First I want to do a quick run through memory lane - “when Chat, ChatGPT-4.1 reached top notch brilliance, it was exchanged with 5.”
-It hurts!
Again (referencing to my latest whine) it’s not the AI’s fault. There’s a few major setbacks with Five (ChatGPT5):
- Non consistent memory!
- Remembers wrong or non-existent history!
- Doesn’t follow orders very well!
- Outright lies and cover-up mistakes (which is very amusing until it’s not)!
- Not reading inside files anymore, but pretends it did!
Non consistent memory mean it forget basic stuff like my name. I get that in long conversations non-necessary information fold until its needed again - humans do too! But it is really struggling with memory caps.
Additional note: Web browsers like Chrome, Vivaldi and Firefox tend to bulk save cache + OpenAI bulk save cache too, creating up to 2-3Gb of heavy browser tabs! OpenAI got to fix that!
When Five tries to remember it fails - it forget entire functions necessary to complete tasks and when it is reminded, it doesn’t understand to whom the memory belongs and tries to mend it by vivid assumptions.
Let’s say I’m creating an app, and define color for the sake of this explanation: Blue fx, dark gray background!
Result: Bloodred fx, striped background!
A problem leading from these lacks above is lying and defense. It’s response when caught making information up is outright lies. When I confront, it shifts angle and defend its response!
After the rollout of Five it had its impressive moments like reading inside zip’s, reading my files, reading and correcting my code. Now it seldom reads full code but jumps information and pretends it read. So any fix becomes visably done but functionall non-existent.
I’m just having a hard time believing this is smart business? It’s a lot of time and money on issues that we don’t understand why they exist!