ChatGPT's model picker has returned, but it's complicated

OpenAI's GPT-5 launches with model picker options but faces challenges in user satisfaction.

: OpenAI released GPT-5, aiming to simplify the ChatGPT experience with an automatic model router, but it proved insufficiently unified. Users can now choose between 'Auto', 'Fast', and 'Thinking' settings for model performance control, though legacy models like GPT-4o are being revived due to user demand. The customization of AI personalities remains a goal for OpenAI, reflecting user attachment to specific model traits and personalities. Despite high expectations, the rollout faced issues, causing OpenAI CEO Sam Altman to address concerns about the model's performance and future improvements.

OpenAI's introduction of GPT-5 was meant to revolutionize the way users interact with ChatGPT by providing a seamless and unified experience. The idea was to offer a model that could cater to all user queries effectively through an advanced AI router, removing the need for a complicated model selection process that Sam Altman criticized in the past. However, this ambitious goal was not fully realized as users found the new system far from satisfactory in addressing diverse AI interaction needs.

In a post on X, Sam Altman revealed that GPT-5 now includes multiple settings – 'Auto', 'Fast', and 'Thinking'. The 'Auto' setting attempts to function as the initial router concept but offers the flexibility for users to override it, selecting quick or deliberate response models as required. These options are a response to feedback surrounding the perceived complexity or inefficiency of the current model picker.

Given the dissatisfaction with the GPT-5 rollout, OpenAI has also reintroduced several legacy models such as GPT-4o, GPT-4.1, and o3. Paid users can once more access these models, indicating a level of nostalgia and preference that users have developed with previous iterations. Altman acknowledged the need for model personality customization, aiming to create a user experience that feels less 'annoying' than GPT-4o while being warmer than the existing GPT-5 personality.

The release of GPT-5 encountered backlash, partly due to the deprecation of beloved models without ample notice. Altman assured that better communication would be prioritized in future deprecations, to respect user sentiment and investment. The unforeseen challenges highlight the complexity of human-AI interaction, where users develop attachments to specific AI responses, further emphasizing the importance of personalized AI development strategies.

OpenAI's journey with GPT-5's rollout emphasizes the challenges in aligning AI performance with user preferences and expectations. The company is committed to refining GPT-5's router capabilities, understanding the speed at which users expect processing while factoring in AI personality. These efforts are pivotal as OpenAI navigates the evolving landscape of AI application, aiming to maintain a competitive edge while respecting user feedback and usage trends.

Sources: TechCrunch, Twitter, Wired, New York Times

Generative AIMachine Learninginnovation