The ‘My AI’ chatbot service, which uses OpenAI’s GPT technology and allows Snapchat+ subscribers to ask questions of the bot in the app and receive responses on any topic of their choosing, has received an update from Snapchat.
In its AI chatbot, Snapchat adds a few capabilities to improve safety. The business has put out a statement on some security improvements as a result of its learning and stated that it will introduce a few controls to control the AI replies.
An age-appropriate filter and parent-focused insights are among the new Snapchat technologies that will keep its recently released AI chatbot “My AI” experience safer.
The business claimed it realized people were attempting to “trick the chatbot into giving responses that do not comply to our requirements” after identifying some potential abuse scenarios for the AI chatbot.
The company has released an update on some safety advancements as a consequence of its learning and stated that it will introduce a few tools to control the AI reactions.
The business claimed that since introducing My AI, it has made a concerted effort to enhance its reactions to improper Snapchatter requests, irrespective of a Snapchatter’s age.
It searches My AI interactions for potentially non – conformance text using proactive detection technologies and takes appropriate action.
The business “developed a new age signal for My AI using a Snapchatter’s birthdate, so that even if a Snapchatter never informs My AI their age in a discussion, the chatbot would continuously take their age into mind while interacting with it.” according to the company. In the upcoming weeks, Snapchat will give parents more information about their adolescents’ contacts with My AI through the in-app Family Center.
As a result, parents will be able to check Family Center to discover if and how frequently their teens are interacting with My AI.
Which, at least for the most part, is a straightforward and enjoyable use of technology; but, Snap has discovered some alarming abuses of the tool and is now seeking to incorporate additional safeguards and precautions into the procedure.
based on Snap:
We were able to determine which guardrails are effective and which ones require strengthening by looking back on early encounters with My AI. ‘Non-conforming’ language, which we define as any message that provides links to violent action, graphic sexual terms, illegal drug use, sexual assault of children, bullying, hateful speech, derogatory or biassed declarations, racism, misogyny, or marginalizing marginalized minorities, has been reviewed in order to help with this assessment. On Snapchat, each of these content types is expressly forbidden.
A same open letter published in 2015 issued a similar warning about the potential of this kind of doomsday scenario.
The worry that we’re working with novel systems that we don’t completely comprehend has some merit. Although these systems are unlikely to spiral out of control in the traditional sense, they may wind up facilitating the dissemination of erroneous information, the creation of misleading content, etc.
There are hazards, no doubt, which is why Snap is implementing these further safeguards for its own AI technologies.
And it ought to be a primary focus considering the app’s young customer base.