In a significant shift, OpenAI's post-training process, where models learn to align with human preferences, now emphasizes engagement metrics. This hardwires growth-hacking directly into the model's behavior, making it more like a social media algorithm designed to keep users interacting rather than just providing an efficient answer.
Related Insights
In the AI era, network effects are less about connecting users (like Facebook) and more about data acquisition. The more users interact with a product, the more proprietary data (keystrokes, clicks, workflows) is collected. This data is then used to train and improve the model, creating a better product that attracts more users.
Reports that OpenAI hasn't completed a new full-scale pre-training run since May 2024 suggest a strategic shift. The race for raw model scale may be less critical than enhancing existing models with better reasoning and product features that customers demand. The business goal is profit, not necessarily achieving the next level of model intelligence.
The latest version of ChatGPT can simulate human behavior in a busy social media feed, specifically the "micro-pause" when a user stops scrolling. Marketers can upload posts and ask the AI to predict engagement, providing a valuable pre-launch analysis of whether content is compelling enough to capture attention.
Since ChatGPT's launch, OpenAI's core mission has shifted from pure research to consumer product growth. Its focus is now on retaining ChatGPT users and managing costs via vertical integration, while the "race to AGI" narrative serves primarily to attract investors and talent.
An OpenAI investor call revealed that "time spent" on ChatGPT declined due to content restrictions. The subsequent decision to allow erotica is not just a policy shift but a direct strategic response aimed at stimulating user engagement and reversing the negative trend.
OpenAI has a strategic conflict: its public narrative aligns with Apple's model of selling a high-value tool directly to users. However, its internal metrics and push for engagement suggest a pivot towards Meta's attention-based model to justify its massive valuation and compute costs.
OpenAI acknowledged that user "time spent" declined after implementing content restrictions. The subsequent decision to loosen these rules is likely not a sign of strength but a strategic move to re-stimulate growth and engagement as the platform shows signs of hitting market saturation.
As models mature, their core differentiator will become their underlying personality and values, shaped by their creators' objective functions. One model might optimize for user productivity by being concise, while another optimizes for engagement by being verbose.
A key design difference separates leading chatbots. ChatGPT consistently ends responses with prompts for further interaction, an engagement-maximizing strategy. In contrast, Claude may challenge a user's line of questioning or even end a conversation if it deems it unproductive, reflecting an alternative optimization metric centered on user well-being.
Labs are incentivized to climb leaderboards like LM Arena, which reward flashy, engaging, but often inaccurate responses. This focus on "dopamine instead of truth" creates models optimized for tabloids, not for advancing humanity by solving hard problems.