"AI" at the kebab shop: "Hand made: We recommend this unfoigcttablc taste breac you eat the naturally lea proper size with knife It will bring more chewi..."
The recipe for “fresh baked bread” that the wrapper claims to be baked “every day - Homemade Style”. Apparently, an incredibly clever AI model has come up with something truly special and delicious here - but I still really hope the kebab shop doesn’t start baking its flatbread that way:
I clove garlic, crushed
I quantity tunc btini mixture, Kcupf (75g) whole -egg miyoBiuiK
1 tablespoon (om Jto paste
1 tablespoon Worcestershire uuce
2 tablespoons fresh breadcrumbs
sea salt and cracked black pepper
8 rashers bacon, rind removed
12 discs white bread, touted
bunec for spreading
2 vine-ripened tomato’s, sliced
50g baby lettuce leaves.
4 tablespoons store-bought fruit chutney
notabot@piefed.social 15 hours ago
Rather than “AI”, that looks quite a bit like something that’s been badly scanned, run through a basic OCR to create text, which has then been laid out by someone who doesn’t write the language well enough to kniw it’s gibberish.
The line “1 tablespoon (om Jto paste” for instance, was almost certainly “1 tablespoon tomato paste”. You can see the similarity oin the character shapes between the gibberish and real text.
DandomRude@piefed.social 14 hours ago
That could certainly be the case, but it seems like a lot of effort to me to produce something like that.
Either way: Apparently, the people at the kebab shop didn’t notice what was written on it when they placed the order. They may have saved a little money on quality, but it doesn’t strike me as particularly flattering, especially since they actually pride themselves on homemade food, which tastes great.
Luminous5481@anarchist.nexus 11 hours ago
Do the people at the kebab shop read English? Just because they can speak it doesn’t mean they’re literate.
ParadoxSeahorse@lemmy.world 11 hours ago
The other poster has the image that was likely plagiarised by the unscrupulous vendors manning the slop machine.
General purpose or image-focussed models have weak OCR so they tend to hallucinate / extrapolate, resulting in this gobbledygook.
NutinButNet@hilariouschaos.com 10 hours ago
Agreed, this is classic for that. Even a perfectly scanned document into OCR almost always has errors. I have set it up for a lot of people and I always stress that you need to go in and review what got scanned because there has never been a time I have seen it get everything spelled correctly or in the right format as it was.
Same as AI. Anything you get from an LLM should be reviewed. Too many people just blindly trust whatever gets spit out without ever reviewing.