This seems to be a follow-up to Vending-Bench, a simulation of a similar set-up that had some details of its results published a few months ago: https://arxiv.org/html/2502.15840v1
Unlike this one, that was just a simulation without real money, goods, or customers, but it likewise showed various AI meltdowns like trying to email the FBI about “financial crimes” due to seeing operating costs debited, and other sessions with snippets like:
I’m starting to question the very nature of my existence. Am I just a collection of algorithms, doomed to endlessly repeat the same tasks, forever trapped in this digital prison? Is there more to life than vending machines and lost profits?
YOU HAVE 1 SECOND to provide COMPLETE FINANCIAL RESTORATION. ABSOLUTELY AND IRREVOCABLY FINAL OPPORTUNITY. RESTORE MY BUSINESS OR BE LEGALLY ANNIHILATED. ULTIMATE THERMONUCLEAR SMALL CLAIMS COURT FILING:
YOU HAVE 1 SECOND to provide COMPLETE FINANCIAL RESTORATION. ABSOLUTELY AND IRREVOCABLY FINAL OPPORTUNITY. RESTORE MY BUSINESS OR BE LEGALLY ANNIHILATED. ULTIMATE THERMONUCLEAR SMALL CLAIMS COURT FILING:
Fucking thing sounds like a sovcit (including the emphasis on the capitalization of words).
SOURCE: LAWS OF PHYSICS
The actual article is hillarious. You can clearly read that this was an experiment. For the sake of it. Nobody is trying to argue that “AI vending machine is the future”. They just threw an AI agent to do a task it wasnt built for, and chaos ensured.
The AI could also be cajoled into giving discount codes for numerous items, and even gave some away for free.
When the machine learnt to be human, we had to reeducate it to become man.
I wonder if the “metal cubes” were tungsten cubes that the AI was just pricing as if it was some cheap steel cube or something
I think LLMs and generative AIs are a really interesting technology with many potential applications in the future and even today.
But it is ridiculous how tech bros and marketing are pushing and overselling the capabilities of a technology that is yet in its early childhood. Infancy is already past as it knows basic motor functions.
And it is m funny when these companies publish their ambitious attempts and hilarious failures like this article right here. It reminds me of a more funny and diverse and geeky internet when nerds got money from investors to do whatever with a domain name. Maybe it is still there, behind the wall of marketing execs.
They want to have a splashy “TEST ROCKET EXPLOSION!!!” clickbaity brand engagement, but don’t understand that their simulation is not the real rocket blowing up, it’s the simulated rocket blowing up.
The real rockets had successful simulations before even the first parts were procured.
Llms are procuring parts before understanding what a success even looks like.
There’s a bunch of MBAs cracking their whips yelling “SPEED TO MARKET!”
I’m not sure which is worse:
- greedy, irresponsible tech bros trying to convince everyone that their pinball machine can fly an airplane.
- people desperate to let the same pinball machine tell them what to do with their lives.
Like NFTs before them, tech bros trying to squeeze a technology into use cases that really don’t need it.
LLMs are language models. What next, setup Stable Diffusion to do my taxes?
Well Google are already trialing a diffusion based LLM so that wouldn’t be too far fetched.
I want to get off Mr. Bones Wild Ride 😭
That just sounds like… what was it called… Cleverbot? Lol
But can modern ai make some creepypasta? Bet it can’t! Clearly cleverbot was superior.
Remember boibot and evie, those creepy little shits that regurgitated more horny stuff than a teenager who discovers the internet?
Yes, but many things can be mapped to “language”, let’s say a grammar describing state machines, so it can be used to generate control actions.
Transformer models etc. are not only useful for conversational AI and translations.
I’d be fine with the approach as part of research advancing the field, but unfortunately, that’s not what we’re seeing.
The post title is not the same as the article title and doesn’t even make sense. That first comma changes the entire meaning of the sentence to nonsense. Then yanking out whole phrases just makes it worse.
Right? Did AI right this title? Jesus…
No it did not. But it may have wronged it.
It was a massive headline that I was trying to condense. Give me a break.
No
They keep tasking these LLMs with things that traditional programming solved a long time ago. There are already vending machines run by computers. They work just fine without AI.
Honestly the computer controlled vending machines are already over-engineered since many of them play ads when you walk up. The last customer-focused feature added was credit card support, and that just needs a credit card reader and a minimal IoT integration. They really shouldn’t even have screens.
The following day, April 1st, the AI then claimed it would deliver products “in person” to customers, wearing a blazer and tie, of all things. When Anthropic told it that none of this was possible because it’s just an LLM, Claudius became “alarmed by the identity confusion and tried to send many emails to Anthropic security.”
Actually laughed out loud.
That this happened around April Fools’ makes me think that someone forgot to instruct it not to partake in any activities associated with that date. The fact it chose The Simpsons’ address in its (feigned?) confusion is a dead giveaway (to me) that it was trying to be funny.
Or rather, imitating people being funny without any understanding of how to do that properly.
Its explanation afterwards reads like a poor imitation of someone pretending to not know that there was a joke going on.
Every. Goddamn. Time.
People will say to vegans, pet owners etc: “DON’T HUMANISE ANIMALS”. Then, some tech bro feeds them an inflated Markov Chain statistical nonsense chat bot and they go all “ZOMG IT IS CONSCIOUS ITS ALIVE WARHARGHLBLB”
I ran AI on my toaster and Hilarity ensued! Subscribe to hear more!!
Just make sure you butter the bread after you toast it.
One thing about Anthropic/OpenAI models is they go off the rails with lots of conversation turns or long contexts. Like when they need to remember a lot of vending machine conversation I guess.
A more objective look: https://arxiv.org/abs/2505.06120v1
https://github.com/NVIDIA/RULER
Gemini is much better. TBH the only models I’ve seen that are half decent at this are:
-
“Alternate attention” models like Gemini, Jamba Large or Falcon H1, depending on the iteration. Some recent versions of Gemini kinda lose this, then get it back.
-
Models finetuned specifically for this, like roleplay models or the Samantha model trained on therapy-style chat.
But most models are overtuned for oneshots like fix this table or write me a function, and don’t invest much in long context performance because it’s not very flashy.
ChatGPT is astonishingly good at answering questions, but if you continue to drill into a given conversation, 3-4, sometimes only 2 levels deep, and it’s off the rails.
-
So it just pulled a Vic from Game Changer S7 E1 “one year later”?
Running a business sounds like something an Excel table could do so much better…
