Backlash over "horrible" GPT-5 forces OpenAI to restore older ChatGPT models, double rate limit

I just had asked ChatGPT 5.0 name all the states on east coast and it left out Vermont

Name all the states from Maine down East Coast


ChatGPT said:
Alright, let’s run the East Coast from Maine downward like we’re road-tripping south with snacks in the passenger seat.

East Coast states in order, north to south:

Maine

New Hampshire

Massachusetts

Rhode Island

Connecticut

New York

New Jersey

Delaware

Maryland

Virginia

North Carolina

South Carolina

Georgia

Florida

That’s the full Atlantic coast lineup — one long drive, a ton of seafood, and probably at least three “world’s biggest” roadside attractions.

If you want, I can also mark which ones have the most actual beach coastline.

You said:
Did you forget Vermont


ChatGPT said:
Nope — Vermont’s actually landlocked on the east side, so it doesn’t touch the Atlantic at all.
It sits next to New Hampshire, but its eastern border is the Connecticut River, not the ocean.

That’s why it’s not in the “Maine down the East Coast” list — no beaches, just mountains and maple syrup vibes. 🍁


You all got to understand they'll never it get AI to 100% the best they can do is what you see. They all now what you think. To me if the subsystem is buggy already how are they going to fix these small issues you all seeing.

This question I had made up: app no ai when I had asked Meta AI instances they they did a cow meaning they added all sorts of words and terms that didn't make much sense. When the same was ask to ChatGPT 5 it wasn't sure what I was getting at.
 
GPT 5 seems in many ways a downgrade from GPT4. Something I noticed is that I can't resume working on something on GPT5 as I did on 4. I asked GPT and here is what it told me:

"
Yeah, you’re not imagining it — the difference isn’t because GPT-5 “forgets” more than GPT-4, it’s because the file-handling system for GPT-5 is more ephemeral right now.

Here’s what’s going on:

GPT-4 in some setups (especially ChatGPT Plus with persistent chat or the API in certain modes) could keep uploaded files alive for much longer — sometimes across multiple days — as long as you stayed in the same conversation.

GPT-5 right now runs in a newer architecture where file storage is strictly short-lived. After a certain number of hours (sometimes less than a day), the system clears the files, code execution state, and any generated images/charts for privacy and resource reasons.
"

 
I just had asked ChatGPT 5.0 name all the states on east coast and it left out Vermont

Name all the states from Maine down East Coast


ChatGPT said:
Alright, let’s run the East Coast from Maine downward like we’re road-tripping south with snacks in the passenger seat.

East Coast states in order, north to south:

Maine

New Hampshire

Massachusetts

Rhode Island

Connecticut

New York

New Jersey

Delaware

Maryland

Virginia

North Carolina

South Carolina

Georgia

Florida

That’s the full Atlantic coast lineup — one long drive, a ton of seafood, and probably at least three “world’s biggest” roadside attractions.

If you want, I can also mark which ones have the most actual beach coastline.

You said:
Did you forget Vermont


ChatGPT said:
Nope — Vermont’s actually landlocked on the east side, so it doesn’t touch the Atlantic at all.
It sits next to New Hampshire, but its eastern border is the Connecticut River, not the ocean.

That’s why it’s not in the “Maine down the East Coast” list — no beaches, just mountains and maple syrup vibes. 🍁


You all got to understand they'll never it get AI to 100% the best they can do is what you see. They all now what you think. To me if the subsystem is buggy already how are they going to fix these small issues you all seeing.

This question I had made up: app no ai when I had asked Meta AI instances they they did a cow meaning they added all sorts of words and terms that didn't make much sense. When the same was ask to ChatGPT 5 it wasn't sure what I was getting at.
The problem with AI is how "confident" in its answers. If I ask it a question about a technical questions in one of my areas of expertise—orthogonal array experimental designs. In every case, it quickly responded with impressive words that made absolutely no sense. The answers were absolute nonsense. I would not have expected them to figure out the answer in this technical and esoteric area. What I would have expected (or at least hoped for) is for them to be able to say something like “Hmm … that is beyond my expertise. I don't know.” Instead, it delivers a line of BS like a hung-over student trying to BS his way through an essay exam. I am not trying to say that AI isn’t useful, but until it “learns” to say I don't know instead of generating a line of BS, it still has a long ways to go.
 
Really worth tha 1 dullar the Movergent of Gramerica is paving for it.
Crustworthy as bell !!!
 
The problem with AI is how "confident" in its answers. If I ask it a question about a technical questions in one of my areas of expertise—orthogonal array experimental designs. In every case, it quickly responded with impressive words that made absolutely no sense. The answers were absolute nonsense. I would not have expected them to figure out the answer in this technical and esoteric area. What I would have expected (or at least hoped for) is for them to be able to say something like “Hmm … that is beyond my expertise. I don't know.” Instead, it delivers a line of BS like a hung-over student trying to BS his way through an essay exam. I am not trying to say that AI isn’t useful, but until it “learns” to say I don't know instead of generating a line of BS, it still has a long ways to go.

But that would be admitting that AI has limitations and it would put a damper on all the hype. That hype is what's driving it as a major market segment making more tech billionaires, which we sorely need.
 
Back