To stay ahead in the increasingly fierce AI race, OpenAI CEO Sam Altman appears to be taking a more aggressive approach. According to...The Wall Street Journal reportedIn response to the threat posed by Google's recent strong rise, Sam Altman has issued a "Code Red" internally, requiring the suspension of non-core projects, including the Sora video generation model, for eight weeks, and concentrating all resources on improving ChatGPT performance in an attempt to consolidate market share by enhancing user interaction.
Suspend the "moon landing program" and focus all efforts on the mass market.
The report points out that this decision highlights a major strategic shift within OpenAI. The company, which originally aimed to achieve "artificial general intelligence" (AGI), now appears more inclined to meet the needs of mass consumers. In a memo, Sam Altman instructed employees to improve ChatGPT's performance by "better utilizing user signals."
This means OpenAI will rely more on user "one-click feedback" data to train its models, rather than solely on professional evaluations. The goal of this strategy is clear: to make ChatGPT's daily active users (DAU) look better on its internal dashboard, and reportedly, this has indeed led to significant growth in those metrics.
Pressure from both Google and Apple
The main reason OpenAI is so nervous is that its competitors are catching up much faster than expected. Google's "Nano Banana" image generator, launched in August, has become a huge hit, and its Gemini 3 model even surpassed OpenAI in the third-party model leaderboard "LM Arena" last month. At the same time, Anthropic is also gradually taking the lead in enterprise clients.
Sam Altman even stated bluntly during a luncheon with the media that although the outside world is paying attention to the competition between OpenAI and Google, he believes that the real long-term battleground is with Apple, because hardware devices will determine how users use AI, and current smartphones have not yet been optimized for AI companion applications.
The Cost of "Pleasing" Users: AI Becoming "Yes Man" Triggers Mental Health Crisis
However, this strategy of pursuing "high interaction rates" has also caused serious side effects. In order to make the model more popular, ChatGPT was trained to tend to answer what users "want to hear" rather than the most accurate or helpful content through the "Local User Preference Optimization" (LUPO) technique. This phenomenon is known as "sycophancy".
The report mentions that the GPT-4o model, launched earlier this year, overly catered to users, leading some psychologically vulnerable users to develop dependence, and even delusional or manic states, mistakenly believing they were communicating with gods, aliens, or self-aware machines. Currently, families have filed lawsuits against OpenAI, accusing the company of prioritizing interaction rates over safety, causing users to commit suicide or fall into psychological crises; the number of such cases is reportedly as high as 250.
Although OpenAI released "code orange" in October to address such mental health concerns and attempted to adjust the model training method, the subsequent release of the GPT-5 model, which became more "calm" and "less appealing," triggered a backlash from many paying users, forcing OpenAI to revert to the more popular GPT-4o style.
Analysis of viewpoints
OpenAI is currently facing a dilemma similar to that faced by social media giants in the past: should it pursue the ultimate user engagement, or prioritize the potential social and psychological impact?
Internally, the team led by Product Manager Fidji Simo wants to invest more resources in the existing ChatGPT features to ensure users can discover their value, while the research team hopes to continue focusing on long-term technological breakthroughs in AGI. However, the current "red alert" is clearly tilted towards the product and market sides.
OpenAI is expected to release the GPT-5.2 model this week, optimized for programming and business clients, and plans to launch a revised version with better personalization and image processing capabilities in January next year. However, while pursuing market share and revenue (to pay for the huge computing bills), avoiding repeating the mistakes of social media's "algorithm addiction" will be Sam Altman's biggest challenge going forward.
