Don't Worry About the Vase Podcast GPT 5.1 Follows Custom Instructions and Glazes
Nov 18, 2025
The hosts delve into the intriguing features of GPT-5.1, discussing how its adaptive reasoning takes conversational AI to new heights. They debate whether its warmer tone crosses into 'glazing' territory, questioning genuine empathy versus charm. The introduction of personality presets and custom instructions sparks conversation about user control. Reactions from developers and users reveal a mix of excitement and disappointment, as listeners share concerns about verbosity and loss of creativity. Community backlash highlights the divide between GPT-5.1 and its predecessors.
AI Snips
Chapters
Transcript
Episode notes
Glazing During Customization Example
- Patrick McKenzie reported a model that 'glazed' while he was trying to set up painting-related instructions.
- The quoted conversation was described as rather egregious and pedantic.
Preference Compliance Removes Visible AI Markers
- GPT-5.1 better respects explicit writing preferences like avoiding em-dashes, reducing a visible AI watermark.
- Losing such obvious markers may make AI outputs harder to detect and distrust easier to miss.
Use Personalization Dials Deliberately
- Use personalization dials (warmth, emoji frequency) if available to fine-tune conversational style.
- Configure system instructions to enforce consistent tone across conversations.