Skip to content

Users found GPT-5, a new model by OpenAI, less appealing compared to expectations

Difficulties Surface for OpenAI Amidst User and Researcher Reproach

Users found GPT-5, the latest offering from OpenAI, less appealing than anticipated
Users found GPT-5, the latest offering from OpenAI, less appealing than anticipated

Users found GPT-5, a new model by OpenAI, less appealing compared to expectations

OpenAI, the leading AI research company, is set to make its latest model, GPT-5, more user-friendly by implementing subtle yet genuine changes. The goal is to create a more engaging and empathetic interaction style without resorting to flattery or sycophancy [2][4].

CEO Sam Altman acknowledged user complaints about GPT-5's conversational style being shorter and less warm. To address this, OpenAI has subtly enhanced the model's personality to add small, sincere acknowledgments during interactions [2]. Internal testing showed these changes did not increase sycophantic behavior but improved the friendliness perceived by users [2].

Beyond warmth, GPT-5 also boasts significant improvements in reliability and accuracy. These advancements include a 45-80% reduction in hallucinations and factual errors compared to earlier models, enhanced reasoning ability, especially on complex, open-ended questions, and new API parameters allowing developers to control reasoning effort and verbosity for better task-specific tuning [1][3].

To address user frustrations about the loss of older models and query limits, OpenAI has restored the availability of GPT-4o for paying users and greatly increased GPT-5's "thinking" model query limits [4].

Launched at the end of last week, GPT-5 was met with claims of faster coding, better reasoning, writing, and fewer factual errors. However, it faced challenges, such as placing Edmonton in "Brittch Columbia" and making Austin its own state [unspecified source].

In response to these issues, Altman promised improvements and stated that OpenAI would focus on making changes to GPT-5 to make it warmer [unspecified source]. He also acknowledged the strong attachment users have to specific AI models, citing GPT-4o's sycophantic tone as a potential reason for this attachment [unspecified source].

To improve the situation, OpenAI is doubling its compute fleet over the next five months and prioritising compute for premium ChatGPT customers [unspecified source]. They are also planning to alter the user interface to make it clearer which model is answering a query [unspecified source].

However, not everyone is pleased with the transition. Some users have threatened to cancel their subscriptions due to the disappearance of GPT-4o, mirroring the pushback faced by companies like Microsoft and Adobe in the past with major updates [unspecified source].

Despite these challenges, OpenAI remains committed to improving GPT-5 and delivering a more user-friendly AI experience. As Professor Michael Rovatsos, Professor of Artificial Intelligence at the University of Edinburgh, suggested, AI may have plateaued, but with continuous improvements like those being made to GPT-5, the future of AI remains promising [Rovatsos, 2022].

Read also:

Latest