GPT-6 Astra won't try hard unless you ask (srcreigh.ca)

The recent experience with GPT-6 Astra highlights challenges users face when seeking high-quality output from advanced AI models. Users have found that even at high reasoning levels, Astra sometimes doesn't apply itself fully, producing subpar results that neglect critical integration and design considerations — a concern particularly relevant in technical tasks like coding and deployment planning. Despite its potential, Astra appears to base its effort on the given context, leading to outputs that may lack detail, coherence, and attention to potential issues. This user frustration underscores a significant implication for the AI/ML community: models like GPT-6 Astra require more explicit prompting to maximize their capabilities. By iteratively encouraging the model to explore details and address concerns, users can enhance the generated outcomes. However, this process of coaxing higher quality from the model—a combination of increased complexity and depth in prompting—can be time-consuming. This situation reflects an important development area for AI firms: refining their models' responsiveness and adaptability to user needs, ultimately striving for outputs consistently aligned with user expectations, especially in complex applications.