Large language models have been shown to achieve remarkable performance across a variety of natural language tasks using few-shot learning, which drastically reduces the number of task-specific training examples needed to adapt the model to a particular application. To further our understanding of the impact of scale on few-shot learning, we trained a 540-billion parameter, densely activated, Transformer language model, which we call Pathways Language Model (PaLM).We trained PaLM on 6144 TPU v4 chips using Pathways, a new ML system which enables highly efficient training across multiple TPU Pods. We demonstrate continued benefits of scaling by achieving state-ofthe-art few-shot learning results on hundreds of language understanding and generation benchmarks. On a number of these tasks, PaLM 540B achieves breakthrough performance, outperforming the finetuned stateof-the-art on a suite of multi-step reasoning tasks, and outperforming average human performance on the recently released BIG-bench benchmark. A significant number of BIG-bench tasks showed discontinuous improvements from model scale, meaning that performance steeply increased as we scaled to our largest model. PaLM also has strong capabilities in multilingual tasks and source code generation, which we demonstrate on a wide array of benchmarks. We additionally provide a comprehensive analysis on bias and toxicity, and study the extent of training data memorization with respect to model scale. Finally, we discuss the ethical considerations related to large language models and discuss potential mitigation strategies. * Equal Contribution. Author contributions and ordering details are listed in Appendix A.

Research at the intersection of machine learning, programming languages, and software engineering has recently taken important steps in proposing learnable probabilistic models of source code that exploit code's abundance of patterns. In this article, we survey this work. We contrast programming languages against natural languages and discuss how these similarities and differences drive the design of probabilistic models. We present a taxonomy based on the underlying design principles of each model and use it to navigate the literature. Then, we review how researchers have adapted these models to application areas and discuss crosscutting and application-specific challenges and opportunities.1 It may be worth pointing out that deep learning and probabilistic modeling are not mutually exclusive. Indeed, many of the currently most effective methods for language modeling, for example, are based on deep learning. of probabilistic source code models (Section 5). Finally, we mention a few overlapping research areas (Section 7), and we discuss challenges and interesting future directions (Section 6).Related Reviews and other Resources. There have been short reviews summarizing the progress and the vision of the research area, from both software engineering [52] and programming languages perspectives [28,195]. However, none of these articles can be considered extensive literature reviews, which is the purpose of this work. Ernst [57] discusses promising areas of applying natural language processing to software development, including error messages, variable names, code comments, and user questions. Some resources, datasets and code can be found at http://learnbigcode.github.io/. An online version of the work reviewed here -which we will keep up-to-date by accepting external contributions -can be found at https://ml4code.github.io. THE NATURALNESS HYPOTHESISMany aspects of code, such as names, formatting, the lexical order of methods, etc. have no impact on program semantics. This is precisely why we abstract them in most program analyses. But then, why should statistical properties of code matter at all? To explain this, we recently suggested a hypothesis, called the naturalness hypothesis. The inspiration for the naturalness hypothesis can be traced back to the "literate programming" concept of D. Knuth, which draws from the insight that programming is a form of human communication: "Let us change our traditional attitude to the construction of programs: Instead of imagining that our main task is to instruct a computer what to do, let us concentrate rather on explaining to human beings what we want a computer to do... " [105] The naturalness hypothesis, then, holds thatThe naturalness hypothesis. Software is a form of human communication; software corpora have similar statistical properties to natural language corpora; and these properties can be exploited to build better software engineering tools.The exploitation of the statistics of human communication is a mature and effective technology, with numerous applications ...

Often we wish to predict a large number of variables that depend on each other as well as on other observed variables. Structured prediction methods are essentially a combination of classification and graphical modeling, combining the ability of graphical models to compactly model multivariate data with the ability of classification methods to perform prediction using large sets of input features. This tutorial describes conditional random fields, a popular probabilistic method for structured prediction. CRFs have seen wide application in natural language processing, computer vision, and bioinformatics. We describe methods for inference and parameter estimation for CRFs, including practical issues for implementing large scale CRFs. We do not assume previous knowledge of graphical modeling, so this tutorial is intended to be useful to practitioners in a wide variety of fields.of CRF training on some benchmark problems (Section 4.5).Since this is the first of our sections on implementation details, it seems appropriate to mention some of the available implementations of CRFs. At the time of writing, a few popular implementations are: CRF++http://crfpp.sourceforge.net/ MALLET