All the problems being reported about GPT-4 are correlated:
- The decreasing speed
- The early cut off during responses
- The server errors where the model gives up completely and doesn’t respond at all.
The slower it gets the earlier it cuts off, which compounds the issue because we’re all eating through our usage cap quicker for the same amount of generated content. What’s worse is that, for the past two days in my case, it’s gotten to the point where it won’t even generate a response due to server errors, and even then the error message eats into the already low usage cap. All of these problems have the same root cause and it’s not user error, browser settings or anything else. The service is simply unable to deliver at this point.
I just hope it gets fixed because at this rate the subscription increasingly seems like wasted money.