← sources

Silent Amnesia?

source: reddit (reddit:ChatGPT)published 2026-11-07 04:19ingested 2026-11-08 02:36

retained text

Seeking Advice: ChatGPT Has Become Increasingly Unreliable Despite Detailed Instructions and Repeated Corrections

I'm a ChatGPT Plus subscriber, and I've been using ChatGPT extensively to help manage household projects, develop a small creator business, and build organizational systems.

For approximately the first two weeks, the experience was genuinely useful. ChatGPT helped me think through complicated projects, organize information, and develop workflows.

However, over the past several days (particularly November 3–6, 2026), its performance has deteriorated significantly. What started as occasional mistakes has become a nearly constant problem.

The recurring issues

1. Failure to retain and apply established context

ChatGPT repeatedly overlooks information I've already provided, sometimes many times. It reopens decisions we've already made, contradicts previous instructions, and makes recommendations that ignore the underlying reasons for the project.

This occurs even when the relevant information is available in the conversation or saved context.

2. Reactive rather than proactive reasoning

This is perhaps the most significant problem.

Instead of analyzing a task, anticipating problems, and evaluating solutions, ChatGPT frequently follows my immediate instructions without considering whether the proposed approach will actually work.

It often identifies obvious problems only after I recognize them myself and ask exactly the right question.

3. Repeated corrections without lasting improvement

When I identify a mistake, ChatGPT generally acknowledges it, accurately explains what went wrong, and promises to approach things differently.

However, the same behavior frequently returns within the same conversation or the next one.

I've spent an enormous amount of time repeating instructions, correcting assumptions, and troubleshooting the assistant itself.

4. Difficulty confirming or using High thinking

I've tried selecting High thinking to improve the quality of reasoning. However, the interface doesn't display the model information where ChatGPT repeatedly tells me to look.

I've provided screenshots demonstrating this, but it continues giving instructions that don't match my interface.

I cannot verify whether the selected thinking setting is remaining active or changing between conversations.

A concrete example: The Goodnotes project

I have aphantasia, meaning I cannot voluntarily visualize mental images. I've explained this repeatedly because it significantly influences how I organize information.

I wanted to build a visual content management system where I could see the balance of content pillars, tones, and formats across an entire month.

ChatGPT helped design a color-coded Post-it system and Goodnotes calendar. We spent hours creating templates, code maps, reusable elements, and learning how to use them.

Throughout the process, I repeatedly questioned whether the system was becoming too complicated.

Only after spending hours implementing it did I specifically ask ChatGPT to consider my iPad Pro's screen dimensions and the size of the templates it had created.

It then predicted that the visual codes would likely be too small to distinguish reliably on a full-month calendar.

That was a fundamental design consideration that should have been evaluated before recommending and building the system.

Even more frustrating, ChatGPT had repeatedly misunderstood the system's central purpose, despite my explaining that the entire point was to compare multiple visual codes simultaneously.

The prediction about readability was an estimate, not a confirmed failure of the design. But that analysis should have happened at the beginning, not after hours of work.

Why this is so concerning

I'm not expecting perfection, and I understand that AI can make mistakes.

What I need is an assistant capable of consistently applying established requirements, evaluating a proposed approach before recommending it, recognizing technical limitations, and exercising independent judgment.

Instead, the burden of analysis and quality control has increasingly shifted to me.

The assistant often gives an excellent explanation of its mistakes afterward, but those explanations don't translate into reliable improvements.

I've tried providing detailed instructions, organizing work into Projects, starting new conversations, explicitly requesting deeper reasoning, and repeatedly correcting the same behavior.

The amount of time I've spent managing ChatGPT is now exceeding the time it saves me.

What I'm hoping to learn

\- Has anyone else experienced a noticeable deterioration in ChatGPT's consistency or reasoning?

\- Is this potentially related to long conversations, Projects, context handling, model settings, or something else?

\- Is there a reliable way to make ChatGPT evaluate feasibility and identify potential problems before beginning implementation?

\- How can I verify that High thinking is actually selected and being used?

\- Are there settings, workflows, or troubleshooting steps that have genuinely resolved similar issues?

\- At what point does it make sense to try a different AI assistant?

I'm especially interested in practical solutions rather than simply being told to write better prompts. I've already invested considerable effort in detailed instructions, and the assistant frequently demonstrates that it understands those instructions before failing to apply them consistently.

My central concern is that I need ChatGPT to reduce my cognitive workload, not create another system that requires constant supervision and troubleshooting.

Has anyone experienced this, and what actually helped?

document fetched 2026-11-08 00:19 · canonical https://reddit.com/r/ChatGPT/comments/1xpx9pc/silent_amnesia