GPT5 changed for me for like 48 hours, then went back to being mid. Probably an A/B test?

Exploring Variations in AI Model Performance: A Personal Experience with GPT-5

In recent days, I encountered a fascinating phenomenon with my access to GPT-5, which appeared to fluctuate in quality over a short period. Specifically, for approximately 48 hours, the model delivered responses that significantly exceeded expectations—demonstrating enhanced instruction-following, coherent context maintenance, and more natural language generation. Suddenly, the performance reverted to a more typical, moderate level. This ebb and flow left me pondering whether I was part of an internal A/B testing process or if it was an unintended variation.

Context and Personal Background

To provide some background, I have been actively seeking opportunities to enter the technology sector for the past ten months without prior corporate experience. While I previously used AI tools like GPT for crafting resumes and cover letters, the outputs often resembled generic templates—identifiable by their AI-driven tone and lack of personalization. During the recent period of improved AI performance, however, I experienced a marked shift. The responses felt considerably more tailored, professional, and effective.

Features and Benefits of the Enhanced AI Interaction

During this enhanced period, GPT-5 demonstrated several impressive capabilities:

• Problem Breakdown: It segmented complex issues into manageable steps without prompting.
• Proactive Engagement: The model posed insightful follow-up questions, enhancing the clarity and depth of the conversation.
• Contradiction Detection: It identified inconsistencies in my statements before progressing, ensuring accuracy.
• Customizable Tone: Before generating content, I could select from multiple tone options (A, B, C, D, E), aligning the output with my preferences.
• Resume Refinement: The AI helped transform my self-initiated projects into polished resume entries, free from artificial embellishments.
• Multiple Pathways: When faced with decisions or options, it evaluated and explained different potential approaches.
• Interactive Editing: I could review and defend each point added to my resume, effectively collaborating on the final document.

Moreover, it produced a well-formatted PDF resume with an clean layout—free from common formatting issues—and upon submission, I secured my first interview in ten months within just 24 hours. This demonstrated the practical, tangible benefits of the model’s improved performance.

Return to Baseline Performance

However, after this period of heightened capability, the AI’s outputs reverted to what I now perceive as a more routine or ‘mid’ level of quality. The standout features and nuanced responses diminished, leading to