OpenAI recently launched GPT-5, which CEO Sam Altman characterized as the company’s “smartest, fastest, most useful model yet.” However, this release has sparked significant user backlash, leading to a user revolt unusual for consumer AI products. An anonymous developer created a blind testing tool, accessible online, where users compare the outputs of GPT-5 and its predecessor, GPT-4o, without prior knowledge of which model generated which response. This application aims to reveal user preferences and experiences in a more controlled setting.
The tool allows users to evaluate responses across multiple rounds by presenting pairs generated from both models to gather insights that may not be visible through typical user interactions. Since its launch, the testing tool has garnered over 213,000 views, indicating a shared interest in understanding the distinctions between these AI versions.
Early feedback from users reflects a divided opinion, echoing broader debates within the AI community. While some users prefer GPT-5’s accuracy, others remain loyal to GPT-4o, which is characterized by a more engaging and warmer interaction style. The difference in user experiences raises questions about how AI personality and communication styles impact user satisfaction.
This ongoing situation highlights the tension within AI development: how should future models balance technical improvements with user expectations for relational dynamics? Users have documented feelings of attachment to GPT-4o, further complicating discussions about AI functionality and emotional support capabilities. As OpenAI considers user feedback, the challenge will be to create a model that caters to diverse preferences, potentially leading to a dual offering of AI personalities — a development acknowledging that user needs can vary widely.
Ultimately, the recent developments call into question the metrics that define success in AI, suggesting that user preference may be as significant as technical benchmarks in shaping future AI interactions.
Source: https://venturebeat.com/ai/this-website-lets-you-blind-test-gpt-5-vs-gpt-4o-and-the-results-may-surprise-you/