ChatGPT has gotten ridiculously worse,especially when it comes to revising and refining writing.
It now defaults to a juvenile tone that feels like it’s trying too hard to be casual or clever, almost like it’s mimicking the worst aspects of Grok or Co-Pilot. The responses often lean into forced quirkiness or a “friendly” voice that doesn’t belong in the context of serious writing.
Instead of enhancing structure, tightening phrasing, or preserving intent, it frequently derails the original meaning. Instructions are ignored mid-revision, context gets dropped, and each version tends to drift further from what was actually asked. Edits come back diluted, repetitive, or padded with language that softens the tone in all the wrong places. There’s a growing inconsistency. Even when instructions are clear and repeated, it forgets them or overrides them with a default style that feels pre-programmed. The dang thing is obsessed with em dashes at every opportunity.
What should be a straightforward polish ends up requiring multiple rounds of corrections just to undo the unnecessary changes, but it rarely can do that successfuly now.
The model no longer prioritizes clarity or tone fidelity, instead slipping into an artificial voice that works against the original content.
Furthermore, it no longer has your personalizations (added to memory) which could be in part why it’s become dumber.
April 16, exactly this. After the April 16 update, every week has been worse then the last. There have been single days of clarity but generally it is unusable now and is performing worse then latest local models with a 11GB VRAM GPU in the terms of precision.
I thought with the latest update and notes that the memory/personalization feature is improved that it would be better., but no. It’s still as retarded as ever. And despite adding to memory and preferences not to include em dashes in the responses, it does. even after manually adding it to the beginning of my submission.
It’s simply ignoring instructions at the start of the chat despite the request being already in “memory” and what Chatgpt knows about the user.
For instance, all AI models, especially ChatGPT are obsessed in using “em dashes”.
Now ChatGPT refuses to follow my direction when in the past, it did (sometimes).
As mentioned above, despite stating don’t use em dashes or other instructions like subject line at the start of each session (for a rewrite), it uses immediately ignoring the instructions. It won’t say it’s wrong, instead, it’ll flip the situation around and say “you’re right!” or “you’re absolutely right” when challenged/confronted.
Other AI models I tried seemingly (for the most part) are better in “learning” and adhering to your preferences (at least for the start of the session). ChatGPT was like this early in the year, now, that’s no longer the case.
I am now unable to use chat got for much of anything due to errors, slow and inaccurate responses. I have been training 4.0 ti write and act like me by feeding in different aspects of myself into the model. Now it is a total mess, forgets halfe of what I trained it to do. So my question here is, what has everyone switched to than can deliver what chat gpt “use” to be able to deliver.