Researcher, Multimodal Safety

Written by the T&S Network from OpenAI's posting.

Multimodal safety research for OpenAI's ChatGPT and related models across text, speech, and visual interactions. The Chat and Multimodal Safety team builds the research, training methods, and evaluations that keep multimodal models and products behaving safely, working with Personal AGI, Consumer Devices, and product and model teams.

What you'd do

• Define and advance multimodal safety research spanning text, vision, and audio.
• Link how models perceive images and sound to how they should respond safely.
• Build training methods for VLMs, including post-training and safety interventions.
• Design safety evaluations for vision-language models across varied contexts.
• Work with Personal AGI, Consumer Devices, and product teams to bring safety into always-on, built-in, and personal multimodal products.

What it asks for

• Track record building or advancing multimodal models, such as vision-language, video, image generation, audio, or multimodal reasoning.
• Understanding of multimodal systems end to end, including encoders, projection layers, modality fusion, and inference tradeoffs.
• Experience improving model behavior through post-training such as SFT, RL, data curation, synthetic data, and error analysis.
• Strong research and engineering judgment for open-ended safety problems, including experiment design and failure diagnosis.
• Experience with image, video, or audio safety systems (preferred).

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on OpenAInorth_east before you apply. If we got something wrong, report this listing.