My AI is an experimental, friendly, chatbot currently available to Snapchatters.
Our Approach to AI Safety
We care deeply about the safety and wellbeing of our community and work to reflect those values in our products. My AI is powered by Large Language Models like GPT3.5 and GPT4 from OpenAI and built on its technology, and it has been trained on a diverse range of texts. While My AI was programmed to abide by certain guidelines so the information it provides is not harmful (including avoiding responses that are violent, hateful, sexually explicit or otherwise dangerous; and avoiding perpetuating harmful biases), it may not always be successful.
We integrate the same safeguards and tools we use across Snapchat. Blocking results for drug keywords and slang is also part of how we've customized My AI. In addition, if Snapchatters search for a range of mental health issues anywhere on Snapchat, they will be served resources from expert organizations. This tool, called Here For You, will show up in chats with My AI.
We're also working on adding additional tools to our Family Centre around My AI that would give parents more visibility and control around their teen's usage of My AI.
Feedback
Like other types of artificial intelligence and machine learning technologies, AI chatbots improve the more they are used. We encourage Snapchatters to submit feedback to us, especially if something goes wrong, so we can work to evolve the technology along with OpenAI.
Snapchatters can easily send feedback to our ram by long pressing on any response from My AI to share more on what they'd like to see more or less of while we continue to train My Ai.
To report a Chat received from My AI…
- Press and hold on the Chat response and tap 'Report'
- Let us know why you’re reporting the Chat
- Tap 'Submit'
Misinformation & Divisive Topics
My AI is powered by Large Language Models like GPT3.5 and GPT4 from OpenAI, which have been trained on a large and diverse text corpus, including various types of text such as books, articles and web pages. This corpus reflected the diversity of information available on the internet, including misinformation and divisive topics.
The training process was designed to avoid amplifying harmful or inaccurate information, and the model was also fine-tuned to reduce biases in language and to prioritize factual information — though it may not always be successful. Snap has also added additional attributes and safety features unique to Snapchat.
My AI's Limitations
We’re constantly working to improve and evolve My AI, but it’s possible My AI’s responses may include biased, incorrect, harmful, or misleading content. Because My AI is an evolving feature, you should always independently check answers provided by My AI before relying on any advice, and you should not share confidential or sensitive information. Unlike other chats with friends, all messages with My AI will be retained unless you delete them and may be used to improve Snap’s products and to personalize your experience, including ads.
My AI's responses are sometimes limited to the latest information in the underlying models' training data. This means that My AI's responses might not be current on recent news and events. My AI may also avoid certain sensitive subjects as appropriate.