I am working on some code that is several hundred lines long. I have to feed the code to ChatGPT in chunks so that it can see all of it. When I do this and ask for an explanation of code at the beginning of the file, it does not remember any of my previous input. It only remembers the most recent input that I gave it. In older versions, ChatGPT had full recall of all input provided in the conversation. Without full recall, ChatGPT is far less useful. Please allow this to be enabled or reenable this feature.
#Bug Report: Short Term Memory - ChatGPT is forgetting previous input
keen musk
The free version of ChatGPT should have enough memory to remember basic details in conversations with a lot of text input. Additionally, you should consider allowing ChatGPT to have significantly more memory in the paid version. That is the one of the only things that could get me to pay for a subscription. Without more memory, it becomes useless for analyzing larger inputs.
Bug Report: Short Term Memory - ChatGPT is forgetting previous input
Steps to reproduce:
- Submit any input of code that maxes out character limit.
- Submit second input of code that maxes out character limit.
- Ask ChatGPT for details about code contained in the first input.
fervent plover
Confirming that I am receiving this issue on May 24th GPT-4, but with far less than a few hundred lines of code. I'm inputting 54 and getting confused responses.
junior cloud
I can confirm
mortal forge
I am having this exact same issue but only with gpt-4. It's driving me crazy.
cyan scaffold
It can only remember 8k tokens
So if you paste code and asking it to change it over and over it is going to forget things. From the first request
And then it will start to modify your code and forgetting to add previus things and then it is wrong and so on
mortal forge
Is there any work around or will the number of tokens be expanded in the future? Coding and input/output examples can eat up a lot of memory very quickly.
cyan scaffold
GPT-3.5 Can remember 3k tokens.
GPT-4-8k(gpt 4 on chatgpt) Can remember 8k
GPT-4-32k Can remember 32k tokens. You can just upload the code into a diffrent conversation and tell it what to do
What i mean by upload is send
mortal forge
First, thank you for your help. Appreciated. What is the difference between GPT-4-8k and GPT-4-32k? Also, I am using a paid account and I was working with the same code for a couple of weeks with no issue, then suddenly this issue cropped up. So, for me it seemed out of the blue.
cyan scaffold
You will need to get an API(if your not a big company you wont get it in a few years or at all.
You cannot get 32k engine with plus
You cannot use 32k engine in Chatgpt
mortal forge
Thank you. I understand.
How can we count tokens before we submit to gpt-4. Is there a method for doing this? Maybe I could try preformatting my information and count the tokens before submitting.
To help others that may read this. I found https://platform.openai.com/tokenizer
mortal forge
I used the above tokenizer and kept my entire session below 7k, but still had the same issue.
smoky fable
I asked the ChatGPT to convert the code into a code block. But ChatGPT return a non complete version. It is forgetting my input. This happens on gpt 3.5 & gpt 4. It's making me crazy to ran out my gpt-4 cap for this stupid work.
cyan scaffold
I asked the ChatGPT to convert the code into a code block. But ChatGPT return a ...
Gpt-4 on chatgpt is 8k so it can remember 8k tokens
3.5 can remember 3k tokens