#Very slow GPT-4 responses and failing to complete the response

1 messages · Page 1 of 1 (latest)

static timber
#

Bug Report:

When submitting a message in GPT4 on the UI, the responses back are extremely slow and they are largely not completing showing a network error or error in message stream.

Steps to reproduce:

  1. Submit any message into GPT4 UI
  2. Response is forming slowly and then stops with failure message.

Expected result:

  1. Response is performant and is able to finish without errors.

Actual result:

  1. After submitting a message the response fails after about 10-20 seconds and shows red text and either ‘network error or error in message stream’

Additional information

Browser: Can recreate on mobile safari, and on chrome via MacBook.
OS: Ventura 13.6.3 and Sonoma 14.2.1 and chrome version 120.0.6099.109 and 120.0.6099.216 both the same issue. Tried on multiple devices and networks exhibiting the same behaviours. iOS 17.2.1 iPhone 12 Pro.

static timber
#

It’s actually getting worse 😭

strong olive
#

yes, any long message that I'm doing or work that I'm doing consistently, it's always cutting off.

pseudo python
#

Tried to submit yesterday, but i am not allowed to post here, even as a paid customer

waxen moat
#

I have been experiencing this for about two weeks. It will complete about half a response, and then give a network error with red text.

static timber
#

Exactly right @waxen moat @strong olive 👍

rapid crown
#

I'm experiencing this issue now.

safe widget
#

For me, this has been happening for over a day - after the code formatting problem (which is still unresolved). It often requires between 4 to 8 retries to get a message to complete (and even then it is incorrectly formatted). It's a good thing I know what ChatGPT 4 is capable of because if this was my first experience with it - I'd never pay for it.

raw delta
#

I've been getting this for the past 5 days. in that time, i think less than 50% of my responses are coming through ok. everything else errors out.

fierce pivot
mild parrot
#

I am having the same issue. Less than 50% of my responses are coming through without error. I also believe that some of the re-submissions are counting toward my daily allotment. I first noticed the issue three weeks ago; it has only gotten worse. If i try to upload a PDF the error rate goes WAY up.

high reef
#

I'm experiencing similar.
Browser: Firefox 121.0.1 (64-bit)
OS: Ubuntu 20.04.6 LTS

sullen eagle
finite void
waxen moat
#

I've switched back to 3.5 and am unable to use any GPTs.

finite void
waxen moat
snow fjord
#

Almost every prompt I entered today resulted in "network error

Screenshot_2024-01-21_at_12.42.17_PM.png
pulsar ember
proper dagger
#

This is really starting to upset me. I have a low failure rate but the failures still count towards our quota, which is bs. My responses through browser are so slow that the browser prompts me that the page has stopped responding and I can't even scroll through the conversation or type in the text box. My page is frozen for 5-10min for each query. I'm really starting to get pissed off that we're paying customers and this issue continues to be ignored, we don't even get the courtesy of an acknowledgement. They need to redo the webpage, upgrade the servers, and/or prioritize paying customers. If gpt4 itself is being overloaded they shouldn't freeze the whole page waiting for a response. Also, accessing gpt from the app doesn't seem to have such a dramatic problem so it seems like a server issue and for some reason they have the webpage designed to halt all activity while it's waiting for a response and that's about the dumbest approach they could take.

sullen eagle
proper dagger
sullen eagle
proper dagger
#

The company is run by socially childlike brains. I'm pretty sure they throttle us and rather than just delaying the response they halt the whole page. And to make matters worse the response is already generated because the text will start typing on the screen so they send us the responses then make our system freeze up for the delay period so we can't even read it or scroll through previous queries. I wish they would just admit they have no choice but to throttle us because of resource limitations and let the response come through and just put a visible timer on the submit button. It would be the civilized adult way to handle it. Instead they think it's better to deceive and that we're too stupid to notice. They're probably throttling based on usage and complexity as well and I'd be willing to bet most of the people with this issue are computationally demanding users. There will probably be a class action about this in a couple of years.

finite void
#

After the 10th regeneration attempt (because of errors generating a response) i told my custom gpt: "tim i beared so much i have become a bear". his answer:

DALLE_2024-01-22_18.15.08_-_A_whimsical_scene_depicting_a_person_who_has_transformed_into_a_bear_due_to_bearing_so_much._The_bear_is_sitting_in_a_comfortable_living_room_surroun.png
#

Here is a whimsical illustration of the scene you described: a person who has transformed into a bear due to bearing so much. The bear is comfortably situated in a living room, surrounded by family activities, wearing a cozy sweater and holding a cup of tea, encapsulating the humorous and light-hearted essence of your concept.

waxen moat
static timber
placid rose
#

This week and last week no problems until now there is problem

image.png
#

used some image with instruction to transcribe some notes

proper dagger
placid rose
#

idk it seems to be annoying like a casino or you have full generation or some fail

waxen moat
#

Per another thread, I disabled IPv6 on my AT&T modem, and it looks promising. It appears the problem may be wholly client-side and caused by AT&T's poor IPv6 performance.