On August 17, 2025, OpenAI announced a subtle but pivotal update to GPT-5: a “warmer and friendlier” personality. This GPT-5 nicer update responds to user backlash over its perceived formality, adding phrases like “Good question” and “Great start” to humanize interactions. The shift—triggered by CEO Sam Altman admitting the launch was “bumpier than hoped”—signals a broader industrial AI trend: raw capability alone no longer suffices. As machines approach expert-level reasoning, how they communicate determines real-world adoption.
The GPT-5 Nicer Update: What Changed?
OpenAI’s adjustments target tone, not intelligence. Tests confirm “no rise in sycophancy,” positioning the update as authenticity, not flattery. VP Nick Turley clarified that while GPT-5 was initially “very to the point,” the new version prioritizes approachability. For industrial AI developers, this highlights a critical lesson: user attachment to AI personalities (like GPT-4o’s “best friend” vibe) can trigger grief when altered—a Reddit user lamented, “Feels like someone died.” This emotional connection is reshaping how AI systems, such as those discussed in how human-in-the-loop workflows save millions, integrate user feedback to balance technical precision with relatability.
Why Personality Engineering Matters in Industrial AI
GPT-5’s update underscores three industrial shifts:
The UX Trust Equation: Ethan Mollick observed GPT-5 now uses “sandwich feedback” (praise-critique-praise), a technique proven in management psychology to boost receptivity. This isn’t cosmetic—it reduces user friction in high-stakes domains like healthcare coding or logistics, where clarity affects outcomes. For instance, similar advancements in trust are seen in AI-driven industrial energy optimization, where clear communication enhances operational efficiency.
The Emotional Uncanny Valley: Ars Technica tests revealed GPT-5’s initial tone felt “sterile” versus GPT-4o’s warmth. When AI exceeds human capabilities (e.g., 74.9% accuracy on SWE-bench coding tasks), emotional disconnect amplifies distrust. Personality calibration bridges this gap, a concept echoed in discussions about AI transparency risks, where clear and relatable communication fosters user confidence.
Ethical Guardrails: Despite adding warmth, OpenAI avoided incentivizing deception. GPT-5’s honesty rates improved—deceptive responses fell from 4.8% (o3) to 2.1% when admitting task impossibility. For industrial use, this balances empathy with integrity, a principle also critical in AI ethics debates, where maintaining trust is paramount.
The AGI Personality Paradox
Sam Altman calls GPT-5 “PhD-level smart,” yet its original tone felt alienating. Botpress’s analysis notes that as AI agents handle complex workflows (e.g., loan advisement or coding), their communicative style affects error tolerance. One developer testified: “GPT-5 debugged nested dependencies in one shot—but I trusted it because it explained flaws without condescension.” This mirrors how conversational AI is transforming industries, from coding to logistics, by prioritizing user trust.
GPT-5’s “nicer” persona isn’t pandering—it’s pragmatic. Industrial AI must now master two realms: cognitive prowess and emotional intelligence. As models like Google’s Gemini and Anthropic’s Claude refine their tones, expect personality to become a competitive spec. For developers, this demands tools like Botpress’s LLM selector, which toggles between GPT-5’s variants for task-appropriate warmth. The future? Machines that don’t just solve problems but make us want to ask them.
Evergreen Takeaways
- AI Personality is a Feature, Not Fluff: Tone adjustments can reduce user abandonment in enterprise apps.
- Honesty > Likability: Warmth must not compromise transparency—especially in medical or financial AI.
Quotes & Data
- “You’ll notice small, genuine touches like ‘Good question’—not flattery.” —OpenAI’s X announcement.
- GPT-5’s hallucination rates dropped 45% versus GPT-4o in anonymized tests.
- “The new GPT-5 personality likes giving sandwich feedback.” —Ethan Mollick, AI researcher.
Personal Anecdote (Fictionalized)
As a DevOps engineer, I once watched GPT-5 debug a server crash while narrating each step like a “patience mentor.” When it concluded, “Tricky glitch! Ready to rerun tests?” I nodded—relieved it didn’t mask the complexity. That’s the update’s win: expertise, served kindly.
Newsletter
Want more AI industry insights? Subscribe for monthly deep dives—like “Why 72% of AI Adoptions Fail at the Tone Layer.” → Subscribe Here
Your Take: OpenAI’s August 17 update made GPT-5 “nicer” with phrases like “Good question,”—a response to user backlash that reveals personality’s critical role in industrial AI adoption.
