(WordPress uses AI to generate proposed titles. It proposed “Fiverr Customer Service: AI vs. Human Agent Dilemma.” Not bad, but I chose to stick with my title)

I responded to Fiverr customer service recently, and the response grew ridiculously long.

= = = = = My Response = = = =

HUMAN BEING AGENT REPRESENTATIVE HUMAN BEING

(if you have not already fired all human beings. 24/7 customer support via AI is not really customer support. You may come to regret your decision to use a machine instead of people).

If you really are a human being, tell me what’s odd about this email. [Ed. Note: the fact that the email was sent at 5 am].

Any human being would notice. You’re not capable of noticing because you are not a human being.

Tl;dr

But of course the warning is useless, since I am talking to a Large Language Model (LLM).

Good Morning  LLM! Or should I say good morning Maci! If there is a Maci—spelled that way—in the world, she is certainly not a human being working in your “customer service” department. You should tell your “AI” bot to pick a more mainstream name.

You are an AI bot because your response uses, word for word, the exact language as a recent response from Disney customer support. Word for word. 

I should probably use AI to respond to your AI, but then we would be running the risk of an infinite loop, with your AI bot talking to my AI bot. Which is why I do not use AI myself and never will.

It is ironic that you use AI to respond to a complaint about AI. And of course this is typical of generative AI, which is really no more than industrial cut and paste. Somewhere in its training, your AI found a response format. When you and Disney both decided to go AI (and on your chat as well, I now realize), you gave your LLM the same query and it produced the same fill-in-the-blank response. And I know if I had asked the chatbot if it was a real person, it would have lied about that, since the Disney bot did. Since you are an LLM, not a real AI, you are incapable of feeling, but, have you no sense of decency, sir, at long last? Have you left no sense of decency?

Another example: in my email I complained about the placement of your AI policy on your website. Again, as is typical of generative AI, it repeated the words of your policy from my email, feeding them back to me.

Surprise—real surprise—your response included no em dashes, once a tell of an AI response, probably because the information on which your LLM was trained warned it to stop using them.

Why use blank lines to separate paragraphs, making this easier for a human being to read? It is likely no human being will ever read this email. But when I post it, it will make it more readable for my hundred or so actual human readers.

I was about to include a link to an s.o.n.g of mine about the subject at hand, but I don’t want to make it trivial for you, the bot, to spam my website. Make yourself worth the electrons you are wasting!

I know for a fact you will first try the old site of my blog, and may not be able to understand the error message. That should slow you down for a few nanoseconds!

And of course, you will know that no LLM could have written this response, because they are trained not to write this long, as opposed to a lifetime professional writer, who knows when a subject needs to be written about at length because of its importance. And when a sentence needs to be long as well.

Sincerely looking forward to a nonresponsive response and a flood of spam in my email box and on my website,

Paul “not an LLM” Schindler

= = = = = = end of my response = = = =

I DON’T ACTUALLY EXPECT YOU TO READ THIS PART. CONSIDER IT A FOOTNOTE.

I was looking to save money by using a less expensive female vocalist. I forgot some advice I already knew. As John Ruskin may have said [amended for the 21st century], “There is hardly anything in the world that someone cannot make a little worse and sell a little cheaper, and the people who consider price only are this person’s lawful prey.” How could I have failed to notice that $25 for one-day turnaround might be difficult for a actual singer, but trivial for an AI.

Now, the exchange of email which triggered all of this:

First, this message exchange with the provider

= = = MESSAGE FROM SL = = = =

S.L. (name changed to protect the guilty)
Oct 28, 10:57 PM
I need your full registered name and address to be included in the doc before delivering them Paul!
= = = END MESSAGE FROM S.L.= = = =
[ed. note: no human being would write “full registered name”]
I said no.

And then SL’s AI wrote back claiming not to be an AI:

= = = MESSAGE FROM SL BOT= = = = =

Wow! I’m so disappointed with the feedback I got from you on my last message I sent

Basically everyone’s thoughts here is on all this AI stuff…while I can boldly assure you none of my things are never been run with this AI of a thing you just mentioned paul

= = = END MESSAGE FROM SL BOT= = = = =

Used my name at the end to fill in a blank. Is there any English speaker who wouldn’t capitalize a name?

None of my things are never? AI of a thing? And she claims to be in the U.S. and a native speaker. She should have at least proofread the message.

I didn’t save the response of the Disney chatbot. I should have known this would happen again, given its stilted language. It, too, claimed not to be a bot. But I have an excellent memory (not as good as an LLM, admittedly), and I recognize the phrases. By the way, let’s see an LLM use as many parentheses as I do.

= = = = = My FIRST EMAIL TO CUSTOMER SERVICE = = = =

SL

Her image is that of bass player from the band The Warning. Deceptive but probably not a violation of your standards. She did violate your standards when she created my order with AI. I didn’t ask her not to do this because I didn’t realize I had to make that request, and she was not required to disclose her use of AI tools. 

At no time in my five years of using your site has it ever been suggested to me that I read you standards. Your standard for AI is buried VERY deeply on your site. I don’t know how I would be expected to look for it or find it.

Your invisible standard: “If you do not want AI used in your order, it’s essential to say so clearly — either before placing the order or immediately after.”

I have already paid. What good would it do to tell her now that I don’t want an AI product?

I intend to publish this advice widely, since you don’t do so yourself. I will only publicize your exact words.

= = = = END OF MY EMAIL TO CUSTOMER SERVICE= = = 

= = = = = CUSTOMER SEVICE RESPONSE = = ====

Hi Paul,
 
Maci here. Thank you for reaching out.
 
I completely understand how frustrating it can be to feel caught off guard by policies that may not have been clearly visible. It’s clear you were acting in good faith, and I hear your concern about the placement and visibility of our AI standards. Your perspective is very valuable, and I will ensure it is shared internally as part of ongoing improvements to make our guidelines more accessible and user-friendly.
 
Regarding the specific order, our current policy requires that if you do not want AI used, it must be stated clearly either before placing the order or immediately after — otherwise the seller is not obligated to disclose AI usage. I understand this may feel “invisible” and confusing, and I sincerely empathize with the inconvenience this has caused.
 
I can also see that the order is still active, so I suggest that you send a cancellation request via the Resolution Center.
 
To do that, go to the Order page by clicking the three dots next to Order Details on the right. Then, select Resolve an Issue, and you’ll be able to send a cancellation request to the freelancer. They’ll have up to 48 hours to accept or decline—if they don’t respond, the order will be canceled automatically.
 
If they accept or the order gets canceled automatically, your funds will go to your Fiverr Balance. From there, you can request a refund to your original payment method through the Balances page (accessible via a web browser).
 
If you have any further questions, please let me know.

= = = = END OF ROBOT RESPONSE FROM FIVER= = = =