The diffusion of increasingly capable AI systems has produced concern that bad actors could intentionally misuse current or future AI systems for harm. Governments have begun to create new entities—such as AI Safety Institutes—tasked with assessing these risks. However, approaches for risk assessment are currently fragmented and would benefit from broader disciplinary expertise. As it stands, it is often unclear whether concerns about malicious use misestimate the likelihood and severity of the risks. This article advances a conceptual framework to review and structure investigation into the likelihood of an AI system (X) being applied to a malicious use (Y). We introduce a three-stage framework of (1) Plausibility (can X be used to do Y at all?), (2) Performance (how well does X do Y?), and (3) Observed use (do actors use X to do Y in practice?). At each stage, we outline key research questions, methodologies, benefits and limitations, and the types of uncertainty addressed. We also offer ideas for directions to improve risk assessment moving forward.

State-of-the-art computer vision systems are trained to predict a fixed set of predetermined object categories. This restricted form of supervision limits their generality and usability since additional labeled data is needed to specify any other visual concept. Learning directly from raw text about images is a promising alternative which leverages a much broader source of supervision. We demonstrate that the simple pre-training task of predicting which caption goes with which image is an efficient and scalable way to learn SOTA image representations from scratch on a dataset of 400 million (image, text) pairs collected from the internet. After pre-training, natural language is used to reference learned visual concepts (or describe new ones) enabling zero-shot transfer of the model to downstream tasks. We study the performance of this approach by benchmarking on over 30 different existing computer vision datasets, spanning tasks such as OCR, action recognition in videos, geo-localization, and many types of fine-grained object classification. The model transfers non-trivially to most tasks and is often competitive with a fully supervised baseline without the need for any dataset specific training. For instance, we match the accuracy of the original ResNet-50 on ImageNet zero-shot without needing to use any of the 1.28 million training examples it was trained on.

We demonstrate that scaling up language models greatly improves task-agnostic, few-shot performance, sometimes even becoming competitive with prior state-of-the-art fine-tuning approaches. Specifically, we train GPT-3, an autoregressive language model with 175 billion parameters, 10x more than any previous non-sparse language model, and test its performance in the few-shot setting. For all tasks, GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model. GPT-3 achieves strong performance on many NLP datasets, including translation, question-answering, and cloze tasks. We also identify some datasets where GPT-3's few-shot learning still struggles, as well as some datasets where GPT-3 faces methodological issues related to training on large web corpora.