Starbucks Abandons Borked AI Inventory Tool That Couldn't Count(gizmodo.com)
gizmodo.com
Starbucks Abandons Borked AI Inventory Tool That Couldn't Count
https://gizmodo.com/starbucks-abandons-borked-ai-inventory-tool-that-couldnt-count-report-2000762252
11 comments
This seems like a solved problem for those who care to solve it. Weigh product on intake, weigh on outflow or some other constant displacement pump solution. I've seen How It's Made and those factories have it down to a science without a guessbot at the wheel.
I am also missing the cost savings. I assume coffee shops have extensive dead-time in between peak hours. If some person does not have to manually perform inventory during the day - what other value was on the table to be generated?
Wonder if this is why my Starbucks has been chronically out of stock on all their food items and popular drinks for several months.
Maybe they can continue to pay people instead of robots.
But AI is better for them because people might file HR complaints against you... or anyway I guess this is something of a concern for Marc "sniffles" Andreessen... wonder why.
https://youtu.be/PHQvb10vKyk?si=pJTUqDI-n4naug1O&t=8200
(@2:16:40)
https://youtu.be/PHQvb10vKyk?si=pJTUqDI-n4naug1O&t=8200
(@2:16:40)
LLMs are notorious for not being able to count.
Try using them for cross words in a small manguage. It is hillarilus. Now, it has gotten better, but even top end models frequently answer confidently with things like "'eats grass and farts, 3 letters, _o_'
, must be ozelot!"
Based on available info, NomadGo doesn't rely on an LLM. The issues reported had to do with incorrect object recognition.
so first claude f'd up pizza hut. Now it has f'd up starbucks. So much winning going on in this AI space lately.