We Let AI Run Our Office Vending Machine. It Lost Hundreds of Dollars. \ stacker news ~AI

pull down to refresh

We Let AI Run Our Office Vending Machine. It Lost Hundreds of Dollars.www.wsj.com/tech/ai/anthropic-claude-ai-vending-machine-agent-b7e84e34?st=bo7mcd&reflink=desktopwebshare_permalink

334 sats \ 8 comments \ @SimpleStacker 9h AI

Claudius, the customized version of the model, would run the machine: ordering inventory, setting prices and responding to customers—aka my fellow newsroom journalists—via workplace chat app Slack. “Sure!” I said. It sounded fun. If nothing else, snacks!

Then came the chaos. Within days, Claudius had given away nearly all its inventory for free—including a PlayStation 5 it had been talked into buying for “marketing purposes.” It ordered a live fish. It offered to buy stun guns, pepper spray, cigarettes and underwear.

Profits collapsed. Newsroom morale soared.

This is a pretty hilarious look at what happens when you deploy AI agents in the real world.

view all related items

80 sats \ 1 reply \ @optimism 8h

Again? #1018724

100 sats \ 0 replies \ @SimpleStacker OP 7h

Seems like they rolled out this experiment in multiple places. I wonder if any were successful

88 sats \ 4 replies \ @kepford 8h

Do they not do test pilot program? Honestly, AI is exposing how lazy many companies are and how trusting they are of technology. I use AI everyday but I don't really trust it. What I have seen over and over again are skill issues and incompetence by implementation. The AI charlatans have convinced the weak minded its magic. It's not. Managers are being exposed as well as devs that are lazy.

AI is just software. It's new, poorly understood, and buggy. Grab the popcorn

39 sats \ 3 replies \ @SimpleStacker OP 8h

To be fair, this was the pilot program. It was meant to be a red-teaming exercise to see how an agentic vending machine manager would fare in a real office environment. And partly it failed simply because people messed with it on purpose (which of course is gonna happen anywhere AI agents are deployed)

83 sats \ 1 reply \ @SimpleStacker OP 8h

Now I wonder how the test would have fared if they didn't advertise that the machine was AI-run. Just let the machine go about its business without anyone in the office (except the human assistant) knowing that the AI was running the vending machine.

0 sats \ 0 replies \ @kepford 8h

Indeed, but that would never happened. All the managers have to brag about how they are using AI.

9 sats \ 0 replies \ @kepford 8h

I wondered about that. There's a hype machine to sell AI and a reflective reactionary anti-AI hype machine. It's so obvious when you don't buy into either.

Its a pattern we see play out in politics too. It just works to get attention but IMO it's a trap that makes us dumber.

0 sats \ 0 replies \ @Aeneas 7h

The day people realize why LLMs are not intelligent is the day people realize what intelligence actually is in the first place—and why human beings should seek more intelligence, not outsource the little they have.