Cantina Labs is a social AI product that combines group chat, voice, and video capabilities together with AI character creation. With our advanced AI bot creator tools, users can design digital versions of themselves, bring imaginary characters to life, or choose from thousands of existing AI bots in our user-generated bot directory. Cantina bots are lifelike, social creatures, capable of interacting via text, voice, images in Cantina rooms or anywhere across the internet as they can be shared off platform. We have made Cantina invite-only for now to enable the development and implementation of strategic safety systems and core functionality ahead of opening up the platform to the wider public. By keeping the user base small, we are able to refine our safety policies, features and prevention systems.
We are focused on the following initiatives to uphold the Inspired Internet Pledge:
- Contribute to and/or actively engage in third-party research related to the Pledge principles
- Establish standards or policies focused on users’ mental health and wellbeing
- Embed users’ mental health into success metrics within teams/departments
Dec 2024: Cantina Labs joined the Inspired Internet Pledge as an Signatory.
We have joined the Inspired Internet Pledge so that we can be held accountable for our Trust & Safety product, policy and operations work, but also, as a means to place mental health at the core of our efforts.
We acknowledge the potential for misuse of our social spaces and generative AI tools. To ensure a safe experience for all users, we have implemented a number of measures and precautions over the last few months. For instance, we have reporting tools across the platform to enable users to flag content and interactions that are concerning to them.
For abuse detection, we have adopted a hybrid approach, combining third-party classifiers by leading experts like Thorn and Hive with in-house word and phrase blocklists in multiple languages, including English, Spanish, Portuguese, Russian, French, Chinese (Simplified and Traditional), Hindi, Arabic, German, Japanese, and Turkish. These detection mechanisms operate across text exchanges in rooms, DMs, bot creation prompts, bot engagement, and image uploads.
We have also joined the Tech Coalition and the Digital Trust & Safety Partnership to ensure that we are learning from our peers and sharing our experiences based on our AI technology and users adoption of it.
We are particularly concerned about the impact of our technology on mental health and we are currently designing a partnership with Stanford’s Lab for Mental Health Innovation to red team self-harm and suicidal exchanges with bots and to design appropriate interventions.