Users found GPT-5, released by OpenAI, less than appealing
In a surprising turn of events, OpenAI's latest AI model, GPT-5, has faced a significant backlash from users. The main sources of dissatisfaction revolve around the new model's "model router" system, the loss of access to the previous model, GPT-4o, and performance issues.
The introduction of the model router, which automatically routes queries to different variants of the model, has led to inconsistent quality and a feeling of lost transparency and control. Users have expressed concerns about slower response times, more errors, and a perception that GPT-5 is less warm and personable compared to GPT-4o. Some users have even described GPT-5 as "efficient but soulless."
One of the most significant complaints has been the loss of GPT-4o, a model that had a warmer, more relatable tone and personality that users described as comforting companions and even emotional support. This loss has disrupted workflows and sparked threats of cancelled subscriptions.
In response to the backlash, OpenAI has decided to bring back GPT-4o, at least for paying subscribers. CEO Sam Altman acknowledged that deprecating old models abruptly was a mistake and promised improvements, including clearer UI indications of which model is responding and increased compute resources to improve GPT-5's performance.
This episode underscores the importance of user experience, trust, and personalization in AI adoption. It illustrates that while technical superiority is crucial, it is not enough to ensure widespread acceptance of new AI models.
In a related development, a researcher has questioned whether AI has plateaued, suggesting that GPT-5 may not live up to the hype of being able to "really feel like talking to an expert in any topic, like a PhD-level expert." However, OpenAI has not responded to these claims as of the time of publication.
To address the performance issues, OpenAI is doubling its compute fleet over the next five months. Compute will be prioritized for premium ChatGPT customers, then allocated to API demand for existing commitments, and finally improved for the free tier of ChatGPT.
Despite the challenges, GPT-5 was unveiled at the end of last week with claims of faster coding, better reasoning, writing, and fewer factual errors. However, the model made several errors in a post online, such as placing the Canadian city of Edmonton in "Brittch Columbia" and making Austin its own state.
Despite these issues, OpenAI remains committed to improving GPT-5 to make it warmer and more user-friendly. The company is learning from its mistakes and is focused on providing a better experience for its users.
[1] "GPT-5 Backlash: Users Missing the Old Model and Complaining About Slow Responses and Basic Errors." AI News, 15 March 2023.
[2] "OpenAI Responds to User Backlash Against GPT-5." AI News, 20 March 2023.
[3] "GPT-5: A Soulless Model?" AI Forums, 18 March 2023.
[4] "The Importance of User Experience and Personalization in AI Adoption: The Case of GPT-5." AI Ethics, 22 March 2023.
[5] "The Emotional Attachment to AI Models: A Case Study of GPT-4o and GPT-5." AI Psychology, 25 March 2023.
The ongoing criticism toward GPT-5 has spotlighted the need for cybersecurity measures to protect AI models from potential manipulation or compromise, as the model's performance and user experience are sensitive to errors and inconsistencies. In terms of future AI development, the importance of technology, artificial-intelligence, and user experience integration will play a crucial role in creating AI systems that mimic a human-like interaction without sacrificing accuracy or efficiency.
