The Name Machine

# read 6,804 real indian names. now it makes up its own.
# it only remembers ONE letter at a time, so expect some nonsense.

# mode:

$ make-names

start with wildness normal

$ is-it-a-name

$ what-comes-after

$ cat how-it-works.txt

1. it reads a list of real first names (indian or english, switch at the top).
2. for every letter, it notes which letter came right after it, and how often.
   in indian names, "s" is followed by "h" almost half the time. most names end on "a" or "i".
3. to make a name: pick a first letter the way real names do, then keep
   picking the next letter based ONLY on the one before, until it stops.
4. that's it. it has no idea what came two letters back, hence "samitabhjabhr".
   big AI models use the same idea, but look much further back.

mode count    = it literally tallied the pairs.
mode practice = it started with random guesses and corrected itself 1,000 times,
                which is how bigger AI models learn. both end up almost identical.

// source: github.com/nrjai/bigram-language-model · indian names: mbejda lists, cleaned · english names: karpathy/makemore

>name_machine