Cselle says he recognizes the dangers of offering users AI-generated text and that users are free to edit or ignore the suggestions. “We don’t want a situation where robots are pretending to be humans and the entire platform is them talking to each other,” he says.
To protect the integrity of the community as it opens the door to more than 300 million people, Pebble will also use generative AI to monitor new registrations. The system will use OpenAI’s GPT-3.5 model to compare people’s X bio and recent posts to Pebble’s community guidelines, which, unlike Musk’s service, prohibit all nudity and violent content.
Mike Greer, Pebble’s CTO, says the goal is to determine “whether someone is inherently toxic and treats others poorly.” Those that are or do so will be blocked and reviewed manually. Pebble intends to compare potential users to “other sources of truth” online once it opens up signups further, it says, to include people without X accounts.
Newcomers to Pebble will find that its AI-generated suggestions include posts to introduce yourself to the service or welcome newcomers. Ideas also encourages playful banter with the official Pebble account, which itself is a bot powered by generative AI. It can also suggest replies to other people’s posts.
During my experiments last week, Ideas suggested responding to a post from Katie Harbath, who runs the tech policy consultancy Anchor Change. She had linked to her newsletter discussing the election-related content policies of social media companies. I posted Pebble’s suggested answer in full: “Interesting analysis! It is crucial to understand how platforms evolve in the political landscape. As the 2024 elections approach, their approaches will undoubtedly shape the digital political sphere. Thanks for sharing! #PlatformPolitics.
Although I was concerned that this response would be considered spam, Harbath says it “didn’t call AI out on me.”
Early testers of the AI feature, which is also powered by OpenAI’s GPT-3.5, changed 85% of its suggestions before releasing them, Cselle says. The system will learn from user feedback, so it hopes ideas will improve and be more personalized over time. “We really want these tools to help people be the best version of themselves and allow them to gain a pure sense of their voice, without replacing it or removing it,” says Sarah Oh, COO of Pebble and responsible for trust and security policy.
