In 2015, Stanford computer science PhD candidate Andrej Karpathy decided to test out some neural network tools he'd been experimenting with, and set them to generating plausible baby names.
Karpathy trained the net with a data set of 8,000 baby names and turned it loose. It produced a lovely collection of names, 90% of which were not in the training set.
Mellina
Edongia
Lorant
Baires
Dasher
Quengy
Kens
Paac
Jeencia
Jaiep
Deua
Analine
Mifalia
Conette
Dovanice
Tyberr
Irda
Gerfand
Auberro
Thamosa
Babrield
Kait
Stephepoa
Abi
Chrristen
Stephanna
Racky
Dones
Lys
Vira
Phela
Deney
Dorah
Avet
Ca
iunia
Jasqquosius
Esty
Solumote
Hilling
Gredie
Dynae
Berdiah
Suynn
Tile
Blian
Mancie
Tovey
JotaLine
Meiger
Galrucd
Ferdeor
Soligo
Joan-Rodelina
Ettellon
Viterine
Sobi
Aderina
Juloina
Karali
Oethreh
Morred
Rangell
Crico
Cervynt
Syere
Ducky
Akie
Emilya
Epheny
Malda
Wetenn
Jaqueteg
The Unreasonable Effectiveness of Recurrent Neural Networks
[Andrej Karpathy/Github]
#RandomExperimentSundays [Andrej Karpathy/G+]
(via Diane Duane)
