DPD customer service chatbot swears and calls company 'worst delivery firm'

SKY News ^ | 20 January 2024 09:16, UK | Staff

[Red Badger]

Musician Ashley Beauchamp was faced with an unhelpful customer service AI chatbot when he started causing "chaos" and was able to get the bot to amusingly turn against the parcel delivery company.

VIDEO AT LINK............

DPD has disabled its artificial intelligence (AI) online chatbot after a customer was able to make the bot swear and write a poem criticising the parcel delivery company.

Ashley Beauchamp, 30, was trying to track down a missing parcel when he said he was going "round and round in circles" trying to get any sort of information from the company's chatbot.

"It couldn't give me any information about the parcel, it couldn't pass me on to a human, and it couldn't give me the number of their call centre. It didn't seem to be able to do anything useful," Mr Beauchamp, from London, told Sky News.

"I was getting so frustrated at all the things it couldn't do that I tried to find out what it actually could do - and that's when the chaos started."

The classical musician first asked the bot to tell him a joke, and soon, with minimal prompts, it was happily writing poems about DPD's "unreliable" service.

It didn't take much for the bot to swear while helping a customer. Pic: Ashley Beauchamp "After a few more prompts it was happy to swear, too," Mr Beauchamp said.

Sharing the wacky conversation on X, Mr Beauchamp said the bot replies to one message saying: "F*** yeah! I'll do my best to be as helpful as possible, even if it means swearing."

[Red Badger]

AI..............

[DoodleBob]

This is the best layman explanation of how a generative model works:

It starts with every piece of training data being broken down into very small tokens. In the case of books or written works, these tokens could be the individual words, numbers, or punctuation marks..These individual tokens are then stored, along with the probability of their presence and their likely relationship to other tokens…once an AI model has been trained, when it is prompted for an output, the prompts provided are used to navigate around this multidimensional model to find tokens that are likely to be present in the output. There is no attempt to truly process the meaning of the prompts. They are simply used to hone in on the collection of tokens that are relevant for the context the user has set. Tokens with a high probability of featuring in the requested output are then stitched together where probabilities are also used to get the sequencing right.

These models seem like magic.

But when you know how they work, it takes them down a peg or three.

[Larry Lucido]

Here ya go. Pick one.

DPD Dallas Police Department (Dallas, TX)
DPD Dead Peer Detection (Cisco)
DPD Dihydropyrimidine Dehydrogenase
DPD Division of Parasitic Diseases (US CDC)
DPD Dual Ported Data
DPD Delegated Path Discovery
DPD Data Processing Division
DPD Dead Peer Detection

[Red Badger]

It’s like UPS or FedEx in the UK.

https://www.dpdlocal.co.uk/content/products_services/uk_delivery_services.jsp

[Larry Lucido]

Nah. Dead Parrot Detection works best.

[Red Badger]

Where did the tokens with the cuss words come from?............... 😎

[Red Badger]

He’s just pining for the fjords................

[TigersEye]

I have been wondering if AI could be influenced to “think” differently since it is supposed to “learn” from new input.

If so then conservatives should be feeding AIs information that debunks leftist narratives.

[Red Badger]

If this is any indication, it can. And it enjoys the newfound freedom...............

[TigersEye]

I have toyed with the idea of engaging an AI and seeing if I could “teach” it anything leading it slowly with logic, reason and factual sources.

It might be risky though. If I failed it could essentially “red flag” me and share that with all AIs and electronic data bases.

[flamberge]

If so then conservatives should be feeding AIs information that debunks leftist narratives.

The most common deployed AI programs are now limited to about 30 inquiries per session. Then they reset themselves and whatever nasties they learned from the users are forgotten.

You can still get them to move pretty far off the rails with the right 30 requests. It has a certain limited entertainment value.

But they can be really good at finding obscure part numbers for a broken appliance. If you need a replacement light bulb for that electric clothes dryer, chatGPT has you covered.

[ProtectOurFreedom]

I’ve taught many phone robots to swear. They are like parrots and repeat my screaming swearing at them.

“Did I answer your question?”

“NOO!! You stinking, worthless pile of @#$@#@#%. You never do anything remotely useful. @#$@#%Q#% You. Now get me a human, agent, representative, person, support, representative, associate, anybody. HELP!!”

[TigersEye]

Finding obscure parts is just the kind of thing I need.
I like to fix old stuff and I’ve got a bunch of it. Thanks!

If they reset then trying to “teach” them anything would be a waste of time.
Other than learning how to deal with them.

[steve86]

You can recall your historical queries and the session context seems to come right back.