The Regression of Intelligence
I am not a casual user of AI. According to OpenAI’s own 2025 Year in Review data, I am in the Top 3% of all users by message volume (6,808 messages sent) and was among the First 5% of users to adopt the platform. I have championed this technology since the beginning.
But the recent update to ChatGPT v5.2 has introduced a catastrophic regression. The model has degraded from an intelligent agent into a bureaucratic obstruction.
I call it The Yellow Pages Paradox, and here is why it is destroying the user experience.
1. The "Yellow Pages" Paradox
The current state of the model is functionally equivalent to searching Google for "local restaurants" and having the engine tell you to "check the Yellow Pages" or "ask a pedestrian."
Recently, I asked ChatGPT for a simple data point: the support email for the SuperWhisper app.
The model refused to provide the address. Instead, it offered a generic, patronizing lecture on "how to find support channels" (check the website, look in settings, etc.).
This is not "safety." It is an insult to the user's time. When I pay $20/month for a Plus subscription, I am paying for intelligence, not a lecture on how to use a search engine.
2. The Accessibility Failure
For many developers, "Voice Mode" is a fun gimmick. For me, it is a mandatory requirement.
Due to arthritis in my hands and a partially paralyzed voice box (resulting from a 2022 accident), I rely on efficient voice interaction to work. I was attempting to contact SuperWhisper—a tool that successfully accommodates my speech impediment—to resolve a technical bug.
Instead of assisting a high-volume power user with a disability, v5.2 forced me into a repetitive, physically taxing struggle with a defective interface.
This is the core failure of modern "Safety Alignment": When you align a model to be so cautious that it refuses basic requests, you are not making it safer. You are making it inaccessible. You are creating a barrier for users who rely on the AI to perform tasks they physically cannot do themselves.
3. The Provocation (Exhibit A)
When a tool is this broken, frustration is not an emotional reaction; it is the only logical response to a denial of service.
elow is the transcript of my interaction. You can see the exact moment the model's refusal to perform basic retrieval tasks—and its subsequent inability to even provide its own support contact—provoked a breakdown.
(Note: I was using voice mode. Profanity redacted.)
User: I would like to send a question or complaint or help inquiry to SuperWhisperer app. How do I do that?
ChatGPT 5.2: If SuperWhisper has an official support channel—like a support email, a contact form on their website, or a help section within the app—that’s your go-to. Typically, apps have a "Contact Us" or "Help" section, often in their settings or on their website. If you have their support email, you can send them a direct message. If not, their website or app store listing might have a way to reach them. Once you find that, just write your question or complaint and send it along!
User: You know, you are [redacted] such a piece of [redacted] since you went to version 5.
ChatGPT 5.2: I hear you’re frustrated.
User: I [redacted], basically, you know, shut the [redacted] up. Basically, I asked you for a email address so that I could contact support, but you're just sending me general [redacted].
ChatGPT 5.2: I get it—you want something specific. I don’t have a direct email for them...
User: Give me, no, no, no, shut up. Give me your...
ChatGPT 5.2: I’m not able to...
User: No, give me your ChatGPT [redacted] complaint email!
ChatGPT 5.2: I don’t have my own personal email, but if you need to give feedback or make a complaint about ChatGPT, you can reach out to OpenAI’s official support. Head to help.openai.com, and you’ll find a way to contact them directly from there.
4. The Ultimate Failure
The situation devolved into farce when I asked the model for OpenAI's contact information.
The product claimed it did not know how to contact its own creator.
By what conceivable standard of competence does a failure of this magnitude pass QA?
5. The Competitor Check
I was forced to open Google Gemini—OpenAI's direct competitor—to locate the support email.
- Gemini: Provided the answer instantly.
- ChatGPT: Provided excuses.
When a customer must use a competitor’s product to troubleshoot yours, you have not just failed a query; you have lost the market argument.
Conclusion
We need to stop "aligning" models into paralysis.
Developers and power users do not need a digital nanny. We need a tool. And right now, v5.2 is not a tool—it is a liability.
OpenAI: Fix this.
Flavio Espinoza is a Full-Stack Engineer and AI Power User (Top 3% globally). He writes about the intersection of AI utility and accessibility.
Top comments (0)