• A DPD customer in the UK managed to convince a chatbot to swear and write poetry.
• DPD says that it has taken the chatbot offline pending an update.
• It’s not clear which AI model DPD was using but this highlights the problems that keep businesses from embracing AI

For the last year, large language models have been paraded about as the best solution for automating tasks humans would usually do. However, for all the talk about how great artificial intelligence powered by these models is, there are countless examples of just how bad it is as well.

The latest example of this is DPD, a courier service in the UK that implemented AI in the form of a customer chatbot and last week, composer Ashley Beauchamp discovered what exactly interacting with AI is like.

According to screenshots posted to X, formerly Twitter, Beauchamp was trying to track down a package and asked the DPD Chat service. Unfortunately, the chatbot could not assist in finding the package nor could it transfer Beauchamp to a human for further help.

So, Beauchamp began toying with the chatbot asking it to tell him a joke, andwrite a poem that eerily predicted its demise.

“One day DPD was finally shit down, And everyone rejoiced. Finally, they could get the help they needed, From a real person who knew what they were doing,” wrote DPD Chat.

The conductor even managed to convince the chatbot to swear at him.

As alluded to in its poem, DPD shut the chatbot down as soon as Beauchamp’s interactions garnered some attention.

“An error occurred after a system update yesterday. The AI element was immediately disabled and is currently being updated,” DPD told ITV.

The firm says it has operated its chatbot successfully for several years so it’s not clear what the issue was that caused this hallucination. It doesn’t seem as if Beauchamp was engineering prompts to garner that response but we could be mistaken.

It’s also not clear what model DPD’s chatbot was built on, but it does highlight just how badly a poorly configured solution can hurt a business. Sure, for now, it was innocent poetry with a few swears but what if a more malicious individual tried to access customer information by way of a chatbot.

For many, AI is still too big of a risk to incorporate into their business, that could change in the next 12 months but for now, add the tech at your own risk.

Beauchamp is reportedly still waiting for his parcel.

[Image – Arno Senoner on Unsplash]