6 ms·
The 1000 Indians story was completely incorrect though. They has 1000 people in India tagging the videos after the fact for improvements to the machine learnin
by jedberg 3y ago
The 1000 Indians story was completely incorrect though. They has 1000 people in India tagging the videos after the fact for improvements to the machine learning algorithms. They weren't watching in real time.
- smnrchrds 3y agoI have a question: was the customer billed based on whatever the machine learning algorithm determined, or based on what humans decided after the fact? In other words, was the humans' job to just provide training data for future versions of the machine learning algorithm, or to manually fix the algorithm's errors and prevent over-billing?
- entangledqubit 3y agoIIRC, they could correct your bill after the fact. I used to joke with others that shoplifted items should be retroactively free - as long as you self report after the fact. (Extra useful training data...)
- antiterra 3y agoFrom my experience you just weren’t actually billed until they were done, it could be many hours after you left the store.
- paxys 3y agoThere are reports that 70% of the sales had to be manually reviewed by the team in India (https://www.businessinsider.com/amazons-just-walk-out-actually-1-000-people-in-india-2024-4 https://www.businessinsider.com/amazons-just-walk-out-actual...), so no, they weren't just training algorithms.
- jedberg 3y agoThat article in particular was the one that was debunked. It was based on a false rumor.
- paxys 3y agoDebunked where? The only response is from Amazon (in the same article), who are disputing the number but still admitting that the team did intervene when needed.
- Dylan16807 3y agoAnd they did not give their own number, so even the dispute is very weak.
- jedberg 3y agohttps://www.theverge.com/2024/4/17/24133029/amazon-just-walk-out-cashierless-ai-india https://www.theverge.com/2024/4/17/24133029/amazon-just-walk...
- jnaina 3y agoThe 1000 Indians working to track all activities remotely was/is a false story. See link below: https://www.linkedin.com/feed/update/urn:li:activity:7186447512973574144/ https://www.linkedin.com/feed/update/urn:li:activity:7186447...
- Culonavirus 3y ago[flagged]
- sillysaurusx 3y agoNo need to make this personal. This is about Amazon, not jedberg.
- jedberg 3y agoI'm certainly not defending Amazon, I'm defending AI. What they build was impressive and worked for the most part (I saw the behind the scenes). It was just really expensive because you needed high resolution full coverage of the entire store. I just find it disingenuous for people to pull out this example as "why AI doesn't actually work" when it's not an example of it.
- waltbosz 3y agoHaving not read the original story, this makes more sense. "1000 Indians that were watching people shop at the Amazon Go stores" ... Conjured up picture in my head of an Indian man inside an Amazon Go store sitting behind a registerless counter. What his purpose would be, I don't know.
- WWLink 3y agoFunny story. The amazon go stores do have a register at the counter now. But as far as I can tell, they rotate positions. There's never really anyone standing at that counter. If you walk up to the counter someone will come from the sandwich counter or restocking the shelves or whatever.
- aimazon 3y agoFor what it’s worth, I purchased from one of these Amazon stores every day for over a year. After leaving the store, it would typically take 2+ hours before my purchases were registered in the Amazon app. I agree that there probably was a large degree of automation but based on how the store worked, there didn’t need to be an army of “real time” reviewers for humans to be involved in each individual purchase… it’s totally plausible based on what I experienced that Amazon did have an army of people involved in the standard purchase path.
- mFixman 3y agoSame here. My theory is that purchase tracking is done automatically via AI, while humans are on the loop to check the feed of people taking each item and verifying that the amounts are correct before charging. It makes sense. Even if your entire AI system is 99.9% accurate, you want a human to detect and remove that one-in-a-thousand mistake.
- jkhanlar 3y agoI think it was from a video I watched a day or two ago, describing how with those uber and lyft and other order delivery services that consumers apparently are placing those orders wtih the services, picking up the items themselves and walking out, and then when the actual delivery person arrives, order is gone, reported as missing, the consumer also reports similarly, even though they picked up the orders themselves, that this is apparently happening more widespread commonly that still is vulnerable with business infrastructures to not prevent the exploit from taking place, I am curious how this may affect industry standards with regards to this concept because I never witnessed it but I wouldn't be surprised if it is not even that widely realized about the issue yet thereby enabling more theft and loss and damages and whatnot.