[OPEN] ChatGPT Broken (Images, Files, Web Search, Long Msgs, etc.)

[gunnerjnr]

Hi all,

The Issues:

As the title suggests, ChatGPT is broken. I have been having some weird behaviour since yesterday (02-12-2024) on both the web UI version and the Desktop version (Windows); oddly, it still works on mobile (iOS). I am situated in the U.K.

I cannot paste longer messages (even though they are inside of the max token limit); I get this error message:

Note: Those with a keen eye will note it says 2/2 messages. What’s odd is that it worked fine for the first message (and, in fact, for the past few months, as it is a reusable prompt) but then decided all of a sudden it was too long.

If I split the longer prompt up into 2 messages, then it completely ignores any previous messages; it also ignores memory and custom instructions.

I can’t read any file whatsoever; if I upload even a simple text file saying ‘Hello’ for example, then it just gives some message about being unable to read files.

I can’t use the search feature for anything, and it tells me it cannot search the web:

I can’t create images; same kind of result as above:

And I have no way to contact live help, as any option I click in the chat gives an automated response from a bot that seems to just completely vanish after a page refresh.

So I am paying for a product that I cannot use, and have no way to report it. So these forums are my last hope. I have also had to purchase a month of Claude AI so that I can at least get some work done until OpenAI can hopefully solve the issue.

Things I have tried:

• Clear Cache
• Clear Browser History
• Log out all devices
• Multiple Browsers: (Chrome, Firefox, and Edge) - This also includes incognito windows.
• Reset network and flush DNS settings
• Disable VPN
• Disable Anti-Virus
• Add web address to VPN/Firewall settings to make sure it is allowed through.
• The Desktop app for Windows (same results as above)
• Mobile app (iOS) - Surprisingly, this is the only version that works, but it’s not very practical to try to work from mobile.

Chrome Console Logs (uncertain if these are directly related, but I will just paste them all):

Plain Chat Window Error/Warning Logs:

Uncaught Error: Minified React error #418; visit https://react.dev/errors/418?args[]= for the full message or use the non-minified dev environment for full errors and additional helpful warnings.
    at $a (d7gqhmxyzfulu8ms.js:24:30728)
    at x1 (d7gqhmxyzfulu8ms.js:24:75176)
    at Sv (d7gqhmxyzfulu8ms.js:24:114974)
    at oh (d7gqhmxyzfulu8ms.js:24:114947)
    at Uh (d7gqhmxyzfulu8ms.js:24:114814)
    at yv (d7gqhmxyzfulu8ms.js:24:110696)
    at Ev (d7gqhmxyzfulu8ms.js:24:120306)
    at MessagePort.B (d7gqhmxyzfulu8ms.js:9:1537)

howrtlj6kg41ou0a.js:1 Intercom not booted or setup incomplete.

ga842uuse918os7p.js:52 [DatadogProfiler] Timing chatgpt.web.pageLoad.render_model_switcher already logged

---

Other Logs/Warnings (such as those for the actions taken that are mentioned above):

1. Send a longer message:

howrtlj6kg41ou0a.js:1 Intercom not booted or setup incomplete.

2. Even though I am logged in and using 2FA, when I try to upload the text files, this message appears in the console:

client_id and scope must both be provided to initialize OAuth.

3. This is the message when trying a web search:

[DatadogProfiler] Timing chatgpt.web.pageLoad.render_model_switcher already logged

4. Try to generate an image:

ga842uuse918os7p.js:52 [DatadogProfiler] Timing chatgpt.web.pageLoad.render_model_switcher already logged

Other than that, I am at a loss, and the software is in an unusable state.

EDIT:
I also forgot to mention that it doesn’t seem to have any effect when switching models either, and comes up a warning about hitting the limit for GPT4 requests, even though the selected model is 4o; I tried 01-mini, 01-preview (which no longer has the thinking process, just instantly outputs a result), and 4o-Canvas (which again, doesn’t have a canvas, just outputs a result instantly). So, I think the models don’t work either.

Further Edit:

A couple of new errors have come to light:

Uncaught (in promise) RequestError: Something went wrong. If this issue persists please contact us through our help center at help.openai.com.
    at br.fetch (lrt789z2rawt9v6n.js:267:23754)Understand this errorAI
chatgpt.com/:1

Uncaught (in promise) RequestError: Something went wrong. If this issue persists please contact us through our help center at help.openai.com.
    at br.fetch (lrt789z2rawt9v6n.js:267:23754)

Kind regards,

[e6524554b610819b9c67]

Same Here, already 6 days without OpenAI Team serious respond.
please @OpenAI what do you do ?

[gunnerjnr]

@e6524554b610819b9c67 Wow, 6 days with no response already?

That is very poor on their part. I guess I won’t get my hopes up just yet about any fixes, then.

[gunnerjnr]

It seems that since a fix was put into place during the early hours for the: ‘broken images’, everything else has started to work again.

I have tested and verified each issue from my main post, and ALL seems to be resolved.

---

Image Generation:

• - Tested and Working as of 07:33 am (GMT) 04-12-2024
  
  

---

File Uploads:

• - Tested and Working as of 07:33 am (GMT) 04-12-2024
  
  

---

Web Search:

• - Tested and Working as of 07:33 am (GMT) 04-12-2024
  
  

---

o1-Preview:

• - Tested and Working as of 07:33 am (GMT) 04-12-2024
  
  

---

4o-Canvas:

• - Tested and Working as of 07:33 am (GMT) 04-12-2024
  
  

---

Larger Prompts:

• - Tested and Working as of 07:33 am (GMT) 04-12-2024
  
  

---

Kind regards,

[virap1]

This is not resolved for me. Plus account, still getting ‘I cannot generate actual images directly’… error. This has been going on for 3-4 days.

[gunnerjnr]

@virap1,

You are correct; I have reopened the thread and reversed it from being marked as a solution.

It worked for about 1 hour this morning, then it stopped working again completely. I have since cancelled my subscription and will most likely use other AI providers moving forward.

Maybe I will try the API from time to time. I hope they manage to solve it for everyone.

[virap1]

They fixed it yesterday, but today is again broken and not working at all for me.

[indiefieldnotes]

Imagine buying an $1,800 subscription (Team, 60 profiles * $30) only to discover that chat model selection is broken in WebUI on all Windows and Linux systems. As a result, it rolls back to the GPT-4-Turbo model, which can’t handle files, images, or web access and provides subpar answers.
We have 22 users who work on Mac, and there is no issue there. For the others, the chat simply doesn’t work.

The model_switcher and rollback to gtp-4-turbo are the root of the problem.

Since 26.11 to this day, there has been no fix, not even an acknowledgment of the problem. The subscription is about to renew. It seems we won’t be renewing it either.

Edit:
I’ve asked it about knowledge cutoff date and now I think it’s gpt-3.5-turbo

[e6524554b610819b9c67]

@OpenAI please fix !

we dont deserve this

[virap1]

This is simply unacceptable

[momo0505]

check this bro

GPTs unable to read and process PDFs in messages anymore

Hello Boss , This is part of an email I sent earlier to the OpenAI team, in which we organized several issues and captured screenshots for each of them. We also recorded a video using NVIDIA’s screen recording software. In the process, we discovered that OpenAI had replaced the models for some Plus users who were experiencing issues. You can refer to the video link and images I have provided to understand the situation. Although it is unclear whether this is intentional on OpenAI’s p…

[jjhuang0087]

So all the computing power resources and multimodal capabilities have been given to the new version of Pro?

[gunnerjnr]

Hi @momo0505

I completely agree, I have since been using Claude, but here is part of what I sent to OpenAI support in an email after I rage quit the other day:

Sorry, but it’s still in an unusable state, it doesn’t seem to use the correct models when selecting them, ignores custom instructions, memory, previous messages, even if you try to correct it in a follow up pointing out exact issues and changes, it just completely ignores them.

You get a quick flash on refresh of it jumping from 4o-mini to 4o (even though 4o is selected) but then get a warning that you’ve hit gpt-4 limits, so all messages moving forward will use the 4o model until the limit resets, even though 4o was the model selected and not gpt-4.

The o1-preview used to think for a good 30, 60, 90, 120+ seconds and produce great results but now it barely thinks for 8-10 seconds, and outputs complete rubbish as well as ignore 99% of the instructions and guidelines in the prompt.

It’s almost as if all the models are secretly switching to 4o-mini in the background.

I had to use Claude for the entire day yesterday, as I will have to do again today, and any other day until the issues are found and resolved.

Also, other users have been commenting in the thread I linked about things not working for them, so it might be worth a look and to start investigating.

So we shall see what they come back with, its usually 1 -2 days for a response.

Kind regards,

[momo0505]

i already did this i can sent you our email

[momo0505]

hi@gunnerjnr

I can take screenshots of all our email correspondence and send them to you. If it’s convenient for you, please provide me with your email address. If you’re concerned about your personal privacy, you can register a completely new email address.

Kind regards,

[hassan164]

Maybe upload them on a drive, so we all can access?

[momo0505]

hi

you mean here?

Kind regards,

[hassan164]

Yes, and sending the link here.

[momo0505]

@hassan164 @gunnerjnr

I don’t need to put the first email on it. It’s an ai reply. All of them here are open ai’s real person replies.

1 open ai：

2 us：

3 open ai：

———————————————————————————
4 us：

———————————————————————————
5 open ai：

———————————————————————————
6 us：

———————————————————————————
7 open ai：

———————————————————————————
8:us：

———————————————————————————
9 openai：

———————————————————————————

Actually, in the end, I was still very sad because from my communication with OpenAI, I felt that they didn’t realize this was their own problem, nor did they provide a positive response to the issues I raised in my email. Although they eventually offered a positive resolution with a refund, I was still very disappointed with them. In my view, they didn’t have an attitude of wanting to solve the problem, nor did they seem interested in resolving it. Instead, they seemed more focused on ‘dealing with the person who raised the issue.’

Ps: I completely followed the requirements of the other party and used NVIDIA’s screen recording software (Geforce Experience) to record the video.

[virap1]

I canceled my membership and moved to claude until open ai decides to take major issues like this seriously.