Option to limit default models & HuggingFaces search (for Corp)

[JonVojtush]

I was interested in proposing GPT4All as an AI Chat solution in the corporation that I work for. If theoretically approved, they could provide software to employee laptops. Although there are certain models that pose risk (when considering PHI, proprietary info, & ethics) such as the following:

• 3rd party API required
• Hermes (Uncensored)
• HuggingFaces Search - expands potential risk

The work may not be worth the investment, but it may also open the door to enterprise licensing. Just an idea.

[manyoso]

Yes, it would be fairly easy to turn these things off. Presumably if a corporation wanted to deploy internally and wanted to turn these things off ... i wonder if they'd want a change to the source or do they want an 'option' in what sense? In a settings file? I imagine a build time option would be best don't you think?

[JonVojtush]

II like the settings file idea, to give some flexibility on specific models & features to disable. My mind went straight to a -secure parameter when running the .exe because I'm a noob at desktop programming with Go :P

[manyoso]

The settings file is problematic because it would be very easy for people to change the setting. If a corp really wanted to make it secure and disable these things the best thing would be build time settings. We could wrap the features in C pre-processor defines... but we'd need a list of the specific features that a corp would want to disable and then assess how complicated this would make the code. Can you share what corp this is?

[JonVojtush]

Here are my recommendations based on our data & ethical requirements.

• Access to models that require a 3rd party api (to prevent risk of unauthorized data access)
• Access to models trained by Uncensored data (to prevent potentially unethical response)
• Access to the new Hugging Faces search tool (to prevent risk of downloading models with potentially malicious code or was trained on uncensored data)