A multimodal model works with several types of data. This expands how we interact with it, but does not eliminate errors.
What happened
OpenAI described GPT-4o as a model trained jointly on text, images and audio. The announcement demonstrated voice and visual interactions.
Why it matters
Different inputs provide more context: an image can complement a question, while a voice can convey intonation. What an application can do still depends on the features available to it.
The bigger picture
Multimodality does not mean flawless perception. Responses involving images and audio still need to be checked.
Archive article. Product capabilities may have changed since the announcement.