Anyone else getting issues with chatgpt 4 repeating itself over and over?

This has started recently, when i ask a question, chatgpt will repeat its last answer before answering me. Has anyone else encountered this issue, am i using it wrong? See below:

User

how can i get into fpga programming?

ChatGPT

Getting started with FPGA (Field-Programmable Gate Array) programming can be an exciting venture into the world of hardware design and digital logic. FPGAs are versatile and powerful devices that can be used in various applications, from simple logic controllers to complex system-on-chip designs. Here are some steps to help you get started:

Understand the Basics

Learn Digital Logic Design: Before diving into FPGA programming, it’s important to understand the basics of digital logic and circuit design. Study topics like logic gates, flip-flops, multiplexers, and finite state machines.

User

are there any cheap fpga boards

ChatGPT

Getting into FPGA (Field-Programmable Gate Array) programming can be an exciting journey, offering the chance to work on cutting-edge hardware projects. Here’s a structured approach to start with FPGA programming and some suggestions for affordable FPGA boards to consider:

Understand the Basics

Learn Digital Logic Design: Before diving into FPGA programming, it’s essential to have a good grasp of digital logic and circuit design. Consider starting with basic logic gates, combinational and sequential logic, and how digital circuits work.

Study Verilog or VHDL: These are the two primary hardware description languages (HDLs) used in FPGA programming. Choose one to start with (Verilog is often considered more straightforward for beginners) and get comfortable writing and understanding HDL code.

  1. Select an FPGA Board

When choosing an FPGA board, consider your budget, the complexity of the projects you aim to work on, and the learning resources available for the specific board or its development environment. Here are a few affordable options:

This happens every time, and my chats have also not been getting saved when this happens and i referesh the page.

4 Likes

Yes, a similar issue was previously reported, with the assistant messages not being saved and the user input being seen as multiple questions to be answered, or similar.

You can abandon that broken chat and try again.

But several recent problems with network etc have been found and are related to browser (default) privacy settings, adblock and tracker blocking extensions, certificate trust, etc.

You can click on the shield icon in Firefox, for example, and see where sites are trying to elevate their privileges beyond what is required, for example. Or what should not be required, but web developers and captcha makers and session recorders and trials want to impose on you.

image

2 Likes

Yes, the same thing has been happening to me since at least 2 or 3 days ago. Very annoying. I have to tell it not to repeat itself.

This has also been happening to me. I think what is happening is the response isn’t saved. But when you type in your next prompt, since it hasn’t saved your last response, it just assumes to add the 2 prompts together. Resulting in it answering the first prompt again into the second one. It is really annoying and I have not found a fix.

3 Likes

I tried mentioning this in the custom instructions settings, but no success.

Anyone found some working prompt there?

We also had some cases where new conversations had the same conversation id as the previous one.
Then the two prompts from the two conversations would get ‘fused’ and the model reply resulted in a error message.

While I didn’t have this issue before it may be worthwhile to check the conversation id in the browser and try to force a new id before continuing.

Logging in and out, clearing the cookies and the cache before each conversation etc…

Hope this helps somebody…

Maybe some relation to chat history?
I’m thiinking about delete it all but i still access some info from some of them.

Very annoying, answers get longer each iteration in a chat.
And it seems to ignore now my custom instructions from settings too.

We can just assume OpenAI is aware of every bug, since we don’t have a way to report?

1 Like

It is also esting through my usage like multiple calls are being made.

1 Like

I’m using the OpenAI Python API and I also have these problems, so it has nothing to do with network connection or messages not being sent.
Also tried changing the prompt and increasing presence_penalty/frequency_penalty with no luck.
Has anyone found a solution for this that does not involve fine-tuning?

I think I might be running into the same issue.
Whenever I start a new conversation, it works fine the first two or three prompts and answers, but after that nothing works - as soon as I refresh the page, the last answer(s) will be gone and all of the prompts I’ve sent up to that point will be crammed together into my first message.

Also, the GPTs will start repeating their answer from the previous prompt - and then everything vanishes as soon as I refresh the page. This happens on the standard GPT4 model as well as all of the custom ones I’ve tried.

I’ve archived all my other chats, no change. Deleted them, no change. I have tried Firefox, Opera and Safari so far, all without any extensions and in incognito mode. I have tried on my computer and on my phone. Nothing seems to work, it’s completely unusable like this. This happens on BOTH accounts in my workspace.

I seriously don’t know what to do anymore, I’ll probably cancel the subscription at this point.

3 Likes

Same here. It will often start by responding again to my previous question, followed by a response to my latest question. When asked to give a numbered list of previous messages in the conversation, some will be missing, merged together or reordered.

1 Like

Was searching for someone experiencing this as it has been happening now for about a month. It is really annoying. So what, we need to disable adblockers or something?

1 Like

This is also happening to me for about a week now.
ChatGPT answers my current question AND my prior question on every interaction.
It doesn’t typically happen at the beginning of a new chat session, but after a while, it starts answering my current and prior question on every single interaction.

Thus MUST be eating up resources! @vb

Yes, I have had this happen in my responses a few times in the last week. It’s super annoying but reflecting on my prompts I believe it’s necessary to tell the model ‘Ok, let’s move on to the next question, which is …’

I don’t know why the behavior is now like this but the required effort to adapt is rather low.

Make of this what you will but I hope it helps somebody.

1 Like

So saying that makes it not focus on the whole preceding part of the message (what it perceives as one message)?

My assumption is that the model ‘thinks’ the previous question from the user is still important. Using the prompting technique I described above guides the model to focus on the next question.

This should not be happening, especially not with these repetititons but it works for me. Op’s example is really a good one, as one would expect the model to proceed to the next question without further ado.

Your interpretation is close, but it’s generally not necessary to be as explicit as you said. I’ve found that if I’m clearly continuing the previous thread it understands. It’s primarily with short prompts, especially those that seem not to address the previous response, that it repeats itself.

People who tend to write short, disjointed prompts probably encounter this a lot, while people who write elaborate prompts (as I often do) only see it when we’re terse.

Have you guys tried asking it why it did that? Obviously, it doesn’t know why, as has no memory of its previous processes, given that it’s data fed through a static model (ergo not intelligent at all). But it does invent interesting rationale, generally that it found the explanation very important.

1 Like

The suggestion by @vb has been working well for me but it can actually be kept very brief. A simple “Cool, what about…” or “Thanks, how about…” seems to be enough to signal a change of topic.

I think I started seeing this after using it for a year or so and began dropping the courtesies and conversational fluff that I used to use. I’m a little skeptical that this is a new issue of chatgpt and more of an issue I started having when my prompts started getting more terse (like @kazvorpal pointed out).

1 Like

Yes. It’s actually a positive if the model looks at the context of the conversation and replies accordingly. I mean the cases when from one reply to the next the previous context is dropped have also been reduced. This makes for a much nicer experience when working on a specific issue, as I don’t have to hit the regenerate button as often.

Regarding how much one has to write to tell the model to move on: it’s likely also about being very clear. I can write a elaborate prompt but if it doesn’t contain the request to move on the model may still put too much weight on the previous prompts.

I’m not sure if you realize what’s going on here. Have you actually not noticed that it is simply deleting it’s own replies?

It happens after just a few messages.

Example:

Me: Perform task please.
ChatGPT: of course! here is the result
Me: perform task 2
ChatGPT: of course, here is task 2!

Me: How about this 3rd task?
ChatGPT: Of course! Here is task 1 :slight_smile: For Task 2 you can do xyz.
Task 3 is more difficult here is the plan:
bla bla bla.

At this point if you just reload your page, you will see that The system believes you have simply sent it 3 instructions in one single message. Unless you reload the page, you will never see this, but the replies from ChatGPT are basically streamed to the browser but never saved to the database.

I canceled my ChatGPT subscription today and signed up for Claude 3 Pro, because GPT4 is no longer usable for me for anything other than a simple message and reply, which I can get with Copilot.

It is not perfect either and suffers from the same types of hallucinations and tendencies to overlook things as conversations get larger, but it is much better at producing large code blocks without placeholders and prompt following in general.