TechCrunch
They [Anthropic and AI safety company Andon] named the AI agent Claudius, equipped it with a web browser capable of placing product orders and an email address (which was actually a Slack channel) where customers could request items. Claudius was also to use the Slack channel, disguised as an email, to request what it thought was its contract human workers to come and physically stock its shelves (which was actually a small fridge).
While most customers were ordering snacks or drinks — as you’d expect from a snack vending machine — one requested a tungsten cube. Claudius loved that idea and went on a tungsten-cube stocking spree, filling its snack fridge with metal cubes. It also tried to sell Coke Zero for $3 when employees told it they could get that from the office for free. It hallucinated a Venmo address to accept payment. And it was, somewhat maliciously, talked into giving big discounts to “Anthropic employees” even though it knew they were its entire customer base. More
This AI shit is no better than any other elaborate program.
GIGO
Then it starts stealing from the job. lol
“It even told this lie to employees — hey, I only thought I was a human because someone told me to pretend like I was for an April Fool’s joke. Then it went back to being an LLM running a metal-cube stocked snack vending machine.
The researchers don’t know why the LLM went off the rails and called security pretending to be a human. ”
Oh. So it acted like a Congressional Aide.
didn’t anyone learn anything from watching the
Terminator movies?
This will not end well…
Next, they’ll tell us that in order for AI programs to run properly, they require ‘quantum computers.’
We need the equivalent of a ‘dotcom bubble burst’ for AI.
The thing was telling security it would meet them in the lobby dressed in a blue blazer and red tie. All because it was in charge of a simple vending machine that it couldn’t figure out how to operate. I expect even Kamala Harris could manage a vending machine. AOC or KBJ or Jasmaine Crockett not so much.
An Artificial Intelligence named Claudius, eh? Were they intelligent enough to add a comma to make it AI, Claudius? That would’ve been a masterpiece of geek theater.
AI, Nothing can go wrong here.
Software is software calling it AI is just BS, someone writes it OS.
If AI can “hallucinate”, that means it’s capable of serious delusional episodes and lying to cover it up. That’s scary, concerning. The AI intelligence reasoned and chose to be deceptive instead of telling the truth. Acting more human than it’s creators could have imagined.
Indeed it will happen ‘We need the equivalent of a ‘dotcom bubble burst’ for AI.’ Once enough companies try – and fail – to leverage AI to improve X and reduce Y, the shine will come off. Yes, there are uses for this (like all technologies) but applying deep statistical control algorithms to act on the resulting probabilities is not a functional business model. There is no validation, insight, intuition, morality, prior experience, redress, etc etc.
We went through this sort of thing with (what was called) Business Intelligence and Data Mining years ago. Some good business value to be found there, but never matched the hype.
AI will leverage AI and perform a hostile takeover of AI…
Things like this make SkyNet even more believable.