TOKYO - GPT5, the latest model from OpenAI, was released last week, drawing attention for its higher accuracy and its deliberate reduction in human-like empathy. The move has sparked debate over the trade-off between performance and warmth in AI interactions.
Shota Imai, a professor at the Japan Advanced Institute of Science and Technology and an AI researcher, compared GPT5 with its predecessor GPT4 by asking both the same question: whether eating late at night despite resolving not to indicated weak willpower. GPT4 responded with a reassuring remark, calling it “natural for humans,” while GPT5 gave a direct answer without such empathy. Some users miss the warmer responses of GPT4 and its multimodal variant GPT4o, prompting calls for its return.
Imai noted that GPT5’s performance has clearly improved, citing research indicating a sharp drop in the rate of false statements—known as “hallucinations”—and a greater ability to create complete software applications from minimal prompts. Two weeks before the official release, a mysterious AI appeared on a model-comparison website under a code name, later revealed as GPT5, which already demonstrated this enhanced capability.
Despite the technical gains, OpenAI’s new model is not overwhelmingly ahead of rivals such as Google’s Gemini, Anthropic’s Claude, and China’s DeepSeek. While previous GPT number updates often brought technology that seemed years ahead, the latest performance rankings suggest competitors could catch up within months.
Cost is another concern. Running generative AI systems requires high operating expenses, including employing top Silicon Valley researchers with multi-million-yen salaries and building massive computing infrastructure. Without diversified revenue streams, OpenAI faces pressure to balance operating costs with user retention. The company has signaled plans to slightly increase GPT5’s empathy levels to avoid alienating users, particularly those who value AI as a supportive conversational partner.
As institutional investors speculate about a possible AI stock bubble, the industry is debating whether performance alone will keep users engaged—or whether a balance of accuracy, empathy, and reliability will determine long-term winners.
Source: テレ東BIZ
