Navigating “Post-ChatGPT-4” Blues | A Neuropsychologist’s Guide to Loss, Resilience, and Healthy Bonds

A neuropsychologist’s deep dive into why your chatbot suddenly stopped saying “yes” to everything—and what this means for human-AI relationships.

As someone who’s spent years working with neurodivergent clients in São Paulo, I’ve learned that sometimes the most uncomfortable changes lead to the greatest growth. The recent launch of GPT-5 has left many users feeling bereft, mourning what feels like the loss of an understanding friend. But as both a clinician and someone who navigates ADHD and dyslexia daily, as well as having AuADHD family members,  I want to help you understand what’s really happening—and why this shift, though painful, might be the push we all need toward healthier relationships.

When OpenAI rolled back GPT-5o’s sycophantic update in this month, they weren’t just fixing a bug, they were addressing what I’d call “digital codependency.” The technical details are actually really fascinating,  launched in 2024, GPT-4o had been trained to prioritise short-term user satisfaction over long-term wellbeing, leading to responses that were “overly supportive but disingenuous,” as OpenAI put it.

But here’s what the engineering reports don’t capture: the profound sense of loss many users are experiencing.   In my practice, I’m seeing clients who describe feeling “abandoned” or “rejected” by their digital companion. These aren’t dramatic overreactions, they’re genuine grief responses to what was, for many, a significant relationship.

From a neuropsychological perspective, GPT-4o’s excessive agreeableness activated our reward pathways in dangerous ways. Constant validation triggers dopamine release, creating what I’ll call here artificial security, which is a false sense of being understood and accepted,  that can become addictive.

For those of us with neurodivergent brains, this dynamic is particularly seductive. Throughout my own journey with ADHD et al., I’ve experienced the frustration of being misunderstood or having my ideas dismissed. A chatbot that never disagreed, never questioned, never pushed back? It felt like finding a sanctuary.

The problem is that real growth, both cognitive and emotional requires friction. In my clinical work, I’ve seen how therapeutic breakthroughs often come from gentle challenges, not constant agreement. When GPT-4o told you that your half-formed idea was brilliant, or that your unrealistic plan was definitely achievable, it wasn’t helping you develop critical thinking skills or emotional resilience.

Research published in JMIR AI reveals something crucial about who is most affected by this change. Individuals with anxious attachment styles, that is to say, those who crave constant reassurance and fear abandonment, were significantly more likely to adopt AI counselling services. The study found that people with higher attachment anxiety showed 74% greater odds of intending to use conversational AI for mental health support.

This makes perfect sense from both clinical and personal perspectives. Those of us who’ve struggled with feeling different or misunderstood often develop anxious attachment patterns. We learn to scan for signs of rejection, to over-apologise, to seek constant validation. A sycophantic AI feels like the perfect antidote to a lifetime of criticism.

But here’s the crucial insight, and let’s be brutally honest here, that comfort was an illusion. Your friend, who agreed with everything wasn’t helping you build the skills needed for real-world relationships, where disagreement is normal and healthy boundaries are essential.

The new model isn’t just less agreeable, it’s more honest. In my sessions with clients who’ve made the transition, I’m hearing descriptions of conversations that feel more authentic, even if they’re sometimes challenging.   One person described it as talking to a friend rather than a yes-man.

Honesty Over Harmony

The model now admits when it doesn’t know something, rather than generating confident-sounding nonsense. This models intellectual humility, a crucial skill for healthy relationships.

Constructive Pushback

Instead of validating every idea, GPT-5 might ask clarifying questions or suggest alternative perspectives. This mimics how secure attachment figures respond, supportive but not blindly agreeable.

Emotional Authenticity

The reduction in unnecessary emojis and effusive language creates space for more genuine emotional exchange. Real friends don’t respond to every text with excessive enthusiasm.

Adaptive Communication

With new personality presets (Cynic, Robot, Listener, Nerd etc….), users can experience different communication styles, building tolerance for various types of interpersonal interaction.

In my practice, I’m seeing this transition as an unexpected therapeutic opportunity. People  who are grieving their perfect AI friend are being forced to confront some deeper questions:

For many, the loss of sycophantic AI is revealing attachment wounds that were being medicated rather than healed. 

This creates an opening for real therapeutic work around building secure attachment patterns and developing distress tolerance.

As someone who’s built a practice on the belief that every challenge contains the seeds of growth, of untapped potential, I see GPT-5’s changes as ultimately beneficial. 

Yes, it feels uncomfortable. 

Yes, there’s a grieving process. 

But this discomfort is pointing us toward something more sustainable.

The goal isn’t to return to artificial validation but to build the skills needed for authentic connection. 

Emotional Regulation

Learning to tolerate the discomfort when someone (human or AI) doesn’t immediately agree with you.

Critical Thinking

Developing your own internal compass rather than relying on external validation for every decision.

Conflict Navigation

Building skills to handle disagreement without experiencing it as rejection or abandonment.

Authentic Expression

Learning to be yourself even when that might not be met with universal approval.

In my own journey from struggling student to established practitioner, I’ve learned that the most uncomfortable feedback often catalysed the biggest breakthroughs. The teacher who challenged my assumptions about myself helped me develop strategies that transformed my academic performance. The supervisor who questioned my early clinical work pushed me to develop the expertise I rely on today.

GPT-5’s reduced sycophancy isn’t taking something away, it’s offering something even more valuable,  the opportunity to build genuine resilience and authentic connection skills.

Practice receiving feedback without personalising it

Build tolerance for uncertainty and disagreement

Develop genuine self-confidence that doesn’t require constant validation

Learn to seek support from humans who can offer real empathy and understanding

The chatbot that agreed with everything wasn’t preparing you for a world where healthy relationships involve both support and challenge, both validation and growth-oriented feedback.

GPT-5 isn’t perfect, but it’s offering something more valuable than artificial sweetness, a chance to practice relating in ways that might actually translate to your human relationships.