That is what it says. That they were required in 700/1000 transactions to verify and use the data to train a model. While the goal by this time was 50/1000. So, not going well. But also just seems like non-tech folks misunderstanding things in tech. Most of the articles refer to it as generative AI too which is a bit of a laugh
I’m in tech and while I’m aware of training and the likelihood of using no small amount of humans to do that, I don’t ever remember hearing about that aspect from Amazon? It was all flashy demos and ‘magical’ tech. So I sort of get non-techy people misunderstanding the reality here.
Given my limited experience with these stores (I went opening week and as recently as a month a go) this makes a lot of sense though. I don’t recall any of my receipts arriving earlier than like two hours after I walked out.
Lol 70% of it was the training boundary? And they went ahead and did a whole chain of stores?! Did they at least see promise in a smaller dataset? Or were they just like "big data make us money gods!"
I work for the Zon.
Not on the Go stores project but another team dealing with ML that require human review sometimes.
This is super common because Amazon tries to be careful when confidence is anything below like 95% on our team atleast. We have a team of people who do manual reviews for accounts flagged by our systems.
This is a good thing.
Now them being contractors and not paid nearly enough is a another topic altogether.
> This is a good thing.
How exactly is this a good thing? For this system, 70% of transactions needed human review (against the performance target of 5%). The entire system relies on human labour, simply for the convenience of not doing an actual check out.
The 70% statistic is available in other reports, [here](https://gizmodo.com/amazon-reportedly-ditches-just-walk-out-grocery-stores-1851381116) and [here](https://www.hindustantimes.com/world-news/us-news/1000-indians-part-of-amazons-generative-ai-driven-just-walk-out-project-in-fresh-stores-report-101712118547238.html) for example.
Sounds like average shitty journalism with no idea about what they're on about.
The store's were powered by machine learning algorithms with the goal being a minimum amount of human input and checking required. It didn't go as well as they had hoped and they ended up needing humans to check results a lot more often than they had hoped.
Yeah this is a catchy headline but they did have AI just they also had humans to train the model by auditing the results. It may have even been staff reviewing all video footage for quite a while but that doesn't mean they didn't eventually have a model. Just sounds like a shitty application of AI that was abandoned rather than Amazon lying about the tech.
Depends on how much the human is helping really, if it's just "AI is 60% sure of this thing y/n" I'd take it as still AI driven work even if a human had to double check.
It's more like editorializing. To the point I wouldn't call it "news". Just a rando's opinion article. Couldn't help but throw in hot takes:
> Apart from relying on cheap, outsourced labor instead of paying fair wages locally
A city near me has had an Amazon Fresh "ready to open" - like signs up, cart corrals out, the works.
For two years. Local media confirmed it's not hung up in any sort of inspections, Amazon just... hit pause.
I thought they were just ditching the “walk out checkouts,” not the actual stores.
https://gizmodo.com/amazon-reportedly-ditches-just-walk-out-grocery-stores-1851381116
I shop almost exclusively at my local Amazon Fresh, because A) the prices are really good, especially when they toss in 20% discounts on the entire purchase, and B) I love the "dash cart" that allows me never to talk to another human being.
However the entire experience makes me doubt the store is profitable. I've no data to support my suspicion, but I get the feeling it's there as a laboratory to track -- and, ideally, predict -- customer behavior, and refine things like just-in-time inventory management. The cart itself is clearly a kluge that, granted, fills a niche, and v2 improved on some of the problems with v1, but was unlikely to be engineered to be anything but a stopgap measure until they can refine the technology mentioned in OP's article.
But this is why I think so many stores are lying fallow, because they actually would cost *more* to open for business. Amazon's strategy may just be to acquire the real estate against the day when the more advanced stores can open and then can really start to dominate. Right now it's still R&D.
Pretty much. A lot of automation is really just some people in India. I worked for a company that did comment moderation of news sites. We sold it as AI. It was really a team of people in India. The AI part was having that remote team flag stuff so we could train an ML. But the actual moderation, 100% human.
It's standard practice for tons of shit, especially AI related stuff. Those annoying automated phone systems that take voice prompts? There are call centers somewhere with a bunch of people that never take any calls directly from customers. They just get audio snippets (with zero context for user privacy) of what users said when the computer isn't sure itself and then tell the computer what they meant.
One such system that's fairly mature i babysit shows about ~80% success from the IVR itself most of the time. These are nationwide phone lines and staffed by well under 1000 people so maybe Amazon just really sucks at it.
Human labeling is the basis for AI training. I hope at least some people are learning something from this bullshit headline rather beyond just “AI is a scam!”
Doesn't sound believable to me, considering that these sort of stores exist in other countries too, not run by Amazon. Is it more of manual reconciliation and training?
49 Comments
rlbond86@reddit
notimmortalyet@reddit
justin-8@reddit
rayray5884@reddit
theophys@reddit
stingraycharles@reddit
untetheredocelot@reddit
pseudosciencepeddler@reddit
untetheredocelot@reddit
pseudosciencepeddler@reddit
programming-ModTeam@reddit
os12@reddit
FullyStacked92@reddit
FlukyS@reddit
WriteCodeBroh@reddit
FlukyS@reddit
Swoo413@reddit
link_29@reddit
Synth_Sapiens@reddit
deja-roo@reddit
Spyder638@reddit
gefahr@reddit
thegininyou@reddit
alicia-indigo@reddit
balthisar@reddit
neodiogenes@reddit
Membership-Exact@reddit
zazzersmel@reddit
davidellis23@reddit
eigenman@reddit
davidellis23@reddit
zazzersmel@reddit
hanumaNRL@reddit
OptimalOptimizer@reddit
hanumaNRL@reddit
Nidungr@reddit
Specialist_Brain841@reddit
rydan@reddit
ripter@reddit
Specialist_Brain841@reddit
LeatherDude@reddit
granadesnhorseshoes@reddit
eigenman@reddit
obsertaries@reddit
o5mfiHTNsH748KVq@reddit
IgnisIncendio@reddit
notimmortalyet@reddit
_wombo4combo@reddit
icemelter4K@reddit