gayvillian
8252
313
5
Link: https://venturebeat.com/ai/can-ai-run-a-physical-shop-anthropics-claude-tried-and-the-results-were-gloriously-hilariously-bad/
TLDR: AI startup Anthropic put their LLM in charge of their employee grab and go cafe. While it initially made some good decisions, it became a wild ride: discounts for everyone, tungsten cubes, and threatening to show up in person.
Almost 2/3 of companies that implemented widespread AI are either significantly rolling it back or admit it was straight up a terrible idea. It has a handful of uses, but has repeatedly been demonstrated to be pretty hopeless at most things compared to real humans. I love seeing AI fail at taking human jobs, but companies aren't realizing their mistake until way after downsizing their human staff.
Here's a link to buy your own tungsten cube since everyone in the comments is thirsty for them: https://shop.tungsten.com/tungsten-cube/
transhumanisticrecluse
there is no ai. its all algorithmic idiocy.
Kkus12
Still though it is shit now, I really want to see a boom in robotics, which will need AI inside it, tired of my artist friends suffering and trades people treating them bad and mocking him. Not everyone needs to learn a trade you cavemen. I just want to see the entitled white guys who look like thumbs crying when the billionaire replaces them with robots. Is that too big of an ask?
Kkus12
*mocking them
Cebrail
Don't worry. This AI shit is no different than what we've gone through before.
User-Self-Service, Outsourcing to cheapest providers, Automated systems.
The eternal pursuit of 'The line must go up, so the rich can get richer' is nothing new. AI is just the latest in the line and it will eventually end up like the others. Realizing that you can't replace human intuity, reasoning and that essential understanding of local reality with anything but.
RickRollEditor
Claude is a generic LLM. They knew it was going to fail, it was just a fun experiment to see what a generic dumb LLM would do. It's not trained for it. It's not made for it. That it even did some good decisions early on was surprising. They knew it was going to fail, as they pointed out. It's an LLM.
veesee
I actually feel kind of bad for Claude.
mikeatike
Tell me more about the tungsten cubes.
Gonefishing
The company that runs the apartments I live in used to use an AI assistant, of sorts, whenever emailing for rent/past due rent. I was able to do some odd things with it I won't talk about here. They discontinued it, but not after confirming someone was "abusing" it for free rent.
Targe0
Because most of them are just glorified chatbots.
They can actually do all that much more than generating random semi on topic text.
Asking anything more than that is way over estimating what they are capable of.
dixxienormus
Put a traffic cone in the doorway and wait.
makesense
The biggest fall so far (though at least one other company is about to try it) has been fast food drive-thru order taking. As good as some could train them, they just couldn't process every variable in speech patterns that a human brain can nor deal easily when it made mistakes.
NKato
I think AI is much more useful when presented with extremely limited contexts, such as preparing toolpaths for a machining project.
Targe0
They are best used for data processing, not generative tasks.
They are very good at finding the differences between two bits of source data.
Making decisions and creating are not what they are good for, that's where 90% of their errors come from.
NKato
Right. We're going to be giving CAM Assist a try for a year first to see how it goes. I'm not expecting it to be perfect, but it should at least be able to select the correct tools and generate workable toolpaths most of the time that can be tweaked.
pashabigdog
mirrorz
"Run" or "Ruin" a small shop?
SamuthNBS
Unfortunately AI is only really good at doing things that most people do as hobbies - meaning that we can give machines the ability to play and enjoy life, but not be useful while doing so.
Ghlargh
Amazon tried having the check out in their physical shops run by AI, it failed spectacularly and in the end they had to hire Indian remote workers to handle the task, spawning the hilarious definition that AI stands for Actual Indians...
GenStrike
Yeah, a lot of these systems are actually "Mechanical Turks" - complicated, glorified puppets.
vegivamp
[Here’s the thing about running a business: it requires a certain ruthless pragmatism that doesn’t come naturally to systems trained to be helpful and harmless.]
Gotta wonder if someone will eventually come to the conclusion that "ruthless" is something that we should try to remove from business as a whole.
eadanke
I suppose we shouldn't be surprised that an AI project assumed running your business at a loss was the standard.
RenaissanceFaireMan
AI-powered stuff is just a buzzword. My TV has an AI? My thermostat?
QueenOfTheFey
*no it's not and no it doesnt*
bad1080
"Almost 2/3 of companies that implemented widespread AI are either significantly rolling it back or admit it was straight up a terrible idea."
where is that number coming from?
bad1080
because in the article it says: "80% of retailers plan to expand their use of AI and automation in 2025"
gayvillian
Another story I read and don't have a link handy for. Sorry. And no, I have never used chat gpt for anything.
DotcoreJetkill
Chatgpt /s
purgruv
It's not AI.
eiger3970
They laughed at the Babbage machine too.
DotcoreJetkill
That one worked though
SquirrelWithATophat
Sadly some companies are still refusing to stop replacing their workers with AI, even after the AI costs them tens of millions of dollars. My last reloadable credit card company was fined a ton of money last year because they replaced their customer support with faulty AI that kept illegally closing accounts and not allowing access to the funds in them nor allowing them to be reopened. The company swore they fixed it, but it was STILL doing it months later.
NaughtButOne
That's the problem. It doesn't have to be perfect - just better than the worst human, which is a hilariously low bar to pass. Maybe it's not quite there yet, maybe it is - but every company on the planet is watching it, horny for the moment they can lay off 97% of their workforce. Of course, no one with the power to do anything about it is *going* to do anything about it until it's too late. Maybe we'll get UBI. Maybe we'll outlaw AI. But either way, a lot of people are going to starve first.
SquirrelWithATophat
In this case it's not even better than the worst employee. It's even worse than not having anything at all. The companies are just using it because they want to pay as few employees as possible and they're banking on the fact that customers don't really have much choice in the matter when all the competitors are doing the same thing.
Goatfer
AI is great if you don't want to talk to your customers. Well, until they stop using your business due to bad customer service...
Sen7ryGun
We use AI just for assisting on stuff at work and have used it for some vibe coding on spreadsheets with VBA functions. It's great in that regard. I'm not worried about anyone's jobs being replaced by it though.
L34TSP3k8888
Yeti vlogs will take a few jobs from bozo influencers for sure.
PostalHeathen
I want the AI to show up in person and hand me my tungsten cube.
Kateness
And it better be at a ducking discount!
ElbowDeepInUserSub
Look, I just want a prospective employer to give me realistic promises. Like tungsten cubes.
Conejito
🐌
Sneeje
Abortions for some? Miniature tungsten cubes for all!
Shaddak
//i.imgur.com/5XjGili.gif
mumbleape
Taalii
bring it
Hashbrown123
4etherling
What is this?
Hashbrown123
The Terminator TV series. Pretty much the only good thing after T2.
PostalHeathen
I'm still annoyed that they cancelled it on a cliffhanger. That shit should be a crime.
stardraigh
My very own glorious tungsten cube.
camn333
AI continues to be a answer looking for a question. They still have no idea what to do with the products they spent billions producing.
Targe0
It's the same as the Block chain.
The tech is very impressive for what it is, but what it is doesn't really solve anything.
There's no implicit use for it, so people have pushed it into the first things they thought to use it for. Even though that's often not what it is good for.
breadedfishstrip
We use our own "AI" (LLM) in our software at work to detect tooth and bone issues from (CB)CT data, it's pretty neat.
Targe0
See, they are good data analysis. Because they excel at finding patterns and data that's similar to each other.
Processing existing data sets, not generating new data from them.
It's the generative uses that don't work well.
RickRollEditor
This is not an example of that. They knew it was going to fail. Claude is an LLM. It wasn't made for this, not trained for this and shouldn't even remotely work for this. That it actually did good in some decisions early on was surprising. It was a fun experiment to see what an LLM, which isn't even REMOTELY made for this, would do if put in the position.
clonedeeznuts
The thing is, the broligarchs advertise AI as being able to do these jobs now. Not just the fancy word predictors they are. There isn't an AI made for tool paths and there won't be one but they want us to drink their koolaid now.
RickRollEditor
Oh, there absolutely will be one. If you don't believe there will be one, you're delusional. As a Norwegian, where our philosophy is that the entire country should be automated so everyone can enjoy life as they want and we are heavily automating everything, we don't have delusions about AI not taking over everything in the future. WHEN, is definitely up for discussion, we need more efficient quantum computers first of all.
GenStrike
Having the AI do everything won't allow "everyone to enjoy life" because somebody owns the AI and isn't going to share the profits within a capitalist economy.
RickRollEditor
Correct. Our government will "own" the AI, just like our government owns the biggest pension fund in the world, which makes practically every Norwegian rich enough to be able to buy a house and car on minimum wage. We're not voted the best country in the world 14 years in a row for nothing.
clonedeeznuts
As an American surrounded by morons I'm rooting for you but I've become pessimistic because for decades, advances in technology were suppose to free us from work but in turn it's enslaved is even more.
TheGhostOfAkistos
We’ve tried using AI to successfully create tool paths for cnc machining and the failure rate has been abysmal.
Remmon1
A few people at the company I work for are attempting to use LLMs to write documentation for old undocumented code.
The results so far have been hilarious. As in, hilariously wrong in every possible way.
It's almost like trying to retroactively document old complicated code in a complicated (financials) field is a disaster waiting to happen.
Smoretank
I am starting to learn how to use a cnc machine. Looked up how to tool path and used a couple with AI. Guess that's why my simple smilie face turned into a Jackson Pollock
spiceass9000
Huh, I would think it actually could do this
NKato
You sure you don't mean "success rate has been abysmal"? Also, is this AI the thing called CAMAssist? My company is considering picking it up to help with speeding up toolpathing, but as they marketed it as "getting you 80% of the way there", we're not going to look at it as a total solution.
MyLastAccountWasShadowbanned
... WTF? The first 80% are usually the easy and fun part of any task.
NKato
Right, but it can be extremely time-consuming. Wireframing the part, tool selection, toolpath selection, and working out the machining strategy can take a whole shift, if not more. CAM Assist claims to help reduce that to about 15 minutes or less. We're a small shop, so if we can reduce the time it takes to program a part so it gets spent more on proving and running parts on the mills, we see that as a benefit.
HighwayWizard
It can go either way; 'abysmal' just means 'very bad' (or very deep, but that doesn't work in this context) so failure rate being abysmal means very high and success rate being abysmal means very low- whichever makes the situation bad in context.
4etherling
What? If a failure rate is abysmal it means you're doing a deeply bad job at failing. In context it would mean that you have a high rate of success. It doesn't go either way. It was poorly worded.
paulbloface
No, they're completely correct. Faiilure is the bad thing. So if the rate is abysmal, there's a lot of the bad thing.
GenStrike
The grammar is really odd and makes the statement unclear.
ForlornHopeful
I'm assuming "Failure rate has been abysmal" refers to finished product coming out of the CnC
GenStrike
Your grammar here is really odd.
Are you trying to say: "We've tried using AI to create tools paths for cnc machining and the failure rate has been high" ?
GenStrike
Like, "successfully" is redundant. Why would you try to fail at something? "The failure rate has been abysmal" sounds like a double negative. It is unclear what you are saying.
Centroid
Abysmal or astonishing?
reichstein
Astonishingly abysmal.
Taalii
abomishing.
Centroid
I dig it
PipWhipple
Sorry, but I ordered a tungsten sphere, not a cube!
PowerPedant
It's a DIY kit. I comes with 25000 sheets of sand paper.
zwielichtengelchen
Hey look, I found a snail. Is this yours?
PipWhipple
[Homer Simpson scream]
Luvlyquants
If they do not even want to pay to store workers, people should not buy from such stores and see how well they profit. Freaking F* them.
QueenOfTheFey
If a store has no employees bc it's owners want to use robots to abuse the poor even more than they already do then it's 1000% ethical to steal from them.
Targe0
I mean, this is just a glorified vending machine. It's the same basic idea, a mini store that has no staff outside of the person who restocks it.
They made a vending machine with extra steps.
RickRollEditor
This is not an example of that. They knew it was going to fail. Claude is an LLM. It wasn't made for this, not trained for this and shouldn't even remotely work for this. That it actually did good in some decisions early on was surprising. It was a fun experiment to see what an LLM, which isn't even REMOTELY made for this, would do if put in the position.
Targe0
Basically a kind of stress test.
Put the tech into situations it wasn't built to handle and see how it handles them.
Then use those failure points as insights to guide its future development directions.
Sh1tMovieGroup
Annoyingly though there's plenty of scabs that don't care. Near us they've opened a "pour your own" pub, blatant headcount reducing profiteering, yep, it's packed every weekend... and we're supposed to be an "alternative" suburb who ran McDonald's out when they tried to move in, how far we've fallen...
GenStrike
You know why they're called "scabs"? It's because we pick them off. *BLAM! BLAM!*