Opera's Aria AI gets local image understanding with 2000+ LLMs for enhanced privacy

A couple of weeks ago Opera introduced Image Understanding to its AI assistant, Aria, and now the company is enhancing its capabilities with local large language models (LLMs) support. Part of Opera Developer’s latest Feature Drop, this update includes over 2000 local LLMs, enabling users to make image queries using models from the Llava and bakllava families. These multi-modal models can process both text and images, offering more comprehensive AI functionalities directly on users' devices.

This local AI feature aims to improve privacy and user control by eliminating the need for cloud servers. To use this feature, users must download Opera Developer, access the Aria icon in the sidebar, select 'choose local AI model,' and download a model through the settings. After downloading, users can initiate a new chat and upload images via a ‘+’ button next to the chat input box.

While local LLMs provide enhanced privacy, they are not without challenges, such as hallucinations where the AI might offer incorrect answers. For example, it might confuse an Apple Vision Pro headset with an Oculus Rift or misidentify the number of dogs in an image. Larger models can reduce these inaccuracies but often require powerful hardware, leading to the need for cloud computing. Opera has also introduced a device Benchmarking tool to help users determine if their hardware can support these larger models.