At Anthropic, we believe the most impactful safety research will require access to frontier AI systems. The most powerful AIs will operate not just on text but also other modes of data, including images, video and audio. Such models have potential to augment human creativity and productivity in exciting ways. However, we are very concerned about the risks introduced by powerful multimodal AIs. The Multimodal team at Anthropic builds and studies multimodal models to better understand and mitigate these risks. Our team works across many parts of a large stack that includes training, inference, system design and data collection. Some of our core focus areas are: Foundational Research We develop new architectures for modeling multimodal data and study how they interact with text-only models at scale. Building Infrastructure We work on many infrastructure projects including: * Complex multimodal reinforcement learning environments. * High-performance RPC servers for processing image inputs. * Sandboxing infrastructure for securely collecting data. Data Ingestion We are more interested in running simple experiments at large scale than smaller complex experiments. This requires access to very large sources of multimodal data. We develop tooling to collect, process and clean multimodal data at scale. Because we focus on so many areas, the team is looking to work with both experienced engineers and strong researchers, and encourage anyone along the researcher/engineer curve to apply. You may be a good fit if you: * Have significant software engineering experience * Are results-oriented, with a bias towards flexibility and impact * Pick up slack, even if it goes outside your job description * Enjoy pair programming (we love to pair!) * Want to learn more about machine learning research * Care about the societal impacts of your work Strong candidates may also have experience with: * High performance, large-scale ML systems * GPUs, Kubernetes, Pytorch, or OS internals * Language modeling with transformers * Reinforcement learning * Large-scale ETL