►
From YouTube: 2013 Hackathon Kickoff
Description
2013 Hackathon Kickoff
B
A
So
everybody
who
is
a
hackathon
staff,
could
you
mind
standing
up
real,
quick
I
want
to
identify
you.
These
are
the
people
that
can
help
you
if
you
get
nifty
staff
shirts
at
end
of
the
hackathon.
If
you
have
a
t-shirt
too,
so
all
of
the
groc
/
you
meant
to
personnel
are
here
hold
on.
Let
me
see
if
I
can
get
this
screen
down
a
little
screen
deck
nope
as
far
as
it
goes.
Okay,
so
you
know
Jeff,
Zubat,
I,
myself
and
then
other
engineers
from
New
menta.
A
We
also
have
some
tagged
personnel
here,
Courtney,
molds
and
I.
Don't
know
if
Johannes
here
or
not,
but
he's
the
CTO
of
tagged
and
he
RSVP'd
as
well
I
want
to
say
a
quick
thank
you
to
Terry
fry
and
Casey
stone
who
helped
out
a
lot
with
all
the
refreshments,
the
logistics.
That
was
a
big
help.
I
don't
think
it
would
not
have
been
as
nice
of
an
event
without
and
thank
you
to
tagged
who
generously
landed
us
this
facility
for
use.
Please
respect
it!
A
It's
a
really
nice
area,
tons
of
room,
good
Wi-Fi,
lots
of
electric
outlets,
so
thank
you
tagged
and
then
we
have
several
members
of
our
community
who
RSVP'd
as
well.
So
these
are
people
that
are
on
our
mailing
list.
If
anyone
is
on
our
mailing
list,
you
mind
standing
up
just
identifying
yourself
saying:
hi
I
recognizes
you
for
coming
very
helpful
awesome.
If
I
missed
anybody,
I'm
sorry
I
didn't.
I
didn't
catch
everyone,
so
I'm
going
to
pass
off
for
a
quick
little
blurb
from
and.
A
C
C
There
you
go
is
now
you're
getting
sound.
Sorry,
all
the
important
things
were
said
for
those
of
you
tuning
in
so
mad,
as
if
I
would
just
say
a
few
words
about
of
introduction,
so
I'm
going
to
do
that
right
now.
So
first,
thank
you
for
coming
here
and
thank
you
for
spending
your
time
and
your
your
intellect
on
this.
This
project,
we
think,
is
very,
very
important.
So
just
few
things
why?
Why
are
we
doing
this?
What
is
new
pic
about?
Why
did
we
do
this
when
I
started
new
menta
about
eight
years
ago?
C
This
is
I've
been
working
in
this
field
for
over
30
years
actually,
but
when
we
started
the
men
to
30
years
ago,
the
mission
of
that
company,
which
is
now
Brock,
it
was
to
be
a
catalyst
for
machine
intelligence,
and
it's
still
the
mission-
and
you
know
in
a
catalyst,
is
something
that
accelerates
a
reaction
that's
happening
anyway,
but
and
it
doesn't
get
consumed
in
the
reaction.
That's
the
definition
of
a
catalyst,
so
you
know
we're
going
to
build
intelligent
machines,
we're
going
to
figure
out
how
brains
work,
and
you
know
what
I
said.
C
What
we
can
do
and
I
could
do
is
we
can
accelerate
that
we
can
make
it
happen
quicker
and,
and
I've
been
spending
a
lot
of
my
time.
Trying
to
do
that,
I
think
it's
the
approach,
we're
taking.
We
just
understand
how
the
neocortex
works
and
base
it
on
biological
principles
is
one
that's
gaining
ground.
It
was
very,
very
unpopular
number
of
years
ago,
but
it's
now
getting
a
lot
of
adherence
more
recently
site,
I'm
totally
confident
than
what
we're
doing
use
the
right
path.
C
So
you
know
this
is
all
about
modeling,
a
neocortex,
but
really
as
a
path
to
machine
intelligence,
and
you
don't
need
to
know
a
lot
of
neuroscience
to
do
this,
but
there's
a
lot
of
concepts
you
have
to
learn
and
those
concepts
are
online
learning
hierarchy,
sequence,
memory,
sparse
distributed
representations
and
so
on
now
that
the
core
of
new
pack,
of
course,
is
this
cortical
learning
algorithm.
That's
the
code.
The
code
is
mostly
implementing,
which
is
a
model
of
a
layer
of
cells
in
the
New
York
cortex.
It's
a
real
building
block.
C
It's
like
the
first
building
block.
You
can
get
beyond
having
a
neuron
and,
you
might
say
yeah,
but
neural
networks
have
been
doing
this
for
a
long
time.
Not
really
neural
networks
aren't
based
on
anything.
That's
like
real
neurons
and
the
not
based
on
any
kind
of
architecture
you
see
in
the
brain.
They
don't
really
do
the
right
things,
mostly
spatial
classifiers,
so
this
is
I.
C
Think
we've
really
achieved
a
very
nice
milestone
in
in
a
building
block
of
how
thousands
or
hundreds
of
thousands
of
neurons
work
together
to
learn
spatial
temporal
patterns
and
to
make
inference
and
predictions
and
so
on.
I
think
this
is
really
important
work
for
a
couple
reasons
it's
important
biologically,
but
most
of
you
don't
care
about
that.
It's
important
because
this
is
a
we
are
working
on
the
path
of
machine
intelligence,
I
kind
of
liken
it
back
to
nineteen
forty-three
in
the
computer
world
I've
been
saying
this
recently.
C
My
talks
some
of
my
talks,
but
you
know
back
in
1943.
They
were
just
starting
to
build
computers,
no
just
trying
to
figure
how
they
work
and
how
to
get
the
technology
to
work
and
find
what's
useful
about
them,
and
so
on
and
I
feel
like
we're
in
that
stage.
Right
now
we're
at
the
beginning
of
the
very
long
period
of
time,
which
is
going
to
become
a
very,
very
big
industry
and
very,
very
influential
in
on
this
planet.
C
A
lot
of
people
are
looking
for
this
I
talk
at
conferences
and
people
are
looking
for
what's
beyond
by
Norman,
the
people
in
the
semiconductor
world
are
looking
for.
You
know
how
do
I
take
advantage
of
the
incredible
to
pass,
they
can
put
on
chips,
and
many
of
these
people
are
interested
in
what
we're
doing
the
CLA
and
our
work.
So
some
really
great
conversations
going
on
there.
We
have
a
partnership
with
IBM,
doing
research
on
building
chips
for
these,
for
our
algorithms
and
so
on.
C
We
put
this
in
an
open
source
environment
because
it
fits
with
our
mission,
we're
not
trying
to
keep
this
private
or
secret
people.
They
asked
us
to
do
this
when
we
first
published
the
CLA
white
paper.
People
said:
would
you
please
make
the
source
code
available?
We
want
to
build
a
community
your
that
this
is
our
second
hackathon
there's
a
lot
bigger
than
our
first
one
and
and
that's
great
I.
Imagine
this
is
going
to
continue
to
grow.
So
so
we
open
with
old,
isn't
community
about
this
around
the
world.
C
Do
whatever
it's
impossible
to
make
people
productive,
advance
the
state
of
the
art,
the
science,
whatever
things
we
need
from,
you,
there's
lots
of
different
people,
type
of
capabilities
in
the
community,
and
we
need
everything
from
simply
as
documentation
and
diagrams
and
and
tools
to
characterize
the
algorithms
to
improve
the
algorithms
apply
them.
The
new
types
of
data
Francisco
Webber's
here
from
sep
than
they've,
been
using
sparsha
stupid
representations
in
language
with
a
bunch
of
you
interested
in
that
here
today
we
can
build
new
products.
C
On
that,
what
we're
doing
I
grok
we're
building
a
product
based
on
it.
I
think
we're
going
to
download
that
product
tomorrow,
it's
very
cool,
it's
totally
built
on
built
on
the
CLA.
We
hope
it's
going
to
be
a
big
product.
It's
important
to
do
that
because
it's
important
to
get
commercial
interest
money
attention
from
people
and
things
like
that
and
then
finally,
there's
some
popper
to
do
some
do
real
research.
Some
of
you
are
interested
in
and
advancing
this
in
very
significant
ways.
C
Now,
I'm
going
to
talk
later
today
about
sensory
motor
integration,
a
few
of
you
we'll
talk
about
that
some
people
here
interested
in
extending
the
hierarchical
concepts
underlying
this.
So
these
are
not
easy
things
to
do,
which
leads
me
to
my
last
point.
This
stuff
is
hard.
It
is
not
easy.
It's
not
like
you
can
just
come
in
here
and
just
flip
a
switch
and
start
doing
something
really
amazing,
but
it
takes
a
little
while
to
get
under
the
skin
of
the
CLA
and
under
these
principles
to
understand
them.
There's
a
lot
of
things.
C
We
understand
a
lot
of
things.
We
don't
understand.
So
you
know
I'm,
not
gonna
sugarcoat,
this
it's
hard
work,
but
we
got
stuff
that
works
really
well
now
and
we're
going
to
go
with
it
and
continue
to
work
on
and
finally
had
to
say,
I
think
it's
worth
it!
It's
worth
me.
I
know
it's
worth
my
time.
I'm
hoping
you
find
it's
worth
your
time.
This
is
really
significant
stuff.
I
think
it's
going
to
you
know.
C
Historically,
people
will
look
back
will
say
you
know
where
the
where
the
beginning
of
machine
intelligence
occurred
and
will
play
a
role
in
that
and
I
think
it
it's
going
to
be
very
beneficial
for
the
future.
So
those
are
my
only
comments
unless
I'm
I
got
any
questions,
I'm
going
to
be
here
all
day
and
tomorrow,
I
won't
be
here
overnight,
but
so,
if
you
want
to
talk
privately,
you
can
do
that
too.
But
if
anyone
wants
to
ask
a
question,
a
group,
that's
fine!
No
all
right!
A
A
Yeah:
okay,
oh
here's,
your
laptop,
alright,
alright!
So
my
name
is
Matt
Taylor
I
am
the
self-proclaimed
community
flag,
bearer
I,
don't
like
to
call
myself
community
manager
because
I'm
not
trying
to
manage
you
guys
I'm,
just
trying
to
help
I'm
gonna
go
over
some
protocol.
Most
important
thing
is
all
of
our
all.
The
resources
for
this
hackathon
are
on
this
little
page
here.
So
I'm
a
it's
easy
to
remember
the
method
org
slash
hack
and
it
shows
wow.
It's
very
large
this.
A
This
is
pretty
much
what
I
would
consider
the
most
important
things
for
you.
During
this
event,
it's
got
a
link
to
our
code
of
conduct,
which
you
probably
saw
on
your
way
up
so
they're,
just
expectations
of
your
behavior.
It's
pretty
standard,
don't
expect
any
problems.
There's
a
link
for
hack
registration,
which
is
required.
I
will
be
turning.
This
on.
I
should
probably
turn
it
on
right
now,
so
I'll
turn
it
on
as
soon
as
I
have
a
free
moment
and.
A
There's
also
the
schedule,
which
is
also
on
the
on
a
monitor
to
the
side
over
here,
which
we're
going
to
try
and
make
figures
in.
We
have
some
announcements
that
will
try
and
keep
up
to
date
with
anything
that
we
think
we
might
need
to
communicate
to
you
in
mass
we're
also
going
to
have
some
online
participants.
So
it's
one
of
the
reasons
I
wanted
to
make
this
page
available
to
anybody
and
there's
a
link
to
our
YouTube
channel,
which
should
have
this
live
event
streaming
right
now
so
again
for
online.
A
Okay,
and
for
communication,
really
encourage
you
to
get
onto
IRC,
there's
going
to
be
people
within
the
community
that
can't
make
it
here
that
are
very
helpful
on
our
mailing
list
and
on
IRC
so
reach
out
to
them.
If
you
need
help,
if
you
can't
find
help
here
of
leak
and
they're
all
aware
of
this
hackathon
and
probably
going
to
be
monitoring
us
so
there's
links
to
our
IRC
channel
and
our
mailing
list,
and
if
you
find
an
issue
that
you
think
is
a
bug
feel
free
to
file,
it
will
take
a
look
at
it.
A
A
So
this
is
our
schedule
today
we're
running
a
little
bit
late,
but
right
after
this
little
kick
off
presentation.
I'll
talk
about
natural
language
processing
will
have
time
for
lunch.
In
a
bit
of
a
Bingle,
you
can
start
on
your
hacks.
All
of
these
are
pretty
much
optional.
If
you
just
want
to
find
a
room
and
get
to
work,
that's
fine
and
we're
live
streaming
and
putting
everything
on
YouTube
is
well
a
reference.
A
So
if
you
want
to
come
back
later
and
take
a
look,
it's
your
prerogative
we're
going
to
have
CL
a
deep
dive
by
our
vp
of
engineering,
super
tile,
med
and
Jeff's
gonna
or
excuse
me.
Ian
Danforth
is
going
to
kind
of
MC
discuss
about
mapping
things
within
the
CLA
into
more
machine
learning,
artificial
intelligence.
Semantics
later
this
effort,
lunch
then
Jeff's
going
to
give
a
sensory
motor
integration.
Talk.
A
I've
got
five
thirty
kind
of
left
open
in
case
anything
pops
up,
I
think
I'm,
going
to
let
what
some
of
our
community
have
some
discussions.
At
that
point,
and
at
eight
o'clock
this
is
Q
a
technical
Q&A.
If
you
guys
are
having
problems
or
you
don't
understand
how
something
should
work,
what
kind
of
just
it's
an
opportunity
for
us
to
get
together
ask
questions,
help
each
other
out
more
formally
than
just
walking
around
asking
people
randomly,
so
you
can
feel
free
to
stay.
The
night
grab
couch,
maybe
brought
a
pillow
sleep
on
floor.
A
Whatever
you
wish,
we
will
have
staff
on
hand
throughout
the
night
tomorrow.
Francisco
is
going
to
talk
some
about
sdrs
with
Incept
word
sdrs.
So
these
gracious
to
come
Fergie
in
Austria,
Thank,
You,
Francisco
I
know.
Some
people
in
our
community
has
some
questions
about
how
those
are
constructed
and
one
of
our
a
product
manager
for
Brock
is
going
to
give
a
little
demo
of
rocks
product
rock.
A
A
A
This
whole
area
here,
there's
actually
people
working
in
here,
so
try
not
to
disturb
them
any
more
than
I'm
disturbing
them
right
now,
and
also
in
the
back
corner
to
some
private
areas.
They're
clearly
marked
just
please
stay
out
of
it.
Stuff
that
we're
not
supposed
to
get
into
the
bathrooms
are
back
this
direction
over
your
left,
shoulder.
There's
men
and
women's
bathroom,
there's
also
a
shower.
A
If
you
need
it,
I
haven't
attempted
to
use
it
yet,
but
hopefully
the
door
locks
and
the
emergency
exits
are
here
to
my
to
my
right
directly
behind
Jeff
and
back
in
this
corner
behind
your
right
shoulder.
If
you
need
to
get
out
quickly,
all
of
these
green
rooms
that
you
see
here,
our
conference
rooms
lining
this
whole
hallway,
it's
a
big
one
in
the
back.
Those
are
free
to
use.
A
So
if
you
get
a
team
together
and
you
want
some
private
space
and
quiet
time
grab
a
room
close
the
door,
they
all
have
whiteboards
a
couple
of
them
have
projectors,
so
its
first
come
first
serve
trying
out
the
hog,
a
big
room.
It's
just
you,
you
know,
share
share
the
wealth,
but
those
are
all
for
us.
These
we've
got
to
set
up
some
work
stations
around
here.
You
can
work
wherever
you
find
space
to
work.
If
you
need
the
power
or
something
like
that,
come
and
see
us,
it
could
probably
help
you
out.
A
The
Wi-Fi
is
posted
on
the
back
of
your
badge.
It's
tagged,
meet
up
and
I
won't
say
the
password
since
we're
streaming
it,
but
it's
on
the
bathroom
badge
be
there
are
some
rec
rooms
as
a
ping-pong
area.
There's
a
video
game
room
back
here.
There's
a
tinker's
pool
foosball
back
by
the
other
emergency
exit
feel
free
to
use
those
just
respect
to
space,
don't
wreck
anything
in
the
rec
room.
A
You
can
come
and
go
whenever
you
wish,
but
we
ask
you
to
please
let
whoever's
at
the
front
desk
know
we're
trying
to
keep
track
of
who's
in
the
facility
just
in
case
there's
an
emergency
or
something
happens.
This
is
not
our
area,
so
we
just
want
to
track
as
much
as
we
can,
but
someone
will
have
to
buzz
you
back
in.
So
let
us
know
if
you
come
and
go
but
you're
free
to
come
and
go
as
you
wish
and
talked
about
the
off-limits
area
to
sleeping
the
Wi-Fi.
A
Okay,
hopefully
everybody
has
a
food
will
be
delivered
for
lunch.
Today,
food
will
show
up
there's
a
drinks
in
this
area.
This
will
be
pretty
much.
The
food
will
all
be
back
here,
so
there's
breakfast
set
up
now.
They'll
be
lunch
w
dinner
later
tonight,
tomorrow,
there'll
be
another
breakfast
and
lunch,
but
no
there
and
please
a
once
again
respect
this
space.
A
It's
not
ours,
so
we're
trying
to
take
good
care
of
it,
and
we
were
hopeful
that
everybody
will
help
us
take
a
care
of
it
so
see,
trash
and
stuff
laying
around
he'll
pick
it
up
turn
away
we're
going
to
be
the
same.
I've
already
mentioned
communication.
This
seems
primitive,
but
this
is
the
best
way.
A
So
I'd
like
to
take
a
quick
opportunity
to
ask
you
guys
if
any
of
you
already
have
ideas
about
what
you
want
to
work
on,
so
this
quick
show
hands.
Anybody
have
an
idea
of
something
they
want
to
do.
There's
one
two:
three:
four:
five:
six:
okay,
how
many
are
try
our
want
to
do
something
with
natural
language
processing,
one
two
three
so
raising
hi
I
want
it
I
want
to
connect
to
people
here.
A
So
if,
if
people
are
interested
in
doing
something
similar,
it
might
be
a
good
idea
to
find
other
people
with
a
similar
interest
in
and
team
up
you're
going
to
get
more
done.
If
you
team
up
and
pull
your
knowledge
experience
and
resources,
that's
it's
a
good
idea
to
team
up,
so
some
people
already
know
what
they
want
to
do.
If
you
don't
know
what
you
want
to
do,
go
talk
to
them
as
who
their
idea
is.
It
might
sound
cool
and.
A
A
A
A
A
All
right
any
other
questions:
okay,
all
right,
I'm,
going
to
talk
about
natural
language,
processing,
I'm,
going
to
do
a
couple
of
demos
and
show
you
how
you
can
get
to
this
code,
so
I
started
playing
with
this
is
because
I
wanted
to
investigate
for
this
hackathon
how
hard
this
would
be
to
do
anything
at
all
with
new
picket
natural
language
processing.
So,
let's
first
talk
about
the
python
natural
language
toolkit
who
has
ever
used
the
python
natural
language
toolkit.
So
there's
a
few
of
us
that
have
so
it's.
B
A
Very
powerful
library,
I
haven't
used
much
of
it
at
all,
but
I've
used
enough
to
just
get
some
of
the
things
I
want
to
get
done
done.
It's
not
necessarily
fast
proficient,
but
it's
very
powerful.
So
what
I
did
with
it
was
part
of
speech
tagging,
so
here's
an
example
of
that
so
I
had
a
corpus
of
text
which
was
a
children's
book
or
several
children's
books
that
I
got
from
our
open
source
community,
which
was
compiled
by
our
community.
So
for
an
example
that
is
the
way
thor
got
was
wonderful
hammer.
A
A
Well,
those
are
categories
and
new
pic
an
encode
categories
and
I
wanted
to
see
how
I
could
take
these
texts
and
not
feed
the
words
into
new,
but
just
feed
the
parts
of
speech
now
in
significant
and
see
if
it
could
recognize
any
significant
patterns.
Just
within
these
texts
of
what
part
of
speech
might
come
after
another
part
of
speech,
so
I'll
show
you
a
quick
demo
of
this
and
I'm
going
to
start
off
from
where
I
told
you
to
go
I'm
answer
that
org
hack.
A
A
Word
association,
pretty
speech
I'll
go
over
that
some
of
these
are
failed
experiments
that
just
don't
work
but
they're.
Still
there
parts
of
speech.
Okay,
so
I
made
this
script
that
reads:
an
input
text
I'm
going
to
grab
the
example
here.
Okay,
so
I
made
a
script
called
run,
POS
experiment.
It
takes
an
input
text.
I
forgot
the
hammer.
A
A
So
this
is
taking
every
word.
Every
sentence
actually
just
reading
the
full
tax
breaking
up
into
sentences
and
for
each
sentence.
Let's
stop
it
there
for
each
sentence.
It
sends
it
into
an
otk
and
said,
give
me
all
the
parts
of
speech
and
then
I
just
take
all
those
parts
of
speech,
one
after
the
other
and
feed
it
into
the
OPF.
The
online
prediction
framework
and.
A
So
let's
take
a
look
at
this,
so
this
output-
this
is
output
from
the
script
e,
and
so
the
sentence
was.
She
did
not
want
forward
to
see
her.
What
NLT
k
said.
These
words
were:
was
pronoun
past
tense,
verb,
adverb
for
proper
noun,
the
word
to
which
is
special
verb
and
thrown
out,
which
it
looks
pretty
much
correct
right,
and
this
is
what
new
thick
predicted
that
part
of
speech
would
be
and,
as
you
can
see,
new
pics
pretty
good
at
predicting
end
of
sentences.
A
It
usually
predicts
more
a
lot
more
ends
of
sentences
than
actually
are
in
the
text,
because
when
you
just
look
at
the
parts
of
speech
but
there's
a
lot
of
subtleties
about
this
whole
project.
But
if
you
just
look
at
the
parts
of
speech,
you
can
end
a
sentence.
A
lot
of
different
places
within
a
sentence
right
so
new
pic
makes
those
predictions
because
it
sees
lawful
small
sentences,
and
then
it
sees
word
phrases
that
could
end
but
then
also
get
extended
into
larger
sentences
with
multiple
word
phrases.
A
So
you
have
to
think
about
language
as
branching
branching
structure
as
a
tree.
So
at
any
point
in
a
sentence,
I
could
go
this
direction
or
that
direction,
but
there
are
still
patterns
within
these
strings
of
parts
of
speech
within
these
word
phrases
that
new
pic
picks
up.
So,
for
instance,
it
did
predict
a
pronoun
is
coming
because
that's
a
very
common
beginning
to
a
sentence
as
a
pronoun.
A
He
she
it
they
as
the
subject
of
sentence
and
it
actually
predicted
a
past
tense
and
past
tense
verb
would
be
after
the
pronoun
once
it's
all
the
pronoun.
So
it's
not
perfect,
but
if
you
think
about
this,
it
will
never
be
perfect.
Even
a
human
being
could
not
predict
the
next
part
of
speech
that
I'm
going
to
say
right
because
I'm,
the
only
one
I'm
the
one
making
that
decision,
but
there's
a
lot
of
interesting
things.
A
Makes
sense
or
not
sending
sentences
matically
incorrect
at
the
new
big?
It
should
increase
the
anomaly
scores
because
it's
seeing
patterns
that
it's
never
seen
before
those
those
word
phrases.
So
that
could
be
a
challenge
that
can
be
a
hack,
and
that
could
be
something
that
you
know
you
could
use
with
auto
correct
software.
A
So
there's
a
lot
of
interesting
subtleties
here
this
this
script
also
outputs
into
a
file.
If
you
want
to
process
it
later
with
the
tags-
and
this
is
all
open-
you
guys
grab
this,
do
it
worth
it
so
I,
don't
I.
Could
I
could
talk
about
this
for
too
long,
so
I'm
just
going
to
cut
it
off.
You
want
to
talk
more
about
this
on,
be
here
all
day,
all
night.
A
Okay,
let's
talk
about
cept
word
STRs,
so
moving
away
from
NLT
k,
that's
one
way
you
could
go
with
NLP
with
you
pick
so
Francisco's
in
the
back
we'll
be
here
to
answer
questions,
but
what
set
does
is
they
have
an
API?
That's
pretty
cool.
Take
the
word
metal,
for
example.
The
term
set
has
an
API
that,
if
you
give
it
a
term,
it
will
return
to
you
in
an
SDR.
A
A
So
if
you
go
to
set,
for
example,
they've
got
this
kind
of
demo,
retina
viewer
type
in
the
word
metal
and
there
it
is,
and
you
can
type
in
aluminum,
for
example,
and
it
will
show
you
the
SDR
bit
mad
for
aluminum
the
side
metal
and
give
you
kind
of
the
overlap.
So
similarity
score
is
which
is
pretty
cool,
so
I'm,
going
to
try
and
use
that
so
I
have
an
experiment
that
tries
to
use
this.
But
let
me
go
a
little
further
with
the
details
of
using
zest
and
stuffed
API.
A
If
you
ask
the
septa
API
for
the
bitmap
for
metal,
it
gives
you
a
bitmap
so
it'll
give
you
a
heightened
with
which
is
always
128
128.
So
16384
bits
and
if
the
sparsity
is
somewhere
between
one
and
five
percent
on
the
dot
one
or
two
for
three
or
four
or
five,
it's
one
of
the
questions
that
I've
got
for
you
and
it
will
give
you
a
list
of
indices
that
are
on
for
for
that
bitmap.
A
So
you
can
convert
that
to
an
SDR
if
you
want,
but
this
isn't
really
very
useful.
Another
thing
that
it
will
do
is
if
you
constructed
an
SDR,
a
bitmap
and
you
send
it
to
set.
This
is
the
interesting
thing
it
will
give
you
back
a
list
of
similar
terms,
so
you
could
just
create
an
SDR
randomly
and
sit
and
say
what
sort
of
seems
like
this
back
a
list
of
things.
A
Usually
so
if
you
give
it
metal,
for
example,
it
gives
you
metal
as
the
number
one,
because
it's
an
exact
match
and
then
it
will
give
you
the
next
best
matching
terms.
I.
Think
100
of
them
are
so
so
there's
some
interesting
things
you
could
do
there
and
I'll
show
you
an
interesting
thing
that
I
did
after
I
talked
about
pie
sex,
so
I,
don't
like
using
HTTP
API
is
directly
so
I
made
this
hi
cept.
Things
like
you
get
to
it
again
from
here:
NOP
resources
from
the
hack
page
tools
and
projects
hi
cept.
A
So
it's
just
a
stepped
API,
client
and
Python.
Here's
where
you
get
it
I'm
not
going
to
demo
it,
but
I
will
kind
of
show
you
how
to
use
it.
It's
pretty
easy
import!
Biceps!
You
need
to
get
an
API,
ID
and
key
from
set.
So
you
can
do
that
online
through
the
through
the
website,
cept
80
and
you
need
to
get
an
upgrade.
A
You
need
to
upgrade
your
account
to
beta
program
and
talk
to
Francisco
in
the
back
to
do
that
or
there's
a
there's,
a
method
to
do
it
through
the
web
interface,
but
I
use
the
one
has
to
prove
it,
but
everybody
here
the
registers
are
going
to
prove
all
right.
Yes,
ok!
So
to
doing
it
to
do
the
interesting
stuff,
like
the
term,
to
our
the
SDR
to
similar
terms,
you
need
to
have
this
upgrade
to
your
account.
A
But
I
like
to
see
that
so
I
put
it
in
what's
useful
is
the
bitmap.
So
if
you
ask
it
for
a
bit
map,
it
will
return
you
an
object
that
contains
exactly
what
I
just
showed
you
earlier
I
with
sparsity
percentage
and
all
of
the
on
positions.
And
then,
if
you
take
that
bitmap
and
tell
it
bitmap
to
terms
it
will
go
to
this
page
and
give
you
back
all
of
the
similar
terms
in
this
format,
which
you
can
translate
into
something
like
this.
So
cat
dog
dogs,
animals
animal.
A
So
these
are
all
the
similar
terms
except
says,
so
you
can
do
some
interesting
things
with
this,
especially
when
you're
creating
the
SDRs
with
new
pic,
not
you
know
just
getting
an
SDR
from
stuff
than
sending
it
back
on.
So
what
is
this
so
I
tried
to
do
this
word
association
experiment,
so
I
took
two
lists
that
I
found.
We
could
be
here.
A
Something
and
I
tried
to
get
as
many
animals
as
possible
in
one
list
and
as
many
vegetables
as
possible
in
the
other
list
and
I
thought
I'm
going
to
take
a
random
animal
I'm
going
to
generate
its
str.
Ask
Seth:
what's
the
SDR
for
this
and
I'm
going
to
take
a
random
vegetable
and
asks
up?
What's
the
SDR
for
those
and
then
beat
it
into
a
pic?
So
here's
appears
an
animal
SDR,
here's
a
vegetable
SDR
and
until
it
reset,
which
basically
means
that's
the
power
of
the
pattern.
A
B
A
Nouns
was
a
failure
that
did
not
work
at
all.
Maybe
somebody
can
get
its
work
here
is
word
association,
so
there's
a
script
in
here.
Basically,
you
give
it
run,
Association,
experiment,
file1,
file2
and
some
of
the
other
options.
You
can
look
up
on
your
own,
but
it
will
take
a
random
term
from
file
one
random
turn
from
file
to
going
over.
A
How
many
iterations
you
tell
it
I
told
of
thousands
and
over
and
over
and
over,
to
send
that
you
know,
go
to
set
to
get
the
SDR
push
it
to
the
new
pic
and
then
stop,
and
then,
after
a
hundred
times
of
that,
so
it's
got
a
chance
to
learn
some
of
those,
the
values
and
those
SDRs.
It
will
start
going
to
the
screen.
So
here
you
can
see
something
happening.
After
a
hundred
times
we
sent
it
bison
we
sent
it
komatsuna,
which
I
that's
some
vegetable
that
came
from
list
too,
but
new
pic
predicted.
A
It
would
be
a
garbanzo
not
how
much
good
the
interesting
thing
about
this
is.
All
of
these
terms
are
plant
terms,
so
I'm
basically
just
asked
telling
an
animal
vegetable,
vegetable,
animal
vegetable
animal
what's
next,
and
it
always
gives
me
back
something
related
to
a
plant
because
it's
it's
basically
just
learning
a
semantic
Association
from
these
s.
Dr's
like
this
is
some
type
of
animal
blob.
A
This
is
some
type
of
vegetable
blob,
and
so,
when
you
ask
it
what's
next,
it
gives
you
back
something
but
sort
of
live
is
a
vegetable
so
I'm
interested
in
seeing
what
these
st
ours
look
like
like
it,
because
you
can
plot
these
bitmaps,
because
that
you
get
from
sat
and
kind
of
see
some
semantic
leaving
some
relationships
or
some
groupings
of
characteristics,
I'm
interested
in
seeing
what
the
SD
are
coming
out
of.
New
pic
looks
like
it
seems
like
it
might
be
sort
of
a
Gaussian
blur
ish,
looking
sort
of
thing
that
generally.
A
Yes,
the
seal
is
not
predicting
turmeric
turmeric,
it's
predicting
something
that
when
I
you
send
it
to
cept
and
say:
what's
the
closest
term
you
can
find
to
this.
It
says
turmeric,
so
so
it's
prob
but
I'm
interested
in
seeing
what
that
SDR
looks
like
and
comparing
it
to
what
Seth
says.
Tumeric
looks
like
all
right,
but
I
haven't
had
time
to
do
that.
Another
hack
opportunity.
A
A
No,
so
the
question
I'll
just
answer
the
question
and
I
previewed
it
I
feed
it.
The
first
thing
I
do
is
look
at
the
list
and
just
get
a
random
animal
and
I
send
out
two
steps
and
say
give
me
an
SDR
for
that.
It
gives
me
an
SDR
and
I
bypass
the
whole
opf
client,
stuff,
the
spatial
pool
or
etc.
You
know
what
I'm
talking
about
and
send
it
directly
into
the
SP.
A
So
I
translate
that
SDR
directly
into
TP
input
and
say
this
is
the
SDR,
the
next
str,
and
I
do
that
for
animal
vegetable,
animal,
vinyl,
vegetable,
but
every
time
you
feed
it
something
it
gives
you
back
prediction.
So
I
take
that
predicted
SDR
that
I
get
back
and
I.
Send
that
deceptive
ask
what
is
this
yeah?
C
A
A
From
what
I
can
tell
no,
so
the
question
was:
does
the
SDR
capture
does
the
semantics
from
the
SDR
capture?
What
part
of
speech
it
is?
There's
I
still
have
some
experiments
to
do.
My
singular
plural
experiment
totally
failed
because
I
don't
think
there's
any
representation
of
singularity
or
plural,
singular
or
plural
within
the
STRs
that
cept
produces
so
and
I.
Don't
know
about
parts
of
speech,
I
haven't
tried,
but
there's
a
man.
You
could
ask
right
back
there.
It
probably
knows
that
answer.
So,
if
you
would
you
like
the
answer,
francisco's.
B
A
B
Basically
takes
every
word
as
a
word,
and
it
tries
to
associate
the
SDR
for
the
word,
so
it
makes
it
makes
no
difference
between.
So
the
into
the
capturing
of
the
semantics
doesn't
make
any
difference
in
terms
of
plural,
a
lot,
because
this
is
supposed
to
be
captured
by
the
language
itself.
So
basically
doing
the
STRs.
One
of
the
goals
was
not
to
capture
anything
within
the
SDR
that
shouldn't
be
there,
because,
it's
just
so
to
say
a
concept
that
we
create
to
classify
words
and
so
on,
but
we
should
by
using
it.
B
B
What
we
do
in
a
second
step
is
that,
as
soon
as
we
have
the
STRs,
we
then
try
to
capture
additional
information
like,
for
example,
the
part
of
speech,
but
this
is
stored
separately,
and
this
is
only
used
say:
I
want
to
have
sdrs
that
are
only
nouns,
for
example,
because
I'm
looking
for
a
noun,
so
I
will
get
the
closest
matching
word,
which
is
a
noun,
but
this
is
just
to
do
further
experiments.
On
top
of
this,
thanks.
B
A
So
I
those
are
the
only
experiments
that
I
had
a
chance
to
do,
but
it's
this
is
really
fun
stuff.
I
wish
that
I
could
just
do
more,
so
that's
I'm
planning
on
doing
some
of
this
stuff
with
this
a
crown
at
that
time.
Some
of
the
tools
I
talked
about
in
ltk
there's
also
an
extension
to
NLT
k.
One
of
my
co-workers
mentioned
one
of
the
problems
I
was
getting
with
an
l
TK
was
it
was
poor,
part-of-speech
tagging.
A
So
if
it's,
if
it
does
a
bad
job,
tagging
at
parts
of
speech
and
I've
son
David
into
new
big
Nick
learns
incorrect
patterns.
So
maybe
this
extension
could
could
help
out.
So
if
anyone
wants
to
try
and
improve
my
in
otk
interface
would
be
great
and
cept
again.
This
is
a
very
cool
tool
for
dealing
with
STRs.
A
One
thing
I'll
note
about
s
quieras
coming
out
of
step
is
that
they
are
not
the
sparsity
level
of
each
str
is
not
normalized
and
for
new
pic
we're
rien
ticipate
that
they
should
all
be
at
least
the
same
sparsity
two
percent
is
what
we
use.
The
SDR
is
coming
from
scepter
anywhere
from
one
to
five
percent.
I
did
not
attempt
to
normalize
them,
so
this
is
really
not
optimized
at
all.
The
code
that
you
saw
that
was
guessing
vegetable,
like
things
is
not
optimized
for
new
pic,
it's
just
raw
SDRs
coming
from
stopped.
A
There
are
other
experiments
that
people
have
done.
Shaitan
has
done
some
one
of
our
community
members
is
done,
one
called
linguist
and
go
back
to
this
resources,
page
tools
and
projects.
So
there's
a
few
other
experiments
here.
Linguists
data
feeds
characters
at
the
time,
so
I
don't
think,
is
dealing
with
words
yet,
but
these
are
all
good
examples
of
implementations
of
in
NOP
type
projects
with
new
pic
linguist
uses
of
the
OPF
right
and
the
new
Big
Ten
Opie
has
one
experiment
that
uses
the
OPF
another
that
goes
directly
to
the
TP.
A
A
Okay,
oh
yeah,
and
there
are
some
challenges
here
to
switching
back
and
forth,
or
was
it
on
this
NOP
wiki
page?
If
you're
looking
for
something
to
do,
I
couldn't
get
this
to
work,
I
really
what
I
really
wanted
to
do
something?
But
I
could
say
cat
cats,
dog
dogs
and
then
till
it
cactus?
No,
it's
a
cacti,
not
cactuses,
so
I
did
this
experiment
totally
failed.
I!
Think
it's
just
because
it's
not
that
florala
t
is
not
represented
in
the
stairs
get
back
from
stuff,
but
this
would
be
a
cool
thing
to
try.
A
Another
thing
is
a
Google
was
bragging
about
their
ability
to
guess,
company
capitals
or
country
capitals
with
their
words
of
X
tool
and
I
tried
to
do
it
too,
but
it
but
failed
with
that.
So
it's
another
thing
that
somebody
could
try
out.
That
code
is
also
in
this
NOP
project
I.
Just
it's
just
a
word.
Association
I
just
took
a
list
of
countries
and
list
of
capitals,
something
like
that.
Where
else
I.
A
Lost
my
place:
oh
there
we
go
so.
Another
thing
part
of
speech
is
anomaly:
detection
which
I
mentioned
earlier.
It
would
be
really
cool.
This
might
be
something
I,
try
and
work
on
it.
I
don't
have
much
time
to
Tina.
My
team
up,
so
I
probably
won't
be
very
effective
at
this,
but
it
would
be
really
cool
if
you
had
like
a
CLI
interface,
where
you
could
just
type
words
type
sentences
feed
it.
A
Some
example
text
effects
that
you've
written
and
then
type
stuff
in,
and
it
predicts
the
next
part
of
speech
that
you're
about
to
type,
and
then
you
could
kind
of
follow
it
down
the
rabbit,
hole
and
see
if
you
can
create
cognizant
english
language
just
based
on
what
it's
telling
you,
you
should
be,
what
speech
you
should
be.
Writing
and
I
think
it
also
would
be
interesting
to
to
write
some
code
that
takes
the
SDRs
that
come
out
of
New
pick
out
of
the
TP.
A
A
So
that's
it!
That's
is
that's
the
much
NOP
guidance
as
I
can
give
you
at
this
point,
so
you
guys
are
effectively
on
your
own
free
to
start
hacking.
We've
got
another
presentation
here
at
the
one
o'clock
I
believe,
and
that
will
be
a
suit
I
talking
about
CLA
ACL,
a
deep
dive,
so
that
could
go
a
little
bit
long,
because
it's
a
big
topic,
but
if
you're
interested
in
understanding
how
the
CLA
works
and
not
just
creating
things
with
it,
you
should
probably
come
to
that
talk.
I
do
I.
C
Don't
know
how
many
people
have
understand
it.
You
know
with
natural
language
processing
is
not
something
we're
working
on
a
croc
at
all,
something
very
curious
about
something
that
francisco's
from
working
on,
and
you
know
the
theory
of
CLA
and
the
hero.
Sparser
sugary
representation
suggests
that
before
you
could
do
a
lot
in
this
space,
I
still
want
them
anymore.
Think
this
is
the
only
thing
you
have
to
work
on
here
and
we've
had
a
long-running
debate
over
the
years.
Whether
you
could.
C
Actually
you
know
in
the
brain,
we
formed
the
spar,
stupid
representations
from
low-level
sensory
input
and
Francesco's
come
away
of
sort
of
generating
through
an
online
process
that
I
can
say,
it'll
be
if
you
want
so
fate
and
whether
you
can
shortchange
what
brains
do
and
you
know,
go
right
to
the
STRs
and
he
hasn't
really
heat
and
he
touched
a
lot
of
success.
Creating
these
but
the
whole
experiments
you
are
about
putting
through
sequences
and
that's
what
we
do
it.
We,
the
CLA,
does
very
well
this
one
make
sure
that
hey
you
don't
think.