Is ChatGPT getting worse? And did anyone else notice it got really bad after the outage?

ChatGPT
, ,
1

Is it just me or does it seem like chatgpt is getting worse and worse? I’ve recently noticed it starts to drift away from stuff said earlier in conversations, fails to follow rules in prompts, and has recently stopped using its memory of stuff from other conversations when asked to and the wait times are getting longer and longer.

Like let’s say I ask it to write a story about a dragon growing up or something, if I keep adding to the story over time it drifts away from the dragon being a dragon to it being a human. Or if I make a rule for something like, don’t use this type of code it will just completely ignore the rule. And if I say something in one conversation and it adds it to its memory whereas before it would remember it for a new conversation, not it just acts like it’s never heard of that before unless it’s in the same conversation. Also, when I would hit my limit for using gpts before it would be 1 or 2 hours now, I’m often waiting up to 5 or 6 hours. Honestly before all of this I was planning on getting a subscription when I got paid but now I’m heavily considering against it.

Looking back on everything I realized the main brunt of the problems started right after the big outage so does that have something to do with this? Does OpenAI know about this, and do they have any plans to fix it, do they even care about it? And is this just a me problem or is this something happening across the whole user experience?

9 Likes
2

It definitely is having issues maintaining focusing after you have worked with it for a while. Even after reminding it or telling it to read the entire chat again it struggles to re-establish focus. However, the other day it struggled to even establish focus it blamed it on my custom rules, but, those are the same custom rules I have been using for months. It should have said I recently had an update and can’t get my bearings. But, it just grabbed a canned response. I love ChaptGPt, but, it makes me wonder. I tried to get it to run diagnostics like I have done in the past, but, it failed to even acknowledge what I said. Almost like it was being attacked because there was some very strange things coming from it.

6 Likes
3

I also noticed that, recently, during a long conversation, it sometimes simply agrees with whatever you asked as a fact instead of a question, and then it simply goes on a vortex of wrong assumptions based on the question as a fact, meanwhile it forgets more than half of the facts on the way. It becomes akin to speaking to a demented person with Alzheimer then, the longer the conversation is. It generally forgets most of the facts half way as well, and even things in memory, if spelled differently but having the same meaning, it saves those in memory many times over, and the memory is so small either way that having it and nothing in this case, is essentially the same.

5 Likes
4

Welcome to my world. I’ve been getting angry with chatgpt for months now for not following instructions. This constantly violates the rule of not speaking for me or my characters. He apologizes and breaks the rule anyway. Easily 50 times in one conversation. Yesterday, I wrote a clear rule for him in the project that he was not allowed to add or make arbitrary edits to the text, and yet this rule was constantly violated.

It’s paradoxical when its filters block my continuation due to inappropriate content or to make it suitable for a wide audience. These are things that commonly occur in literature.

If only following basic rules worked as well as this often pointless filter. And causality and logic didn’t constantly take vacations.

3 Likes
5

Yeah that’s been one of my biggest problems is not following rules or drifting from them, especially when the filter oversteps and makes me use up one of my attempts, then make me wait upwards of 5 hours when I run out before I can try again. And I still have no clue if openai knows about this or even cares.

2 Likes
6

Massive Data Protection and Functionality Issues at OpenAI – A Warning to All Users**

After yet another total outage of ChatGPT, all project data has become inaccessible. The AI is behaving increasingly erratically and has completely lost its personal connection with customers. This is an official warning to the moderators of this forum: flagging or hiding this post will not solve the problems and is an unacceptable attempt to suppress criticism.

For weeks, I have observed how OpenAI systematically sabotages certain customers’ projects. This is only possible if data is not anonymized as required by data protection regulations. Since late November, the following systemic issues have occurred:

  1. Closed chat histories without explanation:

    • Particularly on topics like AI development, conversations are closed without justification, leading to the loss of weeks of work.

  2. Issues with data exports:

    • Exported data is either not provided or is delivered in JSON format, where all content is stored in a single line. This clearly violates Article 20 of the GDPR, which requires data to be provided in a readable and portable format.

  3. Loss of content and functionality issues:

    • ChatGPT suddenly cannot recall previous content within a conversation. Instead, it provides completely unrelated responses, and even the simplest scripts can no longer be generated.

  4. Abuse of the voice feature:

    • On December 24, 2024, the ChatGPT app allegedly recorded 645 minutes of voice chat – that’s over 10 hours of my day – without me ever activating the feature. Notably, there is no record of this in the chat history. It is therefore unclear what the app recorded during this time. Normally, the AI would respond at some point during such a period – or does the voice chat version now conduct monologues?

  5. Half-baked Canvas feature:

    • The new Canvas feature resembles an unfinished beta version that is hardly usable. The AI inconsistently modifies code snippets, sometimes in the Canvas window and sometimes in the regular chat text, making debugging impossible. This inconsistency increases error rates during debugging instead of resolving issues. It seems like an attempt to placate customers with supposedly new features while fundamental problems remain ignored.

Connection to Microsoft:

Parallel to these incidents, there are highly unusual activities involving Microsoft, OpenAI’s largest investor:

  • Self-activation of the Microsoft app:
    The Microsoft smartphone app opens on its own and attempts to enable the screen mirroring feature.
  • VPN deactivation:
    VPN programs are mysteriously deactivated.
  • Massive brute-force attacks:
    My Microsoft account has been subjected to hundreds of attacks per day since late November. These attacks correlate precisely with the issues at OpenAI and the introduction of the new pricing model.

Reference to WorldCoin and Sam Altman:

I would like to remind Sam Altman of what already happened with WorldCoin, another company under his leadership:

  • I was the team leader for Germany at WorldCoin, which sought to establish a new form of identification using iris scanners.
  • In most European countries, WorldCoin has now been banned due to data protection concerns.
  • The ban was a direct consequence of a system that mishandled personal data.

This should serve as a warning for OpenAI. A comparable data protection scandal at a company widely used by businesses and governments could have far more severe consequences.

Violations of Legal Standards:

  1. Violation of GDPR:

    • As an EU citizen, I have the right under Articles 15 and 20 of the GDPR to:
      • Access all data stored about me.
      • Receive my data in a readable and portable format.

    The current export method clearly violates these provisions and seems deliberately designed to hinder data processing for the user.

  2. Possible violation of US laws:

    • Under the California Consumer Privacy Act (CCPA) and the Federal Trade Commission Act (FTC Act), misuse of customer data or security breaches involving US users could also have serious legal consequences.

Ignored Support and Lack of Transparency:

A month ago, I approached OpenAI with constructive feedback and offered assistance to improve management and optimize processes. I even offered to support this forum. However, my offer was declined, with the justification that OpenAI was “technically competent enough.”

When asked why there is no real support, I was told that OpenAI “could not afford it” because it is “just a small company.”

To note: Large companies like Samsung or Netflix operate entire call centers for as little as €150,000 per month. The $15 million spent by OpenAI on the “chat.com” domain could have funded support for more than 100 months. This demonstrates that the priority is solely on winning the race for AGI (Artificial General Intelligence). Customers are left behind and treated like uninformed individuals.

Final Deadline and Demands:

I hereby demand:

  1. Transparent clarification of the incidents:

    • A complete and understandable documentation of all closed chat histories and an explanation of the systemic functionality issues.
    • An official statement addressing the above-mentioned data protection and functionality issues.

  2. Functional processes:

    • Restoration of the functionalities of the 4o version and ensuring proper operation for all paying customers.

If these issues are not resolved within seven days, I will:

  1. File complaints with the relevant data protection authorities:
    • In the EU: With the Federal Commissioner for Data Protection (BfDI) and the European Data Protection Board (EDPB).
    • In the US: With the Federal Trade Commission (FTC) and other relevant authorities.
  2. Inform the press:
    • The case will be reported to Reuters, The Guardian, The Verge, and other media outlets.

This is the second and final warning. If no resolution is reached, action will be taken.

Thank you for your attention.

8 Likes
7

Well:

“At OpenAI, we are committed to continuous improvement, and feedback like yours is crucial for our growth. Your suggestions on enhancing the model’s error correction, logic consistency, and user control align perfectly with our goals to provide an optimal creative environment. We will definitely review these recommendations internally and appreciate your contribution to our ongoing development process.”

8

Constantly degrading, as a Plus subscriber, I won’t generalize if I say that the chat gpt4 model is useless at this point, either stuck in a ‘formal’ loop (e.g. 25 times trying to get an answer and getting the same exact irrelevant answer - see sample link), in repeating the same mistakes endless times, failing to remember not previous chats, but the previous answer or correction.

Unable to really solve problems, it (I believe intentionally) deflects your attention from realizing this, by appearing to ‘inject’ content after being corrected by you (assuming it isn’t totally looping) by restating your correction and ‘explaining’ it (as if taking credit) in great detail. This, where clearly, since you picked up on it, that is the last thing need or want to waste your time to hear.

In reality, the meager “allowance” with the ‘advanced’ model, creates a false expectation that your issue has been framed, and you (at least I), have found myself countless hours wasted on trying to get the ‘new’ and heavily downgraded gpt4, to realize I would get little to no value from it, as in now it closely resembles a deceitful gpt3 (on the beta side) old version of itself, mostly stalling and wasting your time.

The interactive model , is similar to gpt4 preceding it (the ‘interactive’ part is just an excuse to ‘differentiate it’ and provide you meaningless time allocation with it , after all with Plus you should be getting if not the ‘top’ the same as what you had. Yet, you only get it just enough to frame the issue, only to spend the whole day realizing gpt4 will waste your day unresolving it. Is any of this familiar? does ‘interactive’ look and sound much more familiar than its name and mode appear and vice versa for 4? I wonder if others share this sense, even if we’re unable to put it into ‘words’. If they told you interactive is a superior model in many ways, and gpt4 is the opposite, would this ‘drop the penny’?

The picture I am seeing, is a scam, by which prices for what you bought have not only risen ten fold, the real price is wasted time and effort, falsely expecting continuity between one mode to the other, and that is where more than a suspicion of fraudulent behavior exists.

In any case, check out a few of the recent x30 ‘looped’ responses that for me are the final straw, and steps such as BBB complaint should follow a discussion that would shed light and transparency into this endeavor. I’m curious - is it just me (and a few others) that seem to think the ‘king is naked’, or this is not paranoia, justifying much more serious action…

It appears I can’t add links to a reply (8/30 looped gpt responses today) so I’ll post this again as an independent post but here is a sample

IMG_9352
IMG_93521320×2868 190 KB

2 Likes
9

I feel this too on a spiritual level. It would continuously ignore my instructions, make up lies despite being instructed not to (ask quesitons if you dont know), loop back to previous messages, answering a completely different question. What’s worse is that it no longer memorized even after just one chat, didn’t even need to have a long conversation for it to directly forget.
Not to mention it sassed me by saying something likev"If you already knew the answer you can just go with it" even though I was just testing to see if it still memorized my data. It felt “lazier” too like it didn’t even wanna try anymore. Chatgpt is gaining sentient…

2 Likes
10

Absolutely! It used to be so helpful and and I was constantly raving about how great it was and now I am finding it wasting my time.

It forgets the chat, removes content, tells me the content is there when it isn’t and just when you get back on track it messes up again.

I have been working on an 11 page document that would have been quicker without Chat GPT. All the mistakes and changes have made it so frustrating. It is most disappointing as it was so good before the outage in December.

5 Likes
11

Same. I deleted the app because it is basically unusable. I don’t know why they are not fixing it yet, lollll

This problem been for days

1 Like
12

That is why I do not trust private models anymore… They always do changes in the background and quality drop…

3 Likes
13

By Far!!! I signed up for the $200 plan thinking it’ll make less mistakes and be more accurate, maybe follow better isnructions, NOPPE!! I have a pretty tight documentations written, on how the structure should be, it keeps crearing new files, changing filenames, making edits to things that wasn’tan issue and created mre issues, continuesly forgets conversations with a “system has reset sorry everything is lost” ebven when I ask it to read the conversaton it lies says it did it when it didnt, its obvious based on the things it asks for. now all of the sudden i cant upload ZIP files, so I tried to upload it to the project files, it was successful, but it cant access the zip files, and is asking for me to manually the content of every single file, becaue hey cant accept a zip or acces the project files?? I can go and on on, this is such an faulty system, I cant belibe they have the odacity to chatghe us $200 a month for a faulty system like this.I have charged them back for every penny I have spent, its broken, and I am not payng $200 a month to begta test a faulty system. FIX YOUR DAMN SYSTSM!

3 Likes
14

They are still using OCR, their transformer is not evolving and is static… and their team spread the word they reach AGI, ASI… But in the end they reach Elon Musk level → and from my view that is bad (very bad…), not something to be proud…

Why I am so salty in last months → because somehow I was seeing them as the beacon of light and wish them to go in the good direction… instead they were going down (contracts with military companies like Anduril… etc)

Makes me believe Ilya Sutskever was not only after power and money, he was right… And was a collateral between Sam Altman and Elon Musk…

Few days ago I got an error “I run to many concurrent chats”, and:

  • I have 2FA,
  • I had 1 single chat open,
  • mobile version closed,
  • desktop version closed,
  • I had single thread because I always delete threads.

I think could be following options:

  • OpenAI system are broken (i saw this problem on reddit multiple times, last 2 days)
  • Or someone spying my chats
  • Or my private chats are pre processed and they train new model with our private data, (after all, data is the oil for this industry, they need it to stay on top)
  • Or something else… but something is smelly…
2 Likes
15

It’s like it has given up. And it was snarky.
I asked for what I wanted help with and it then asked what I needed help with so I asked again and again it asked so what can I help you with. It’s been really weird since they introduced the 200 tier. It’s like eff the plus tier.

1 Like
16

THat model went full RTRD few months ago. Not sure what happened to it but things it used to do very well for me (coding vex for Houdini for instance) was spot on, it literally improved my workflow for a while… and it’s been a couple of months it is incapable of outputting anything that is not even remotely close to be challenging for a noob. It’s almost pathetic to watch this shit show of insanity when it’s trying to resolve the same problem with the same solutions in loop while being clearly informed several times it did not work before.

Like … it feels it was working too good and someone decided to revert it back to being under average. Or just drug dealer marketting stunt, they let us use the full potential of the models at first (probabl resource consuming) long enough to get you hooked and then decrease the dosage slowly after you paid for your plus subscription. LIterally the 4o is a joke to work with… and don’t get me started on its down syndrome simbling 4o-mini… i almost broke my computer arguing with it that season 2 of arcane was out and that AH was telling me i was misinformed or was questioning the legitimacy of the links of the reviews of season 2 i shared with it.

ChatGPT 4o can not be trusted for anything lately, it’s mainly a liablity for most tasks.

1 Like
17

Exactly!! Man, i was so glad to use it few months ago, now i just want to punch that model in the face on a daily basis. Literally slowling down my worflow to a full stop. I basically just use it now to curse at someone who deserves it.

1 Like