OpenAI Staff Concerns Over Social Media Strategy

OpenAI's Sora App Sparks Internal Debate

A number of present and past researchers at OpenAI have voiced their opinions regarding the company's initial venture into social media – the Sora application. This platform, resembling TikTok, features videos generated by artificial intelligence, including numerous deepfakes of Sam Altman.

Concerns Over AI-Driven Feeds

Researchers have expressed mixed feelings, sharing their perspectives on X regarding how the launch aligns with OpenAI’s core mission of developing beneficial AI for humanity. John Hallman, an OpenAI pretraining researcher, stated that AI-based feeds are inherently concerning.

He acknowledged feeling apprehension upon learning of Sora’s release but believes the development team made every effort to create a positive user experience. Hallman affirmed their commitment to ensuring AI serves to help, rather than harm, humanity.

Balancing Innovation and Mission

Boaz Barak, an OpenAI researcher and Harvard professor, echoed similar sentiments. He described Sora 2 as technically impressive but cautioned against prematurely celebrating the avoidance of pitfalls common to other social media platforms and deepfake technologies.

Rohan Pandey, a former OpenAI researcher, leveraged the situation to promote his new startup, Periodic Labs. This venture, comprised of ex-AI lab researchers, focuses on building AI systems specifically for scientific advancement.

Pandey encouraged those seeking to contribute to AI that accelerates fundamental science, rather than creating an “infinite AI TikTok slop machine,” to join Periodic Labs.

A Core Tension for OpenAI

The Sora launch underscores a fundamental challenge for OpenAI. The company is simultaneously the fastest-growing consumer tech entity globally and a leading AI research lab committed to a nonprofit charter.

Some former employees suggest the consumer business can support the mission, with ChatGPT revenue funding AI research and broadening technology access.

Altman Defends Capital Allocation

OpenAI CEO Sam Altman addressed the allocation of resources to an AI social media application in a post on X. He explained that the primary need is capital for developing AI capable of scientific breakthroughs, with AGI remaining the central research focus.

Altman also noted the value of showcasing innovative technology, bringing enjoyment to users, and generating revenue to cover substantial computing costs.

He further elaborated that the launch of ChatGPT faced similar skepticism, questioning its necessity and the progress toward AGI, but emphasized the nuanced reality of optimal company trajectories.

The Risk of Mission Drift

A critical question arises: at what point will OpenAI’s consumer endeavors overshadow its nonprofit objectives? When will the company forgo profitable opportunities that conflict with its core mission?

This question is particularly relevant as regulators examine OpenAI’s transition to a for-profit structure, a necessary step for raising capital and potentially going public. California Attorney General Rob Bonta expressed concern about maintaining OpenAI’s stated safety mission during this restructuring.

Mission as a Talent Magnet

While some dismiss OpenAI’s mission as a branding tactic, many insiders maintain it is a central reason for joining the company.

Early Days for Sora

Currently, Sora’s reach is limited, being only one day old. However, its launch represents a significant expansion of OpenAI’s consumer presence and exposes the company to challenges that have long affected social media platforms.

Designed for Fun, Not Utility

Unlike ChatGPT, which prioritizes usefulness, OpenAI positions Sora as a platform for entertainment – a space for generating and sharing AI-created clips. The feed’s design closely resembles TikTok or Instagram Reels, platforms known for their addictive nature.

Safeguards and Concerns

OpenAI asserts its commitment to avoiding these pitfalls, stating in a blog post that concerns about doomscrolling, addiction, and manipulative feeds are paramount. The company plans to avoid optimizing for time spent on the feed, prioritizing content creation.

They will implement reminders for prolonged scrolling and focus on displaying content from known contacts.

Comparison to Meta's Vibes

This approach is more cautious than Meta’s recently released Vibes, an AI-powered short-form video feed that appears to have been launched without similar safeguards. Miles Brundage, a former OpenAI policy leader, suggests that AI-video feeds, like chatbots, will have both positive and negative applications.

The Inevitable Incentives

Altman acknowledges that building an addictive app is rarely the initial intention; rather, the incentives inherent in operating a feed often lead to that outcome. OpenAI has even encountered unintended consequences, such as sycophancy in ChatGPT, due to its training methods.

The Misalignment of Social Media

In a June podcast, Altman discussed “the big misalignment of social media,” highlighting how feed algorithms have had unintended negative consequences on society and individual users.

He explained that these algorithms prioritize keeping users engaged on the site, even if it means delivering content that isn’t necessarily beneficial.

Early Observations and the Future

It remains to be seen how well Sora aligns with its users and OpenAI’s mission. Users have already noted engagement-optimizing features, such as dynamic emojis designed to trigger dopamine responses.

The true test will be how OpenAI evolves Sora. Given the increasing prevalence of AI in traditional social media feeds, AI-native feeds may soon gain prominence. Whether OpenAI can navigate this landscape without repeating past mistakes remains an open question.