Given the broad capabilities of large language models, it should be possible to work towards a general-purpose, text-based assistant that is aligned with human values, meaning that it is helpful, honest, and harmless. As an initial foray in this direction we study simple baseline techniques and evaluations, such as prompting. We find that the benefits from modest interventions increase with model size, generalize to a variety of alignment evaluations, and do not compromise the performance of large models. Next we investigate scaling trends for several training objectives relevant to alignment, comparing imitation learning, binary discrimination, and ranked preference modeling. We find that ranked preference modeling performs much better than imitation learning, and often scales more favorably with model size. In contrast, binary discrimination typically performs and scales very similarly to imitation learning. Finally we study a 'preference model pre-training' stage of training, with the goal of improving sample efficiency when finetuning on human preferences.

No abstract

No abstract

The Astropy Project supports and fosters the development of open-source and openly developed Python packages that provide commonly needed functionality to the astronomical community. A key element of the Astropy Project is the core package astropy, which serves as the foundation for more specialized projects and packages. In this article, we summarize key features in the core package as of the recent major release, version 5.0, and provide major updates on the Project. We then discuss supporting a broader ecosystem of interoperable packages, including connections with several astronomical observatories and missions. We also revisit the future outlook of the Astropy Project and the current status of Learn Astropy. We conclude by raising and discussing the current and future challenges facing the Project.

Property-based testing is a style of testing popularised by the QuickCheck family of libraries, first in Haskell (Claessen & Hughes, 2000) and later in Erlang (Arts, Hughes, Johansson, & Wiger, 2006), which integrates generated test cases into existing software testing workflows: Instead of tests that provide examples of a single concrete behaviour, tests specify properties that hold for a wide range of inputs, which the testing library then attempts to generate test cases to refute. For a general introduction to property-based testing, see (MacIver, 2019).Hypothesis is a mature and widely used property-based testing library for Python. It has over 100,000 downloads per week 1 , thousands of open source projects use it 2 , and in 2018 more than 4% of Python users surveyed by the PSF reported using it 3 . It will be of interest both to researchers using Python for developing scientific software, and to software testing researchers as a platform for research in its own right.