IF THERE’S one thing that we’ve had to deal with in these GDPR times, it’s identifying cookies.
But it doesn’t have to be an unpleasant experience, as demonstrated by the latest AI experiment from INQ hero Janelle Shane.
Just as she previously created brand new paint colours, band names and other weird stuff for her blog, she’s now loaded textgenmn, a neural network, with a list of American cookie recipes (432 of them) from a list on the interweb.
Then, using a bit of jiggery-pokery to make the results remotely relatable, a whole bunch of new cookies were born. Tuck in!
Some of our favourites include ‘Spice Biggers’, ‘Merry Hunga Poppers’, ‘Hand Buttersacks’, and the stunning ‘Nutty Loispers *then Wholes’.
Bored of Gingerbread men? How about “Sugar Person Sugar Masts”?
Something a bit more savoury? Maybe some ‘Brown Cheesy Cookies’ will fit the bill.
As ever with neural networks, these flavour combos show what sounds grammatically plausible (to a machine at least) descriptions which have no context, and as such sound ridiculous to us.
A good example of this is the network’s rather haphazard interpretation of the word “balls” which has left us with the likes of ‘Kookia Fried Mind Balls’ and even ‘has Balls’.
There’s some other cookie, but less ball related examples too. We’re quite keen on ‘Cornbeat Fakes Limb Cookies’, ‘Crawhot’s Maple Bischpoids’ and ‘Graff Scotchin Baspy’s Raisin Bars’. The rest can be seen by signing up to Janelle’s mailing list – which we thoroughly recommend you do.
We’re still very much in a stage where this kind of thing is as ridiculous and entertaining as it can be – the better neural networks get, the less fun we’ll have. But it’s not all fun and games, and studying these results is an important way of moving artificial intelligence forward.
Anyway, we’re off to fill our faces on Merry Hunga Poppers. μ
Source : Inquirer