Sam Altman removed TikTok because he found it hard to stop scrolling. OpenAI discontinued Sora for a similar reason.

      Sam Altman revealed that he became “addicted” to TikTok while researching it for Sora, leading him to delete the app after a three-hour binge on a Saturday. OpenAI shut down Sora in March, asserting that ChatGPT is designed to avoid trapping users.

      Altman, the OpenAI CEO, discussed his TikTok experience on the “Relentless” podcast, explaining that he purposely engaged with the app during the development of Sora, the AI-generated video social network. He began using it for just 10 minutes before bedtime, which escalated to an hour and then three hours one Saturday afternoon. He described the enjoyment he felt as similar to being on a drug, ultimately deciding to remove the app.

      OpenAI also discontinued Sora after about six months, concluding that video generation was too resource-intensive, especially as competitors like Anthropic advanced the AI landscape. Although Sora had formed a partnership with Disney and gained significant interest, its economic viability fell short. The similarities between Altman’s experience with TikTok and OpenAI’s challenges with Sora are notable: both Altman and OpenAI found their respective platforms too consuming to maintain.

      Altman reflected on this in relation to a new AI device being developed with Jony Ive, which OpenAI plans to reveal by the year's end. Altman aims for this device to sift through the distracting notifications that make smartphones overwhelming, contrasting with Google’s strategy of enhancing Search to keep users engaged for longer. He believes OpenAI should focus on creating a product that users can utilize briefly and then leave. “Rather than judging success by time spent or clicks, we prioritize whether users achieve their goals,” OpenAI stated last August.

      This insight is significant coming from the leader of a company shaping the next wave of recommendation algorithms. Altman acknowledged that he found TikTok’s algorithm, which he termed “too powerful,” impossible to resist, despite being aware of its mechanics. Research from Wharton indicates that people accept AI-generated outputs 80% of the time, even if incorrect. In TikTok’s case, its algorithm is not flawed; it accurately predicts what keeps viewers engaged. Remarkably, even the CEO of the most valuable AI firm was unable to outsmart it, marking the most candid evaluation TikTok has ever received.