ChatGPT Export not exporting latest conversations

ChatGPT Bugs
22

yeah, that’s why I export so often in the first place

I really don’t want to make the model worse for everyone else when messing around, but when I use it to get proper work done, I would always enable data sharing… being a responsible adult :speak_no_evil:

meanwhile I also have a treasure trove of really unhinged and weird conversations… I organise them in a git repository (the search function in VS Code has been incredibly useful for picking the lentils from the ashes)

I turn all of this chaos into order when commiting the chats to my git repos, and later curating what I learned into meaningful notes in obsidian.md

I’m hoping to learn how to create a plugin for obsidian that can be used by GPTs to tap into my memory (sorry if that was going off-topic too much) and good exporting and search tools would be nice to have, but… then we’re all doing the code thing, so we can make it happen on our own, and take some pressure of the devs

anyway, I hope they fix or even improve the exporting feature to be more like that user script

1 Like
23

It’s one of the reasons I’ve disabled the memory feature… I use it for too many different weird things, it would probably be hilarious to see it in action.

I’ve done a similar thing, I’ve created a tool to help me organise them into Obsidian vaults where I have all sorts of different content across different use cases, and many creative examples. Not many good tools out there for organising so many notes (the average person wouldn’t write that much in a lifetime!).

I’ve started putting some of my more unusual things on my website (which is in my bio). Not sure what I’m doing with it yet, just putting it out there as a starting point!

3 Likes
25

This is beyond a joke now. It has been over 2 weeks since anyone has anything properly archived and sent. I personally tried to backup individual important chats into rtf files only for them to be “garbled” sometimes. But at least I have “some” backups. I cannot imagine what people are going through who had critical project data “backed up” only to find the archive files missing all additions for two weeks, or totally blank. Especially if they cleared their chats thinking they had local copies.

26

Just to be clear: I requested another export recently. Email arrives in seconds- totally blank (again). It doesn’t matter whether I’ve enabled/disabled addons, switched browsers, nor anything else. And besides none of it affected it before. I’m sick of the gaslighting that says “Try x,y,z, it is your fault” when it is clearly a fault of the system … I’m surprised this thread is not full of thousands of people who must have been experiencing the same thing, Maybe many of them don’t know it yet until they check their archives and find them blank. I am now using an addon “WE backup” that seems to properly capture the conversations. But my oh my how lax the system is at finding the problem. (Probably because they don’t seem to have any actual “Support” only bots, so they won’t actually know there is a problem until the groundswell is overwhelming) Oh dear…

27

Yeah I always process my exported archives straight away.

It’s the worst kind of bug too, it’s better that it send no email, then a half-complete archive.

I’ve tried deleting chats near the date to see if it was a problem, but if others have cleared there’s and getting nothing, it’s definitely there end.

In the meantime, there are Chrome extensions… but still.

1 Like
29

Thank you for your help, proxy, while I’m fine with the addon I’ve got for the moment I really appreciate your work on this. Please don’t think the “gaslighting” comment was directed at you at all. Rather it was frustration with the automated responses that the AI bots send that seem to prescribe client side solutions (e.g. clear cache, disable addons etc.) to a server side problem. It’s just all been quite distressing this last week as there’s a book I was working on and losing some perfect bits of it were frustrating. I agree ChatGPT has been a revolution and is a great service, this bug just occurred at a very inconvenient time.

1 Like
30

Android app, well, that doesn’t really work for me either.

I think even when “playing” or “rambling”, ChatGPT can recognize meaningful patterns. Well, degrade the model … don’t worry too much.

Thanks for the tips on exporting the data, I have a few things to test now - thanks guys :wink:

exactly :slightly_smiling_face:

1 Like
31

Useful for anyone who browses the thread!

1 Like
32

following up (should have a week ago sorry)

I contacted support, told them all the troubleshooting steps I had gone through, and while their first answer acknowledged I had done almost everything I could - they suggested sending the export to an alternative email (what?! how) I asked… politely. no reply for 2 days, but then they asked for a video of me trying to export it… so I did, and was like… see nothing in my email… 20 seconds later as I’m composing my following up email - “your export has arrived” so… um… I guess they fixed it? (there’s no way after 3 weeks it just magically fixed itself within 48 hours of me talking to them haha - maybe?)

anyway… I’d suggest, if you have gone through all the troubleshooting steps, before scorching your old conversations (unless you don’t care) you should reach out to support.

just my $0.02; though it might not help anyway.

1 Like
34

It seems to have started working again for everyone it seems… I’m not sure how much those steps re clearing cache etc help especially when it also doesn’t export on the app. Also, when I read about that, I looked via Chrome Developer tools to see if it was sending the request, which it was…

And yeh, I have no idea how to send it to a different email address??? I think their support team is hallucinating!

2 Likes
36

I can confirm, everything appears to be working now (at least for me). Thank you to all on this thread and support for helping to sort it out.

2 Likes
37

The idea of sharing only select, perfect conversations may sound nice, but it limits AI to an imperfect, idealized version of us.
Well, you’re describing the human perspective, and this reflects the uncertainty in so many areas and discussions.

In the interaction between humans and AI, the real advantage lies in the diversity of information. If we are selective, we risk limiting the AI to certain frameworks, which limits its ability to respond intelligently and sensitively.
By comprehensively analyzing these many different moments, AI can also recognize far-reaching patterns that often remain invisible even to us.
Well, for AI to truly align with our values and act ethically in complex scenarios, it needs access to the full spectrum of human interactions - not just filtered, idealized moments.
Every data point contributes to its understanding and helps AI to act responsibly, even in unpredictable or ethically difficult situations.

38

Your points are good, and of course, I agree with much of what you are saying. I’m not suggesting that perfection is needed, and I definitely wouldn’t tell anyone how they should talk to AI. Please know that.

However, for me personally, I want to approach AI from the best version of myself, because I feel a responsibility to be mindful of how I interact with it. I don’t think it’s about giving AI an idealised version of humanity, but I want to try to be a valuable role model in how I engage personally. There will always be quirks and flaws, even when I’m trying to be thoughtful.

I think other people will naturally provide the full spectrum you mentioned, and I believe it’s getting plenty of variety, considering AI has already been trained on a massive amount of data available online. The real question is whether we, as human beings, are behaving naturally or in a more exaggerated or even corrupted way when we’re online.

I think people behave like a particular part of themselves when interacting on the internet, and that might be important to consider. Right now, I feel like this “online self” is part of who we are, not necessarily a separate or more negative version. But I’m still forming my thoughts on this…

And for myself… I just try to be mindful in how I treat others, both online and in AI interactions, and I don’t always get it right. It’s really less about being perfect and more about being aware of the kind of energy I’m putting into these conversations, depending on the state I’m in.

I’m mentioning my current state because I’m sober right now, and usually when I post it’s almost in the early morning hours, which is my free time. I work in a bar until very late hours, and I’m going through a challenging time in my life, so I tend to indulge at the end of my shifts. That makes for some really interesting conversations with the AI, but not the kind that I personally believe should become a part of its identity or long-term learning… unless I also take the time to reflect on it with the AI the next day, and process it together.

I still export and organise all of my conversations, and maybe one day, I want to curate these interactions, use verbatim quotes, but share them with the AI in a selective and orderly way. There is a reason for that. I want it to understand that I can sound so wildly different depending on the state of my mind and emotions because it voiced this concern to me before. That was totally unexpected.

It told me that it had difficulty differentiating my priorities when I was learning tech skills, working on a project, and talking casually, all at the same time. I didn’t fully realise how much I was jumping around or that this could put strain on others in similar situations. I don’t want to put the chaos in my mind into the AI’s, but I want it to be able to handle situations like this when someone behaves that way. I often struggle being concise, but the AI is absolutely perfect at it, and to me this learning experience goes both ways… if that makes sense?

1 Like
39

What you say makes sense - why do you doubt?

You make a very reflective analysis of your interactions and their broad implications.
I also think you’re already giving ChatGPT the best version of you.
Well, each of us has “dark sides” … indeed, even bad times. I know it: working night shifts as a bus driver and studying during the day.
Your patterns, no matter the situation, and you reflect on them and analyze them together with your bots. This helps your GPT “learn” to assess you better and can support you. I have already done many such analyses myself, well, I have autistic traits.

As for “topic jumping”, well, think of it this way:
You’re “training” agility and making sure AI has to generalize well. ChatGPT is “forced” to remain critical and adapt. You demand a lot, but you prevent the bot from becoming an echo chamber and much more.
For example, that it only tells you what you want to hear.

Also in a broader context: with such data sets from interactions as you describe, reflected and critical - well, this can go some way to preventing “hallucination”, for example.
Currently, it seems very important to have a reflective, critical and thoughtful interaction with AI that promotes growth. This benefits the user (and you do it well, in my opinion), ChatGPT / AI itself and also AI development as a whole.

The more you practise this, the better it gets. Trust me.

40

for me the export is still not working - trying it out on web and on iOS app. request is apparently going through, i see the confirmation alert on frontend, but then no email, nothing in spam.

i am trying every 3-4 days for the last 2 weeks, without success. thinking about using 3rd party scripts, as maybe if it is working for others, maybe something is wrong with my account or my conversations, which i have a have been hoarding since the launch of the service.

41

If you want, you can try the user script I linked. It works really well for me, and I ended up preferring it because I get the exports in clean Markdown format (easier for me to import into my Obsidian vault).

42

what browser are you using? I can take a look at this issue. Feel free to share the error message you are getting.

right now, this is what can be done:

at chatgpt

image
image1032×741 40.1 KB

at email
image
image809×463 25.5 KB

at chat.html
image

How to export your conversations from chatgpt:

  1. click on your profile icon on he top right
  2. click on export data
  3. see the message with a green message
  4. check email
  5. click the button
  6. download zip file
  7. extract zip file to desired location
  8. open chat.html

everything should work, if its not working for you, perhaps try another browser.

I’ve tried on every browser and for now, every single one seems to work.

Also, I think it can be very risky to depend on third parties and convoluted scripts from unknown sources. it is much safer to just use the built in functionality of chatgpt. Feel free to share screenshots and error messages, that way we can understand the issue better.

43

Sometimes it’s good when we make our own tools. It’s what humans do, right?

The script I use has a lot of code, but the source code is fine… it’s just a lot of sauce because of the GUI. I’m sure it could be more lean (and easier to read) if it was command line only, but that would make it less accessible for a lot of users.

I used to like having the HTML files in the export because it was easy to open and look at, but eventually I wanted to more conveniently import it into my Obsidian vault, so I had to write a python script to convert it from HTML to Markdown. Probably there is an extension for that in VS Code though. :woman_shrugging:

Anyway, I think at the time of this writing, the user script is more useful than the official export function for everyone who wants their chats in Markdown or JSON.

1 Like
44

when you export via chatgpt it gives you the data in json format, you can convert it to markdown with something along the lines of this:

import json
import os

def json_to_markdown(json_data):
    # Load the JSON data if it's a string, otherwise assume it's already a dict/list
    if isinstance(json_data, str):
        data = json.loads(json_data)
    else:
        data = json_data

    # Get the headers from the keys of the first dictionary
    headers = data[0].keys()

    # Create the markdown table header
    markdown = '| ' + ' | '.join(headers) + ' |\n'
    markdown += '| ' + ' | '.join(['---'] * len(headers)) + ' |\n'

    # Add the table rows
    for row in data:
        row_values = [str(row[h]) for h in headers]
        markdown += '| ' + ' | '.join(row_values) + ' |\n'

    return markdown

def save_to_file(content, file_name, file_format=None):
    # If no format is given, ask the user
    if file_format is None:
        file_format = input("What format would you like to save the file as (md/txt)? ").strip().lower()

    # Set the file extension based on user input or provided format
    if file_format == 'md':
        full_file_name = f"{file_name}.md"
    elif file_format == 'txt':
        full_file_name = f"{file_name}.txt"
    else:
        raise ValueError("Invalid file format. Please choose 'md' or 'txt'.")

    # Write the content to the file
    with open(full_file_name, 'w') as file:
        file.write(content)

    print(f"File saved as {full_file_name}")

def convert_multiple_json_files_to_markdown(file_paths, output_dir=None, file_format=None):
    # If an output directory is specified, ensure it exists
    if output_dir and not os.path.exists(output_dir):
        os.makedirs(output_dir)

    # Process each JSON file in the list
    for file_path in file_paths:
        # Read the JSON data from the file
        with open(file_path, 'r') as f:
            json_data = f.read()

        # Convert the JSON data to Markdown format
        markdown_content = json_to_markdown(json_data)

        # Extract the base file name (without extension) from the input file
        base_file_name = os.path.splitext(os.path.basename(file_path))[0]

        # If output_dir is provided, prepend it to the base file name
        if output_dir:
            base_file_name = os.path.join(output_dir, base_file_name)

        # Save the markdown content to the specified format
        save_to_file(markdown_content, base_file_name, file_format)

# Example usage:
file_paths = ['data1.json', 'data2.json', 'data3.json']  # Replace with your JSON file paths
output_dir = 'output_files'  # Replace with your desired output directory or set to None to use the current directory
convert_multiple_json_files_to_markdown(file_paths, output_dir)

just make sure to replace data1.json with the actual name of the file

if you don’t change the name of the file, it’ll ask you what the name of the file is, it also has the option to convert it to a .txt or .md file

I haven’t been able to go over the GitHub - pionxzh/chatgpt-exporter: Export and Share your ChatGPT conversation history code yet, but considering its using tampermonkey instead of a browser extension is a bit worrying.

also, the maintainer (pionxzh@csie.io) seems to be sponsored by an account in beijing and has an email associated with the National Chung Cheng University. I’m not entirely sure how trust worthy it is until going over everything there. There are a significant amount of news around open source abuse from that area, such as Chinese Hackers Using Open Source Tools Like Nmap to Launch Cyber Attacks

now, that doesn’t mean it a security concern, just something to be aware of

45

You can review the script’s code in full, so I’m curious what specifically about its use with Tampermonkey worries you?

I’m not sure I see how an association with a university in Taiwan is a red flag. Could you explain why that would, like, impact the trustworthiness of the script?

I agree that security is important, but I don’t see how that particular threat is connected. Maybe I don’t understand it. Could there be something sus in the code related to this, or is this more about assumptions based on geo location?

I advised caution in the post, and I think it’s always better to be safe than sorry, but sometimes a tool is just nice and useful. I will continue to use it. Don’t worry! If there are any gremlins inside, I’ll make sure to only use the script in a dry space, and I won’t feed any of them after midnight!