The company still says the bot can’t be trusted for advice.
Snapchat’s ChatGPT-powered AI personality is expanding to all the app’s users. An upgraded version of “My AI,” the in-app chatbot that was previously limited to Snapchat+ subscribers, is now launching globally, Snap CEO Evan Spiegel announced at the company’s Partner Summit event.
With the expansion, My AI has a number of new Snapchat-specific features. It can provide Snapchat users with recommendations for restaurants and other activities based on what’s popular in the Snap Map, and can suggest augmented reality lenses. Users can also add the AI to group chats, and set a custom name and avatar (via Bitmoji) for the AI persona. My AI is also now able to respond to photo and video snaps. For now, it can only respond with text messages, but Snap says that “soon” it will be able to respond to snaps with its own AI-generated art. (That feature will, however, be exclusive to Snapchat+.)
While My AI’s rollout has been limited until now, the company says that My AI has already been a hit with users, who exchange 2 million messages a day with the bot. Snap’s implementation of OpenAI’s technology has also been criticized, though. The Washington Post reported that the bot engaged in inappropriate conversations and gave questionable advice to researchers posing as teens.
Now, Snap says that My AI uses “additional moderation technology” so that users’ can be temporarily restricted from the feature if they “misuse the service.” The company also says the AI has been trained to consider the age of the user it’s messaging with to help prevent chats from veering into unsafe territory. Additionally, parents will eventually be able to use the app’s parental control feature, Family Center, to track how much time their teens are spending chatting with their new AI friend.
Even so, Snap seems to be well-aware of the limitations of My AI, which it has previously described as “prone to hallucination.” Even Spiegel, who has suggested that interacting with AI like ChatGPT will soon be an everyday activity, was somewhat cautious in his endorsement “My AI certainly makes plenty of mistakes, so you can’t rely on it for advice, but it’s definitely entertaining,” he said.