I encountered a significant issue with GPT-4’s user experience. The system incorrectly told me I could use another model (“4o mini”) when the model was actually blocked, and I was unable to send any further requests. Unlike when I’ve attached files or generated images, which explicitly state I must wait due to those actions, here it simply presented the option to switch models without clarifying the actual cause (which was the block).
I also faced issues related to artifact/canvas creation, where GPT-4 generated something without my request and started attempting to write a document, failing repeatedly. Even after asking it explicitly to stop creating “artifacts,” it continued trying to process the incomplete artifact, failing several times. The system kept reading and attempting to complete it, ignoring my instructions, and failed to understand what an artifact was.
Several other behaviors (for a GPT-4) were problematic:
It began from the point where the artifact failed, disregarding my detailed instructions.
It failed to maintain consistency, even after I requested it not to lose information or become inconsistent.
After I requested no more artifacts, it misunderstood my request and created another incomplete one, not adhering to my guidelines.
This sequence of issues made the entire chat unrecoverable, and I suspect the artifact updates were the primary cause of the repeated failures.
As shown in the screenshots, it tried to create an artifact multiple times, which likely consumed the entire free session. I understand it has been recently updated, and this feature is in beta. But the text is what bothers me the most:
“You have reached the limit of the Free plan for GPT-4.
Responses will use another model until the limit is reset after this time…”
Because it is rather trivial, as is often the case with those small, irritating QA oversights.
Although I encountered another related, frustrating issue that I will detail in the following section.
Odyssey of trying to report
Initially, I selected 
‘poor response,’ then ‘more,’ and submitted several comments, including ‘did not fully follow the instructions.’ In the ‘other’ section, I described the UX issue as best as I could at the time, given my frustration.
The help button does not allow me to report. It also does not start any conversation.
GPT-4 mini suggested outdated or things that have changed.
The support site circularly points to the use of a help button to start a conversation.
When I searched for support team contact, I found a GPT-4 support chat, but it was locked due to the Plus subscription.
And then, I was here, in this forum.
I believe the issue is clear, and this must have been the experience of many reading this, so this message is dedicated to you all.