benchmarks, covering both classification and regression, show that sequence models can be an attractive solution for general MCL.

While AlphaZero-style reinforcement learning (RL) algorithms excel in various board games, in this paper we show that they face challenges on impartial games where players share pieces. We present a concrete example of a game - namely the children's game of Nim - and other impartial games that seem to be a stumbling block for AlphaZero-style and similar self-play reinforcement learning algorithms. Our work is built on the challenges posed by the intricacies of data distribution on the ability of neural networks to learn parity functions, exacerbated by the noisy labels issue. Our findings are consistent with recent studies showing that AlphaZero-style algorithms are vulnerable to adversarial attacks and adversarial perturbations, showing the difficulty of learning to master the games in all legal states. We show that Nim can be learned on small boards, but the learning progress of AlphaZero-style algorithms dramatically slows down when the board size increases. Intuitively, the difference between impartial games like Nim and partisan games like Chess and Go can be explained by the fact that if a small part of the board is covered for impartial games it is typically not possible to predict whether the position is won or lost as there is often zero correlation between the visible part of a partly blanked-out position and its correct evaluation. This situation starkly contrasts partisan games where a partly blanked-out board position typically provides abundant or at least non-trifle information about the value of the fully uncovered position.

Finetuning large language models (LLMs) has been empirically effective on a variety of downstream tasks. Existing approaches to finetuning an LLM either focus on parameter-efficient finetuning, which only updates a small number of trainable parameters, or attempt to reduce the memory footprint during the training phase of the finetuning. Typically, the memory footprint during finetuning stems from three contributors: model weights, optimizer states, and intermediate activations. However, existing works still require considerable memory and none can simultaneously mitigate memory footprint for all three sources. In this paper, we present Quantized Side Tuing (QST), which enables memory-efficient and fast finetuning of LLMs by operating through a dual-stage process. First, QST quantizes an LLM's model weights into 4-bit to reduce the memory footprint of the LLM's original weights; QST also introduces a side network separated from the LLM, which utilizes the hidden states of the LLM to make task-specific predictions. Using a separate side network avoids performing backpropagation through the LLM, thus reducing the memory requirement of the intermediate activations. Furthermore, QST leverages several low-rank adaptors and gradient-free downsample modules to significantly reduce the trainable parameters, so as to save the memory footprint of the optimizer states. Experiments show that QST can reduce the total memory footprint by up to 2.3 $\times$ and speed up the finetuning process by up to 3 $\times$ while achieving competent performance compared with the state-of-the-art. When it comes to full finetuning, QST can reduce the total memory footprint up to 7 $\times$.

Recent efforts have witnessed remarkable progress in Satellite Video Super-Resolution (SVSR). However, most SVSR methods usually assume the degradation is fixed and known, e.g., bicubic downsampling, which makes them vulnerable in real-world scenes with multiple and unknown degradations. To alleviate this issue, blind SR has thus become a research hotspot. Nevertheless, existing approaches are mainly engaged in blur kernel estimation while losing sight of another critical aspect for VSR tasks: temporal compensation, especially compensating for blurry and smooth pixels with vital sharpness from severely degraded satellite videos. Therefore, this paper proposes a practical Blind SVSR algorithm (BSVSR) to explore more sharp cues by considering the pixel-wise blur levels in a coarse-to-fine manner. Specifically, we employed multi-scale deformable convolution to coarsely aggregate the temporal redundancy into adjacent frames by window-slid progressive fusion. Then the adjacent features are finely merged into mid-feature using deformable attention, which measures the blur levels of pixels and assigns more weights to the informative pixels, thus inspiring the representation of sharpness. Moreover, we devise a pyramid spatial transformation module to adjust the solution space of sharp mid-feature, resulting in flexible feature adaptation in multi-level domains. Quantitative and qualitative evaluations on both simulated and real-world satellite videos demonstrate that our BSVSR performs favorably against state-of-the-art non-blind and blind SR models. Code will be available at https://github.com/XY-boy/Blind-Satellite-VSR

Conversational agents have been gaining increasing popularity in recent years. Influenced by the widespread adoption of task-oriented agents such as Apple Siri and Amazon Alexa, these agents are being deployed into various applications to enhance user experience. Although these agents promote "ask me anything" functionality, they are typically built to focus on a single or finite set of expertise. Given that complex tasks often require more than one expertise, this results in the users needing to learn and adopt multiple agents. One approach to alleviate this is to abstract the orchestration of agents in the background. However, this removes the option of choice and flexibility, potentially harming the ability to complete tasks. In this paper, we explore these different interaction experiences (one agent for all) vs (user choice of agents) for conversational AI. We design prototypes for each, systematically evaluating their ability to facilitate task completion. Through a series of conducted user studies, we show that users have a significant preference for abstracting agent orchestration in both system usability and system performance. Additionally, we demonstrate that this mode of interaction is able to provide quality responses that are rated within 1% of human-selected answers.

In the rapidly evolving domain of Natural Language Generation (NLG) evaluation, introducing Large Language Models (LLMs) has opened new avenues for assessing generated content quality, e.g., coherence, creativity, and context relevance. This survey aims to provide a thorough overview of leveraging LLMs for NLG evaluation, a burgeoning area that lacks a systematic analysis. We propose a coherent taxonomy for organizing existing LLM-based evaluation metrics, offering a structured framework to understand and compare these methods. Our detailed exploration includes critically assessing various LLM-based methodologies, as well as comparing their strengths and limitations in evaluating NLG outputs. By discussing unresolved challenges, including bias, robustness, domain-specificity, and unified evaluation, this survey seeks to offer insights to researchers and advocate for fairer and more advanced NLG evaluation techniques.

Recently, many studies utilized adversarial examples (AEs) to raise the cost of malicious image editing and copyright violation powered by latent diffusion models (LDMs). Despite their successes, a few have studied the surrogate model they used to generate AEs. In this paper, from the perspective of adversarial transferability, we investigate how the surrogate model's property influences the performance of AEs for LDMs. Specifically, we view the time-step sampling in the Monte-Carlo-based (MC-based) adversarial attack as selecting surrogate models. We find that the smoothness of surrogate models at different time steps differs, and we substantially improve the performance of the MC-based AEs by selecting smoother surrogate models. In the light of the theoretical framework on adversarial transferability in image classification, we also conduct a theoretical analysis to explain why smooth surrogate models can also boost AEs for LDMs.

LLMs have demonstrated remarkable capability for understanding semantics, but they often struggle with understanding pragmatics. To demonstrate this fact, we release a Pragmatics Understanding Benchmark (PUB) dataset consisting of fourteen tasks in four pragmatics phenomena, namely, Implicature, Presupposition, Reference, and Deixis. We curated high-quality test sets for each task, consisting of Multiple Choice Question Answers (MCQA). PUB includes a total of 28k data points, 6.1k of which have been created by us, and the rest are adapted from existing datasets. We evaluated nine models varying in the number of parameters and type of training. Our study indicates that fine-tuning for instruction-following and chat significantly enhances the pragmatics capabilities of smaller language models. However, for larger models, the base versions perform comparably with their chat-adapted counterparts. Additionally, there is a noticeable performance gap between human capabilities and model capabilities. Furthermore, unlike the consistent performance of humans across various tasks, the models demonstrate variability in their proficiency, with performance levels fluctuating due to different hints and the complexities of tasks within the same dataset. Overall, the benchmark aims to provide a comprehensive evaluation of LLM's ability to handle real-world language tasks that require pragmatic reasoning.

Chain-of-thought (CoT) has emerged as a powerful technique to elicit reasoning in large language models and improve a variety of downstream tasks. CoT mainly demonstrates excellent performance in English, but its usage in low-resource languages is constrained due to poor language generalization. To bridge the gap among different languages, we propose a cross-lingual instruction fine-tuning framework (xCOT) to transfer knowledge from high-resource languages to low-resource languages. Specifically, the multilingual instruction training data (xCOT-INSTRUCT) is created to encourage the semantic alignment of multiple languages. We introduce cross-lingual in-context few-shot learning (xICL)) to accelerate multilingual agreement in instruction tuning, where some fragments of source languages in examples are randomly substituted by their counterpart translations of target languages. During multilingual instruction tuning, we adopt the randomly online CoT strategy to enhance the multilingual reasoning ability of the large language model by first translating the query to another language and then answering in English. To further facilitate the language transfer, we leverage the high-resource CoT to supervise the training of low-resource languages with cross-lingual distillation. Experimental results on previous benchmarks demonstrate the superior performance of xCoT in reducing the gap among different languages, highlighting its potential to reduce the cross-lingual gap.