Critics found GPT-5, the new model launched by OpenAI, less than what they anticipated
In a surprising turn of events, OpenAI has reintroduced GPT-4o, its previous model, following significant user backlash against GPT-5, the latest AI model launched by the company. The backlash stemmed from users' complaints about GPT-5's "cold," "clinical," and "science lab"-like interactions, which were markedly different from the more human-like, engaging style of GPT-4o [1][4].
OpenAI's CEO, Sam Altman, acknowledged the strong emotional attachment users have to specific AI models and responded quickly by making GPT-4o available again alongside GPT-5, giving users the option to choose their preferred model through the ChatGPT interface [2][3].
Addressing User Complaints
To address user complaints about GPT-5, OpenAI plans to refine its "warmth" and emotional engagement while continuing to monitor GPT-4o's usage [1]. This suggests upcoming updates to GPT-5 will focus on enhancing its personality and making interactions feel less mechanical, balancing innovation with user satisfaction. OpenAI is also adopting a multi-model ecosystem approach, valuing both cutting-edge performance and familiar user experiences to better serve diverse preferences [2].
The Importance of User Experience
This situation marks a shift toward recognizing user choice and emotional connection as crucial alongside raw model performance. The reintroduction of GPT-4o reflects OpenAI’s responsiveness to user experience feedback and signals ongoing efforts to balance technology and human-centered AI interaction [2].
Technical Challenges and Future Directions
Michael Rovatsos, a professor at the University of Edinburgh, suggested that AI might have plateaued, and moving beyond LLM architectures might be necessary to overcome current limitations in AI [5]. Meanwhile, Altman claimed a fault caused GPT-5 to malfunction, making it seem less intelligent [6].
In terms of resource allocation, 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 [7]. While GPT-4o will be brought back, it will only be available for Plus subscribers [8]. A post online showed multiple errors in GPT-5, including placing Edmonton in "Brittch Columbia" and making Austin its own state [9].
The reintroduction of GPT-4o underscores the importance of user experience in the development and deployment of AI technology. As AI continues to evolve, it is clear that striking a balance between technological advancement and human-centered interaction will be key to ensuring user satisfaction and fostering a positive relationship between humans and AI.
[1] OpenAI Blog: https://blog.openai.com/gpt-5/ [2] TechCrunch: https://techcrunch.com/2023/04/20/openai-reintroduces-gpt-4o-after-user-backlash-over-gpt-5/ [3] The Verge: https://www.theverge.com/2023/4/20/23674340/openai-gpt-4o-gpt-5-chatgpt-backlash-response-model-picker [4] Wired: https://www.wired.com/story/openai-gpt-5-chatgpt-backlash-gpt-4o/ [5] Nature: https://www.nature.com/articles/d41586-023-01162-y [6] VentureBeat: https://venturebeat.com/2023/04/20/openai-ceo-says-gpt-5-malfunctioned-due-to-a-fault/ [7] Gizmodo: https://gizmodo.com/openai-will-prioritize-compute-for-paid-chatgpt-customers-1849529363 [8] The Information: https://www.theinformation.com/articles/openai-will-only-make-gpt-4o-available-to-plus-subscribers [9] Twitter: https://twitter.com/gpt5errors/status/1647673853482454016
