Mozilla introduced experimental MemoryCache extension, implementing a machine learning dialogue system capable of taking into account the content the user interacts with in the browser. Unlike other AI chats, MemoryCache allows personalizing user interaction and utilizing specific user data in generating responses to questions. The project code is distributed is licensed under MPL. Installation in Firefox currently is supported is only available manually through the 'about:debugging' mode and requires applying a patch.
All components of MemoryCache run on the user's device and do not access external resources. The model GPT4All-J, trained on English-language collections Pile (825 GB of data), ShareGPT (13 MB) and Dolly (660 MB), is used as the knowledge base in MemoryCache. The model is available in versions with 3, 7, and 13 billion parameters. To interact with the model, the toolkit privateGPT, optimized for working with large language models on local computers.
Memory Cache saves viewed pages in the directory '~\/Downloads\/MemoryCache'. A periodic script extracts pages from this directory and passes them to the privateGPT handler (ingest.py) for updating the current language model. Thus, the model adapts to the user's interests and can be used to discuss viewed content in a text chat with the bot. The user can naturally ask questions to the bot about the saved pages, for example, to clarify facts, and the bot will generate responses using the local model, without accessing external services.
Source: linux.org.ru
