What really annoys me is the slow hand scanners. It takes about 5 seconds to download the item data for each scan (name, picture, price). And there's no cache so scanning several identical items is painful.
It would be so easy to cache the last few items for repeat scans (after all, the list is right there on the device, so it even has the data in memory). And I guarantee Tesco has a profile of my usually-bought goods tied to the Clubcard that I scanned to get the scanner. After all, that's why they want to railroad everyone into having Clubcards. So why not have the system pushing likely next scans based on what I just scanned, combined with where it is in the shop and what I usually buy.
Even to a hype-weary AI sceptic, that's a textbook example of machine learning: overwhelming amounts of perfectly labelled data, a great feedback loop (cache hits and misses), and negligible penalty for mistakes (cache miss).
We are not even talking Amazon levels of inventory. A store has what, 10k, 20k items for sale? The entire store inventory prices could fit into <1MB of memory (without pictures).
Comments
What really annoys me is the slow hand scanners. It takes about 5 seconds to download the item data for each scan (name, picture, price). And there's no cache so scanning several identical items is painful.
It would be so easy to cache the last few items for repeat scans (after all, the list is right there on the device, so it even has the data in memory). And I guarantee Tesco has a profile of my usually-bought goods tied to the Clubcard that I scanned to get the scanner. After all, that's why they want to railroad everyone into having Clubcards. So why not have the system pushing likely next scans based on what I just scanned, combined with where it is in the shop and what I usually buy.
Even to a hype-weary AI sceptic, that's a textbook example of machine learning: overwhelming amounts of perfectly labelled data, a great feedback loop (cache hits and misses), and negligible penalty for mistakes (cache miss).
We are not even talking Amazon levels of inventory. A store has what, 10k, 20k items for sale? The entire store inventory prices could fit into <1MB of memory (without pictures).
Yeah, but you can shave a couple of pounds off the device cost by using a shitbox CPU with almost no RAM or flash!
It also means that the embedded developers can't start writing web frameworks into the devices, so they might actually ship this decade.