#1070006915414900886 I have been requesting my data for over 4 months from OpenAI, daily. I get a notice to check my email and a link will come, but none ever does. Our email server log shows that OpenAI never sends it. Ironically though, they do send billing and forum updates, so the data export feature is just broken/not working and they have the correct email obviously. Before this gets ugly and legal entities have to get involved - or this turns into a class action, can someone please fix this or send me my data. I will not continue to beg, and after the next 48 hours - things will start turning ugly very rapidly. I have over 20 support tickets going at OpenAI, & constantly have to wait a week just for a response. If you cannot run a business or provide support - close the doors. Kind Regards - Matthew
#DATA EXPORT HAS BEEN BROKEN FOR 4-5 MONTHS
slim saffron
slate wharf
<#1070006915414900886> I have been requesting my data for over 4 months from Ope...
Does it say error or does it say successfull
slim saffron
successful everytime- but it's not.
As a software developer and engineer of going on 30 years - it's not hard to put a toast/alert message in place without any actions actually occurring somewhere - bad programming/development seems to be at work here.
At this point it's negligence - they've been made well aware
slate wharf
Your thw only one with the issue
narrow agate
as a software developer of 40 years i can recommend checking your spam folder - low common sense seems to be at work here.
slate wharf
Maybe your account is connected to a mail that your not checking
narrow agate
also as a software developer and engineer and architect of going on 50 years i can tell you that bug reports that include insults tend to be ignored
slim saffron
as a software developer of 40 years i can recommend checking your spam folder - ...
Pleae learn to read before being snide. Reading is fundamental to engineering. The mail server indicates that OpenAI has NOT sent any messages. There have been no communications apart from billing notifications and developer forum updates/notifications. So - OpenAI HAS NOT send the data export.
slim saffron
also as a software developer and engineer and architect of going on 50 years i c...
Regardless - I've requested my data, and if they're permitting consumers from my state to use their services, they MUST provide a copy of all user data upon request - I believe within 72 hours. We are now 4 months over, 5 in one week. So it's not a bug report - it's now a legal issue. Thanks!
narrow agate
You have access to the open ai mail server?
slim saffron
Oh my - you can run a reverse MX lookup to see where messages should or will originate from. One only then needs to compare the only IP addresses that ALL communications will originate from - and if there are no matches, with 100% certainty, they have not sent any communications. 🙂 I do forensic work and software - so it's a bit more in depth, but I can tell you with 100% certainty, no data export has left OpenAI's server's bound for mine.
As a developer - I can just rip the data from the server if needed through simple scrape mechanisms, but still - if you're going to offer data exports and offer services knowing you must follow the law - they might as well just fix it and stop with the nonesense 🙂
narrow agate
as a matter of fact the email arrives within minutes, i just tried. It's probably easier to change your account to another email than scraping your stuff
slim saffron
They can always hire me - I'll even do the work for free - just so it works. (Sent my resume to Sam 🤣 as well)
If my mail server was misconfigured, I wouldn't be getting their billing notices, and developer forum notifications. There is still wonkiness on the back-end of ChatGPT - some accounts are still mixing data - so there is certainly privacy and data issues still residing. Regardless, we just need all our data to train our local LLM so we can cut OpenAI out of the mix - not joking 🙂
**We have over 20,000 chat logs in ChatGPT 🤣
narrow agate
you want to train an LLM with chatgpt chat logs?
If it helps, the data export comes from noreply@tm.openai.com
which is probably why your mx lookup failed
slim saffron
Not so much - we have a trained LLM already and we need to align it with the alignments that we configured to chatGPT thru prompting - there are over 2000 prompts in the 20,000 message logs, and we need them ripped out for alignment of our LLM - if we rip the alignments we configures and add them locally it'll take a few days, if we have to manually do it by hand, months...
narrow agate
Yeah that will be easy with the export, you get a handy json with roles marked
slim saffron
yes - noreply@tm.openai.com - not a single email has orinated from any of the associated IP's or even attempted to connect to the SMTP server for delivery. These are the ONLY IPS, data exports can originate from 🙂
narrow agate
Process probably hangs up with 20k conversations
In my 80 years of software engineering i learnt that in this case it is easier to scrape instead of waiting for such an edge case to be fixed
slate wharf
Pleae learn to read before being snide. Reading is fundamental to engineering. T...
They respond in discord but if they dont repond here they dont want too
There are employees in the discord
Maybe just the part of them reading dont know what do do because they dont work in that department.
slim saffron
If anyone else has problems with export and wants a tool to automatically export data without OpenAI export feature - I ended up crafting a console based tool that prompts to login to ChatGPT, and then gather's all conversation ID's, and recursively loops thru them parsing all HTML into exports. I even went as far as to add command line parameters to export as PDF conversation or into JSON format while awaiting to hear back from OpenAI. For a quick and dirty method, you can also parse conversations using Javascript and Chrome plugin with the following code: function replaceImagesWithAltText() {
const images = document.querySelectorAll('main img');
images.forEach(image => {
const altText = image.alt || " [Image: " + image.src + "] ";
image.outerHTML = altText;
});
}
function stripHTMLFromMainContent() {
replaceImagesWithAltText();
const mainElement = document.querySelector('main');
if (mainElement) {
const tempDiv = document.createElement('div');
tempDiv.innerHTML = mainElement.innerHTML;
// Remove text content within span elements with class "flex-grow flex-shrink-0"
const spanElements = tempDiv.querySelectorAll('span.flex-grow.flex-shrink-0');
spanElements.forEach(span => {
span.textContent = '';
});
const textContent = tempDiv.textContent;
console.log(textContent);
} else {
console.log("Error: The <main> element was not found on the page.");
}
}
stripHTMLFromMainContent();
Still waiting to hear from OpenAI 
**Only raised a stink because GPT advised I should 🤣 Love the last line....
sly helm
Requested data over 1.5 months ago, and never even received an email.
slate wharf
**Only raised a stink because GPT advised I should 🤣 Love the last line....
Why does it say as an ai what do you think about this?
It doesnt look like chatgpt would write like that.
slim saffron
Why does it say as an ai what do you think about this?
Quite simply because ChatGPT is an AI language model, and I asked, "as an AI, what do you think about this." Otherwise, GPT's directive is to avoid opinions, as it tried to do in its first sentence by avoiding technological discussion, but then eased up to provide an opinion and advice as requested. Part of prompt engineering is to know how to get input out of the system. Garbage in, garbage out. ChatGPT is not a question/answer system, rather a probability engine to determine sequence of tokens, which is also why it is able to hallucinate it's current state. This was straight forward prompt questioning though. This is why we're aligning our private LLM for our business, to prevent coercion like GPT is faulty of innately. OpenAI hasn't even figured out why temperature doesn't work correctly API wise; When the answer is actually simple and has to do with MOE, which we've fixed on our own LLMs. We need our LLMs to be consistent, meaning regurgitate correct information, nearly identically by 99% if asked the same question multiple times. With a random temperature that doesn't adhere to what is specified in an API request, answers become less deterministic; which is a problem. GPT will provide similar answers 80-90% of the time, but 10-20% variances in the Fin sector are a big risk/liability. It's quite easy to get gpt to fly off the guardrails, and produce output that it should not, and we cannot have this. Which is why we've been crafting a number of internal LLM models for the last 4 months to perform a variety of model-specific actions, without data leaving the building. After this month - we'll be able to shutdown our ChatGPT use for good, so their bugs won't really matter - our alignments are being put in place this very moment and should be completed over the next 2 days.
uneven girder
we believe we fixed the issue, can you give it another try?
sly helm
This issue isn’t fixed, Valerie.
I’ve tried a week ago and I’m still awaiting the email. Another user just mentioned it not working as well in another channel (I’ve linked them to this thread maybe they’ll follow up)
And as others have pointed out in this channel, this is a legal issue in at least the whole of the European Union.
Saying it’s a bug might get some wiggle room. Saying it’s the users fault using a wrong browser can be more wiggle room. The facts how ever point to either a broken data processing flow on oai side or intentionally not doing what it should.
I’m not relying on the feature myself. But I’m sure there can be rather annoying consequences if it’s not addressed in legal terms.
narrow agate
I think the problem is that valerie hasn't hired a guy with 90 years experience in software engineering to fix it