Nvidia is the world's leading manufacturer of graphics processing units In the AI gold rush, chip-makers are selling shovels – and they are in short supply. As the latest generation of artificial intelligence models like ChatGPT look set to transform our lives, the hardware that makes it all possible is becoming a strategic asset, with countries, companies and researchers all scrabbling to get hold of supplies. With talk of shortages lasting until at least next year, some would-be AI developers face being left behind.

Pre-trained language models (PLMs) have become a prevalent technique in deep learning for code, utilizing a two-stage pre-training and fine-tuning procedure to acquire general knowledge about code and specialize in a variety of downstream tasks. However, the dynamic nature of software codebases poses a challenge to the effectiveness and robustness of PLMs. In particular, world-realistic scenarios potentially lead to significant differences between the distribution of the pre-training and test data, i.e., distribution shift, resulting in a degradation of the PLM's performance on downstream tasks. In this paper, we stress the need for adapting PLMs of code to software data whose distribution changes over time, a crucial problem that has been overlooked in previous works. The motivation of this work is to consider the PLM in a non-stationary environment, where fine-tuning data evolves over time according to a software evolution scenario. Specifically, we design a scenario where the model needs to learn from a stream of programs containing new, unseen APIs over time. We study two widely used PLM architectures, i.e., a GPT2 decoder and a RoBERTa encoder, on two downstream tasks, API call and API usage prediction. We demonstrate that the most commonly used fine-tuning technique from prior work is not robust enough to handle the dynamic nature of APIs, leading to the loss of previously acquired knowledge i.e., catastrophic forgetting. To address these issues, we implement five continual learning approaches, including replay-based and regularization-based methods. Our findings demonstrate that utilizing these straightforward methods effectively mitigates catastrophic forgetting in PLMs across both downstream tasks while achieving comparable or superior performance.

Error prediction in large language models often relies on domain-specific information. In this paper, we present measures for quantification of error in the response of a large language model based on the diversity of responses to a given prompt - hence independent of the underlying application. We describe how three such measures - based on entropy, Gini impurity, and centroid distance - can be employed. We perform a suite of experiments on multiple datasets and temperature settings to demonstrate that these measures strongly correlate with the probability of failure. Additionally, we present empirical results demonstrating how these measures can be applied to few-shot prompting, chain-of-thought reasoning, and error detection.

The volume, variety, and velocity of change in vulnerabilities and exploits have made incident threat analysis challenging with human expertise and experience along. Tactics, Techniques, and Procedures (TTPs) are to describe how and why attackers exploit vulnerabilities. However, a TTP description written by one security professional can be interpreted very differently by another, leading to confusion in cybersecurity operations or even business, policy, and legal decisions. Meanwhile, advancements in AI have led to the increasing use of Natural Language Processing (NLP) algorithms to assist the various tasks in cyber operations. With the rise of Large Language Models (LLMs), NLP tasks have significantly improved because of the LLM's semantic understanding and scalability. This leads us to question how well LLMs can interpret TTPs or general cyberattack descriptions to inform analysts of the intended purposes of cyberattacks. We propose to analyze and compare the direct use of LLMs (e.g., GPT-3.5) versus supervised fine-tuning (SFT) of small-scale-LLMs (e.g., BERT) to study their capabilities in predicting ATT&CK tactics. Our results reveal that the small-scale-LLMs with SFT provide a more focused and clearer differentiation between the ATT&CK tactics (if such differentiation exists). On the other hand, direct use of LLMs offer a broader interpretation of cyberattack techniques. When treating more general cases, despite the power of LLMs, inherent ambiguity exists and limits their predictive power. We then summarize the challenges and recommend research directions on LLMs to treat the inherent ambiguity of TTP descriptions used in various cyber operations.

The strategy of training the model from scratch in a specific language or domain serves two essential purposes: i) enhancing performance in the particular linguistic or domain context, and ii) ensuring effective tokenization. The main limitation inherent to this approach lies in the associated cost, which can reach six to seven-digit dollar values, depending on the model size and the number of parameters involved. The main solution to overcome the cost challenge is to rely on available pre-trained models, which, despite recent advancements such as the LLaMA and LLaMA-2 models, still demonstrate inefficiency for certain specific domain problems or prove ineffective in scenarios involving conversational memory resources, given the large number of tokens required to represent text. To overcome this issue, we present a methodology named Cabrita, which, as our research demonstrates, successfully addresses the performance and efficient tokenization problem, all at an affordable cost. We believe that this methodology can be applied to any transformer-like architecture model. To validate the study, we conducted continuous pre-training exclusively using Portuguese text on a 3-billion-parameter model known as OpenLLaMA, resulting in a model named openCabrita 3B. The openCabrita 3B also features a new tokenizer that results in a significant reduction in the number of tokens required to represent the text. In our assessment, for few-shot learning tasks, we achieved similar results with this 3B model compared to a traditional continuous pre-training approach as well as to 7B models English pre-trained models.

Large language models such as Open AI's Generative Pre-trained Transformer (GPT) models are proficient at answering questions, but their knowledge is confined to the information present in their training data. This limitation renders them ineffective when confronted with questions about recent developments or non-public documents. Our research proposes a method that enables GPT models to answer questions by employing context from an information source not previously included in their training data. The methodology includes preprocessing of contextual information, the embedding of contexts and queries, constructing prompt through the integration of context embeddings, and generating answers using GPT models. We applied this method in a controlled test scenario using the California Driver's Handbook as the information source. The GPT-3 model achieved a 96% passing score on a set of 50 sample driving knowledge test questions. In contrast, without context, the model's passing score fell to 82%. However, the model still fails to answer some questions correctly even with providing library of context, highlighting room for improvement. The research also examined the impact of prompt length and context format, on the model's performance. Overall, the study provides insights into the limitations and potential improvements for GPT models in question-answering tasks.

Large Language Models (LLMs) have demonstrated impressive capabilities for text rewriting. Nonetheless, the large sizes of these models make them impractical for on-device inference, which would otherwise allow for enhanced privacy and economical inference. Creating a smaller yet potent language model for text rewriting presents a formidable challenge because it requires balancing the need for a small size with the need to retain the emergent capabilities of the LLM, that requires costly data collection. To address the above challenge, we introduce a new instruction tuning approach for building a mobile-centric text rewriting model. Our strategies enable the generation of high quality training data without any human labeling. In addition, we propose a heuristic reinforcement learning framework which substantially enhances performance without requiring preference data. To further bridge the performance gap with the larger server-side model, we propose an effective approach that combines the mobile rewrite agent with the server model using a cascade. To tailor the text rewriting tasks to mobile scenarios, we introduce MessageRewriteEval, a benchmark that focuses on text rewriting for messages through natural language instructions. Through empirical experiments, we demonstrate that our on-device model surpasses the current state-of-the-art LLMs in text rewriting while maintaining a significantly reduced model size. Notably, we show that our proposed cascading approach improves model performance.

This paper illustrates how generative AI could give opportunities for big productivity gains but also opens up questions about the impact of these new powerful technologies on the way we work and share knowledge. More specifically, we explore how ChatGPT changed a fundamental aspect of coding: problem-solving. To do so, we exploit the effect of the sudden release of ChatGPT on the 30th of November 2022 on the usage of the largest online community for coders: Stack Overflow. Using quasi-experimental methods (Difference-in-Difference), we find a significant drop in the number of questions. In addition, the questions are better documented after the release of ChatGPT. Finally, we find evidence that the remaining questions are more complex. These findings suggest not only productivity gains but also a fundamental change in the way we work where routine inquiries are solved by AI allowing humans to focus on more complex tasks.

This paper outlines a natural conversational approach to solving personalized energy-related problems using large language models (LLMs). We focus on customizable optimization problems that necessitate repeated solving with slight variations in modeling and are user-specific, hence posing a challenge to devising a one-size-fits-all model. We put forward a strategy that augments an LLM with an optimization solver, enhancing its proficiency in understanding and responding to user specifications and preferences while providing nonlinear reasoning capabilities. Our approach pioneers the novel concept of human-guided optimization autoformalism, translating a natural language task specification automatically into an optimization instance. This enables LLMs to analyze, explain, and tackle a variety of instance-specific energy-related problems, pushing beyond the limits of current prompt-based techniques. Our research encompasses various commonplace tasks in the energy sector, from electric vehicle charging and Heating, Ventilation, and Air Conditioning (HVAC) control to long-term planning problems such as cost-benefit evaluations for installing rooftop solar photovoltaics (PVs) or heat pumps. This pilot study marks an essential stride towards the context-based formulation of optimization using LLMs, with the potential to democratize optimization processes. As a result, stakeholders are empowered to optimize their energy consumption, promoting sustainable energy practices customized to personal needs and preferences.

We propose MemoChat, a pipeline for refining instructions that enables large language models (LLMs) to effectively employ self-composed memos for maintaining consistent long-range open-domain conversations. We demonstrate a long-range open-domain conversation through iterative "memorization-retrieval-response" cycles. This requires us to carefully design tailored tuning instructions for each distinct stage. The instructions are reconstructed from a collection of public datasets to teach the LLMs to memorize and retrieve past dialogues with structured memos, leading to enhanced consistency when participating in future conversations. We invite experts to manually annotate a test set designed to evaluate the consistency of long-range conversations questions. Experiments on three testing scenarios involving both open-source and API-accessible chatbots at scale verify the efficacy of MemoChat, which outperforms strong baselines. Our codes, data and models are available here: https://github.com/LuJunru/MemoChat.