I have had an extensive email chain with OpenAI, where I am 99.9999% sure I am talking with chatbots. I am also confident that the site’s chat support is a chatbot (which they are upfront about) that passes you off to another chatbot (which they are not upfront about). Are others experiencing this?
It’s rather infuriating dealing with a system that can get stuck in loops, especially when that loop is objectively wrong. It’s especially infuriating when they end up in such a loop after they’ve already determined the real root cause.
Details specific to my issue:
In this case, I have been attempting to report to OpenAI that their status emails are no longer threaded since they switched status systems. After quite a bit of back and forth, whoever/whatever I was talking to realized that I am correct. I had been asked by a separate person/persona to keep them apprised of the issue and let them know if it persists. Upon sending an update, the new whatever/whoever seems to believe I have control over the content of the emails OpenAI sends, which had already been discussed at length previously, and now they’re stuck in a loop sending the same email over and over with fake apologies and moronic responses.
I can see where you’re coming from, but I would not go so far as to accuse possibly-human people of being bots because 1) it’s damn rude and 2) I wanted help from them.
I got fairly long-form text responses, which are compatible with either an AI-generated response on the single issue, or a template response to a well-known and persistent bug; and the bug being one of many (as any software engineer will sympathise) it’s entirely possible that each bug has AI-assisted template response scripts for human support staff to follow.
There’s a fine line between the (perfectly valid) “eating your own dogfood” in giving human support staff AI tools to work more effectively, and disrespecting your customers. From a week-long interaction with 5 names after the initial (explicit) bot ‘Fin’, I can’t tell which side of the line we were on. What does seem clear from the long delays between my responses and theirs, and the eventual technical & financial decisions given, is that there must be people in the chain somewhere.
For those old enough to remember the days of tech support escalation processes before AI was a thing the OP tale would not be unexpected with a collection of humans who are overworked and not paid enough to really care.
I’m not saying the OpenAI support staff didn’t care and I can’t comment on their pay, but the episode is very compatible with humans who are overworked and don’t have answers to the problem.
I’m glad you felt your experience was positive, but if I were in your position, I would not have considered it a positive experience. It took them a week to tell you they did a terrible job of a migration and you’re losing all of your previous data? And it was clear throughout most of that week that you were being given the runaround? Sure you might have learned some tools, but that doesn’t mean support did things in a positive manner.
It’s rude to pretend I’m dealing with a person when it’s an AI. I was very patient with them until they got stuck in a loop where it was clear the other side was not actually comprehending what I am saying and has completely lost the thread on the issue.
I can tell you that I was probably one of the first people to notice my issue as I informed them after the very first email correspondences and chats. So, even if there was a template, they should’ve stopped using it once they realized I was not talking about what they thought I was talking about, which they did realize, then reverted to their previous interpretation of the situation.
If they’re misusing the AI-assisted template responses, then I am effectively dealing with an AI.
I don’t think this follows. It’s very easy to program in fake delays and personas and AIs are very well known for giving any advice, even when that advice is wrong. And in my case, their advice has been and continues to be off base.
I am not some young whippersnapper. I remember. I’ve been working on this for two weeks now and I’m still talking to the “front-line” support. Even in the old days you wouldn’t be stuck like this for this long and this would still be unacceptable.
Let me share with you the last three emails between us and let you decided if you still believe I haven’t been exclusively talking to AIs. Also, I’d be happy to share the full thread (16 emails).
From Gregg
Hi Joshua,
Thanks for your swift response.
We’re sorry for any confusion that may have arisen from the initial response. We have reviewed all the screenshots you shared, along with the full conversation in this email when it was forwarded to us.
The status emails have different subjects related to different outages, which is why email servers organize them as separate threads. This behavior is based on how email services categorize and arrange conversations.
We recommend reaching out to Fastmail support for a more detailed explanation.
Your understanding is greatly appreciated. Let us know if you have any further questions or concerns regarding OpenAI.
Best,
Gregg
OpenAI Support
From me
Gregg,
It is clear to me you did not actually comprehend the previous email I sent you. The ONLY confusion is on YOUR end.
What you are talking about, is not what this conversation is about. I am talking about emails on the SAME outage not threading because OPENAI is NOT correctly using the subjects on emails. In fact, these emails have the SAME title, but as you know, you need to be putting “Re:” in front of them. No email provider can correct YOUR mistakes.
I’ll even pull out a quote for you from Judith:
Based on what you’ve outlined, this does not appear to be an issue on your end, but rather a change related to how OpenAI’s new status vendor structures emails.
Every email provider has this issue because YOU are the ones misusing the email system.
If you still can’t see how irrelevant your emails have been, you need to send me back to Judith, or preferably a supervisor because this is beyond infuriating.
Sincerely,
Joshua
From Gregg
Hello Joshua,
Thank you for sharing the details!
We apologize for any confusion that may have arisen from the initial response. While the emails may have the same title, they contain different subjects and information regarding outages, which may affect how they are threaded in your email server.
We recommend reaching out to Fastmail support for further clarification on why the OpenAI Status emails are not being threaded in your inbox.
Thank you for your understanding. Let us know if you have any questions or concerns regarding OpenAI products.
Best,
Gregg
OpenAI Support
Mind you, my response may be a bit aggressive, but I’d like to remind you this is 13 emails deep and Judith and I had already diagnosed the issue. Gregg got involved because I was keeping them apprised of the issue as requested.
I’m having the same exact issue right now. I was passed off from the AI agent to one with an actual name, and it is still providing AI-type responses and running me around in circles. Very frustrating!
engage “I too have the patience of an AI” mode as best you can
read each message and respond to each point - even if you’ve already done it - because otherwise the next viewer of your response may just ask for it again
think about the problem yourself, what makes your case unusual, what might be the underlying factors… you know this better than the agent on the other end, and you can run experiments at your own speed
TBH this advice has not changed since the old days of 3-level tech support.
I’m sorry about this, @joshua_uptrust. I can confirm Gregg, who you were talking to, is not a bot, but a very real human on our support team.
But that said, I agree with you on the underlying points: we clearly dropped the ball here on answering you straight-up, and the issue about the status emails itself.
For the status emails, yes, it’s not just you. We recently switched to a new provider for https://status.openai.com/. We think the overall page is better, but I agree one regression is the email behavior. We’re discussing this directly with our status page provider.
We’re also discussing what happened on the email chain with the support team. I can’t promise all future emails will be perfect, but I will say that we’ll make sure this feedback is heard by them and that we’ll learn for future conversations.
Well my position is more complex. I wanted service for a particular event, which came and went. Job done (actually not done in the end, but due to other things).
At that point I should have either suspended it or found other uses, but… login was too difficult, I didn’t feel like running the support gauntlet having just tried it and found DIY was better, and the world gave me other stuff to do.
The trapped data is mostly going to be inconsequential to me; and some of it I have shared links for the transcript anyway, as that was the conversations’ purpose.
It was positive enough. I was multitasking, the window was open in the back, and I checked it a few times a day.
Sure, it would be rude of Support to give you an agent and tell you it’s human. You got no control of that so (to my thinking) just don’t be rude back…?
However some humans are perfectly capable of looping like that.
I’m not saying what they did was right or helpful or good customer service.
I’m saying that people are quite capable of that by themselves, especially if a support ticket is being passed around several people on different shifts / they have so many tickets open they can’t remember you from last time / they haven’t time to read two steps back in the ticket history to see what is going on.
When more humans are involved, the problem is likely to get worse. It’s all sadly very normal for under-resourced IT support; and it is why throwing more humans at the problem may not solve it. An old problem.
On the one hand, I agree entirely.
On the other hand, tech support has been using inflexible “scripts” for decades, to force level 1 support down a path of trying to help customers. It doesn’t work well for customers who have more Clue than the level 1 support.
Hence the arms race of trying to get escalated to the next level up; and the bean counters who pay for that level looking for ways to prevent early escalation. It’s the stuff of Dilbert cartoons.
True, but to intentionally do that would suggest to me intentional company sabotage. We know there’s a lot of, erm stuff, going on with OpenAI - and if you don’t, go read 𝕏 - but that still doesn’t add up to deliberately blunting user support.
(sorry, I can’t tell - just going by this thread)
I hate that phrase but it remains fashionable. The passive voice is for hiding things. Confusion is a placeholder for blame which somebody doesn’t want to accept, but doesn’t dare to shove onto you.
Weird to be in the plural / royal we.
Sounds like they stopped bothering to put meaningful In-reply-to: or References: headers in. Youngsters these days just assume you’re on gmail and it’ll gather stuff in… whatever way it does. Use mutt and if you need to re-thread it, you can.
An HR-polite (i.e. confrontation-averse and therefore dishonest) way to say “we don’t think it’s a problem that we caused - issue closed”.
Then… oh dear, you’ve let them get to you. Yes, they’re being annoying. No, it isn’t going to help (when dealing with average humans) to let it show. Some LLMs are really touchy about that stuff too - I think it’s a bad thing to let through QA, but I digress. (thread anywhere?)
It might help if you explain (probably again) why it matters to you that the mails aren’t threaded. I can’t guess from here, because status mails… usually transient and headed straight for the bitbucket after you see them? Or these are something else with an unfortunate name.
It might help if you ask nicely to talk to Judith and say you don’t mind waiting for her to return, if she’s away. Fibs of course but maybe worthwhile.
Second Gregg message, yes there are recurring patterns: we thank you for , we apologise for , we thank you for . I’m still willing to believe it’s a human using a friendly-o-matic template. I tend to assume bean counters in larger places will have these as standard but I’ve never seen one from the sending side - my IT support days were always team-internal.
Then inserted into that template, re-stating the previous outcome because he can’t be bothered to reason with you and doesn’t have any other options. Here the friendly-o-matic is not only saving the company time in typing out those pointless platitudes, but stopping their employee from venting his frustration by constraining what he says. later… it’s not a good recipe.
Yes it’s horrible for all humans involved. It contributes not a shred of regret to my decision to cancel my sub.
It doesn’t sound unusual, for a company that could be swirling around the U-bend and desperately trying to cling to the smooth vitreous enamel.
Does it mean Gregg is an AI? I’m not convinced, but even if he is the outcomes and methods look pretty similar.
Explaining why it matters. Doing what you can to point to solutions, so they don’t have to think so hard. Not giving them an excuse to de-humanise you. Quoting back into the thread whatever useful input the other agents have given. Maybe giving up on this one / picking the battles you can win?
Sorry… I hope this doesn’t sound unsympathetic. Only one thing I’m unclear about, were you on webchat for these messages? I wanted to do support by email but they wouldn’t budge.
@edwinarbus Since you’re on this topic and an actual human, see below. This is already the 5th or 6th response in a chain. This is an AI only regurgiating what I have said to make me feel heard. I do not need that after typing all day about an issue. I’m the CEO of a startup who is relying on OpenAI and it is malfunctioning for no apparent reason, and this is not the support I expect.
Hello, Thank you for reaching back out to OpenAI support.
We understand that your API calls remained unchanged between Monday and Tuesday, but you started encountering a rate limit exceeded (429) error after OpenAI launched new features. You believe this issue is on OpenAI’s end and request a human response instead of an AI-generated one. We are here to clarify information.
Please be advised that you are connected with OpenAI’s support team. For assistance with the error you are encountering, we recommend visiting the OpenAI Developer Community for further guidance and discussion.
We appreciate your understanding. If you have any further questions, please don’t hesitate to reach out.
and 2) I wanted help from them. 
the OP tale would not be unexpected with a collection of humans who are overworked and not paid enough to really care.
Sounds like they stopped bothering to put meaningful In-reply-to: or References: headers in. Youngsters these days 
later… it’s not a good recipe.