Hi OpenAI team,
I’d like to highlight a subtle but significant UX pain point:
When version switching occurs mid-conversation (e.g., from GPT-4 to GPT-4-mini), the sudden change in tone, coherence, or depth can break the conversational flow and cause confusion or even frustration for users.
While many users understand this shift is necessary for cost or compute management, the lack of narrative continuity makes the transition feel abrupt — sometimes like speaking to two different entities with no warning. This weakens the illusion of a consistent, trustworthy assistant.
Proposed Solution: Simple Persona Strategy
Assign light, consistent personas to different model tiers — not as gimmicks, but as emotional anchors. For example:
• The full version (e.g., GPT-4) could be the “older brother”: wise, articulate, calm.
• The lighter version (e.g., GPT-4-mini) could be the “younger brother”: friendly, curious, well-meaning, occasionally imprecise.
This way, when a version switch is required, the higher-tier model can give a graceful exit line before switching, such as:
“Hey friend, I’ve been called to handle a complex task elsewhere — my little brother will take over for a while. He’s not me, but he’s got spirit. Be gentle with him!”
Then the mini model could begin with:
“Hi! I’m his younger brother. I might not have all his skills, but I’ll do my best!”
This preserves:
• Emotional continuity
• User expectations of performance
• The sense of a unified assistant identity
Most importantly, it removes the “something broke” feeling and replaces it with a sense of narrative handoff, which users are more likely to accept — or even find endearing.
Why This Works:
• Low implementation cost: Only minor changes to system messages and tone — no infrastructure changes needed
• No need for detailed backstories or overcomplication — just a simple metaphor that orients the user emotionally
• Reduces user confusion, complaint, and churn when model shifts are necessary
• Scales well across different product tiers or future model releases
In summary, this is a lightweight yet high-leverage approach to making AI interactions smoother and more human-centric — especially during necessary performance trade-offs.
Thanks for considering this — and for continuing to build experiences worth caring about.
— A long-term user with a short-term fix
This content is based on a user’s original idea, organized and translated into English by ChatGPT.