O1 Forgetting messages in same conversation

ChatGPT
1

I am using both o1 pro and o1. When ever I have it coding a file it seems like its unable to even remember the past few messages in the same conversation. Last week it helped me in coding a multi file wordpress plugin, now it cant even remember a basic function and it randomly put the number “35” its nt even humorous where it should have been 42…

Is anyone else finding it seemingly lazy and lacking memory the last week?

1 Like
2

The rise of MaliceGPT, as long as you have not pointed your finger on it’s mistake gpt is AI.

And the moment you did, “Let’s play dumb” mode intiates.

1 Like
3

I did mention the code wasn’t working… looks like I am down a slippery slope… its just going in circles…

4

@CoCoBunnii i have the same issue, im working in C# with a unity 3d class and we are talking about changes and all of a sudden i have to repeat myself constantly, as soon as it runs out of memory (as a guess) it because the most annoying tool, up until that point its great but as the chat becomes longer there is a point of decline and reopen a new chat, realistically the Pro mode needs more memory to store data and not just think

5

Hey, I hope you’re doing progress with the issues highlighted. I recently learned few things here in this forum and I just wanted to clarify before any further confusions arise.

I am new here and just like you, I came to seek some solutions.

You may try to see if these 2 things help or not:

1. Change Methods/Approach to get things done:

  • On Network Chuck’s YouTube video “You’ve been using AI Wrong”, the interview with Daniel Miessler about Fabric AI. They talk about how Fabric works.
  • It aims to augment human capabilities by integrating AI into daily tasks and workflows, offering a modular framework for solving specific problems using a crowdsourced set of AI prompts.
  • Which I found to be useful, and it works with Adobe Photoshop’s - Generative AI. And that thing flags every single thing that is either direct or close to description of human body parts
  • Simple things like distorted fingers due to motion blur etc. it won’t fix.
  • And somewhere in the video I heard them say “for some reason, it responds better to a linguistic approach” or something like that.
  • And I changed my approach from prompts to talks, e.g.: “Oh man, my fingers look like crippled and horrifying I wish some way it looked natural and nice” that worked for me.

2. Analytical Tools:

  • Lean Manufacturing/Operational Excellence: I have been using them since 2010, and I find the Why-Why Analysis Tool is a something you should try when getting refusals or stonewalled by gpt.
    **stonewalling: Repeatedly responding with the same term or phrase over and over again like “Understood” instead of addressing the actual question, ignoring the user’s explicit requests for proper answers.

  • You may try to ask if it know what Why-Why Analysis is first, then ask why it refused, and keep adding why after every given reason till you feel that u got your answer.

  • Just a quick heads up, WWA says almost every problem can be solved by asking 5 times why and 1 how. And you ask how to a tangible one.

  • Rest you will explore and I’m sure reveal much more things with this tool.

  • Do read about why-why analysis once before asking GPT, because It now highly alters the content to its convenience or defense.

Do let me know how it goes if you choose to try them out.

Good Luck and Happy new Year Ahead.

6

I thought this same thing, but that’s not the case.

Start a new conversation and if you have used custom instructions, enabled then try asking it to repeat them back.

I believe it’s been heavily tweaked ever *** and the tweak is not done properly. that’s why its refrains from repeating its own response immediate response in exact same manner.

And custom instructions/memory are not like regular tokens that gets exchanged in turns.

7

I find I have to make a new chat and then copy and paste the file structure and code into the new chat, almost as if I am retraining the chat. Using a project folder gives results that are less than satisfactory and I don’t think I can use pro mode in the project folder. I also add a comment section in each of the files i copy and paste to tell what the file does and where it is in relation to the plugin I am working on…

Breaking the code into smaller functions sometimes works as well…

8

update:
It’s now adding mock numbers in the code when I ask it to pull a number from a variable earlier on in the code.

Peer laziness