►
From YouTube: IPFS All Hands 🙌🏽📞 Nov 6, 2017
B
C
D
B
B
So
we,
a
couple
of
us
from
ipfs,
has
been
kept.
Calm,
I
think
maybe
Jeremy
could
give
an
overview
of
depth
on
in
general
at
one
point
later,
but
what
I?
What
I
wanted
to
share
is
that
we
had
a
specific
event
a
couple
of
days
ago,
after
calm,
where
we
talked
specifically
about
how
these
different
decentralized
technologies
can
help
with
censorship
and
other
things
specifically
in
Catalunya
and
trying
to
come
up
with
solutions
could
help.
Try
I
think
the
event
ended
up
being
100
people
more
or
less.
B
There
was
a
huge
interest
in
trying
to
come
up
with
version
which
was
very
cool
to
see,
and
we
split
up
into
smaller
groups,
maybe
10
15
people
in
each
group
and
I
was
leading
a
group
with
focused
on
how
ideas
can
help
with
censorship
things
and
we
came
up.
We
were
pretty
much
quickly
established
that
ipfs
is
ready
to
be
used.
The
anonymity
could
be
better,
but
it
sits
well
enough.
B
Now
the
biggest
problem
comes
on
how
to
discover
a
content
and
where
to
covered
content
from
so
to
be
able
to
offline
share
hashes
with
an
impaired
or
pass
around
USB
sticks
and
give
tools
to
replicate
these
USB
sticks
and
also
leverage
the
existing
mesh
networks
that
exist
in
catalonia
today,
griffie,
which
I
think
is
the
second
biggest
mush
network
in
the
world.
Right
now,
so
we
came
up
with
some
some
ideas
of
how
we
can
move
forward
with
this.
A
Cool
and
is
that
the
mesh
network
is
that
different
from
I
know
and
Laurie
I.
Just
yesterday,
someone
came
to
my
news:
feed
about
mesh
networks
for
and
and
that
I'd
actually
looked
as
LRA's.
Actually
you've
been
involved
in
this
mesh
network
and
is
that
different
for
a
mesh
network
Wi-Fi
or
is
further
from
the
protocol
level
level
layer
two
or
is
it
actually
like
ipfs
feeling
the
content?
On
top
of
the
hardware
mesh
network
yeah.
B
So
we
see
is
it's
working,
bye-bye
antennas,
that's
a
point
to
pearson
on
your
pure
peer
with
antennas
and
ipfs
would
just
be
used
to
be
able
to
pass
around
the
content,
and
we
also
have
been
working
with
the
quick
or
we
were
starting
to
work
now
with
quickly
peep
people
to
build
distributed,
DNS
service
for
their
internal
network
as
well.
So
me
and
Loras
met
with
them
in
the
hack
day.
B
B
It
was
quickly
censured
by
the
Spanish
government
and
eventually
the
Spanish
government
to
control
over
the
doctor,
which
is
or
local
top-level
domain,
so
people
started
mirroring
the
website
on
different
domains
and
the
Spanish
government
continued
to
to
block
other
domains
on
the
ISP
level
and
then
eventually,
I
think
it
was
the
pirate
party
during
colonial
who
put
the
website
on
aquifer
and
the
Catalan
president
tweeted
a
link
to
this
website,
and
he
was
quickly
blocked
as
well.
The
DNS
of
our
food
gateways,
which
also
meant
that
other
things
were
blocked
as
well.
D
B
So
currently
the
when
you
are
pinning
content
or
when
you
are
adding
content
to
ipfs.
We
cannot
share
that
with
your
notes.
So
it's
kind
of
obvious
which
content
you
are
adding
and
things
like
that
people
were
a
bit
worried
that
you
wouldn't
be
able.
It
will
be
easy
to
sniff
and
figure
out
who
is
hosting
what
content
and
also
the
lack
of
audited
tor
transports,
but
some
people
off
as
well.
C
B
E
Yeah
we
get
to
put
a
few
best
in
the
game,
which
is
our
music.
That's
that
I
talked
to
a
few
other
different
blocked
in
groups,
and
they
were
also
all
really
excited,
especially
the
ones
who
are
reading
something
I'll
have
to
look
at
that:
I'd
be
really
cool,
and
so
hopefully
there's
a
lot
increased
interest
soon.
E
A
A
So
one
thing
that
we
were
going
to
talk
about
with
the
four
demos
was
the
JSI
PFS
is
working
on
Safari,
which
I
saw
leave
posted
on
Twitter,
so
I
asked
someone
to
give
a
demo
and
so
I
think
Victor
head
I'm
here
to
discuss
it.
But
actually,
if
someone
has
Safari
Dawid,
if
you
have
Safari
running
I
should
be
great
to
go.
Yeah.
F
F
D
F
Okay,
like
so,
and
you
see
that
it
says
online,
so
if
you
go
to
the
JavaScript
console,
can
you
still
see
the
JavaScript
console
yep?
Alright?
So
before
you
would
see
a
bunch
of
errors
here,
it
would
say
something
with
web.
Crypto
is
not
there
or
something
with
WebRTC
was
not
there,
but
now
this
is
like
a
stream
and
a
pure
material.
It's
just
a
connection
that
was
close.
That's
fine
but
like
there
is
no
error,
so
you
can
open
the
same
link
now
I'm
going
to
stop
screen
sharing.
F
F
F
It
was
pretty
much
and
I
could
play
it
on
Safari,
so
Safari.
Basically,
the
support
team,
the
WebKit
iam
as
I've
been
working
on
web
RTC
and
apparently
they
not
only
implemented
web
RTC.
They
also
implemented
the
rest
of
the
web,
crypto
primitives
that
were
missing
and
we
just
happened
to
discover
it
by
surprise.
Dream
was
fast
because
someone
grabs
an
iPhone
and
tested
it
on
their
mobile
device.
F
F
So
there
is
like
two
things
that
are
being
shared
their
gear
pair.
That's
two
modes:
one
is
like
read-only
another
one
is
right:
Plus
read
so
to
have
the
read-only
mode.
You
just
need
a
key
to
decrypt
the
messages
that
go
through
the
pub/sub
channel
so
that
the
try
it
gets.
Private,
like
you,
need
to
have
a
key
to
I,
have
access
to
the
messages
of
the
room,
otherwise
just
receive
random
lobs
of
data,
and
then,
in
order
for
you
to
be
able
to
write
the
mechanism,
to
give
permission
to
write
in
fear
pad.
F
F
So
so,
as
always
explaining,
there
is
a
key
for
signing
the
messages
and
and
appear
pairs
instance
will
only
accept
the
messages
that
it
can
validate
the
signature,
and
it
will
only
take
those
messages
and
add
its
to
the
CR
DT
and
then
like.
If
everything
checks
out,
then
it
creates
a
new
state
locally
and
then
injects
had
move
state
into
the
viewer
into
the
editor.
F
So
if
you
don't
have
the
right
key,
so
the
signing
key,
you
can
still
view
if
you
have
a
decrypt
key,
but
you
cannot
you
they
all
appears
will
never
accept
them
your
messages
because
they
will
consider
it
that
me
about
it
did
I
manage
to
like
explain
any
any
questions
on
his
business.
Since
we
last
met
okay.
A
So
I
think
guys
know
I
had
a
demo
working
with
pub
sub
room
I.
Think
I
was
Hector's
work
with
the
IP
ID
respect
that
I
built
for
the
D
D
ID
on
top
of
IP
FS,
so
that
still
isn't
working
and
some
I'm
thinking
and
I
still
I
think
the
web
crypto
shim
file
is
is
still
acting
up
in
Safari.
If
it
works,
fine
in
chrome
and
so
I
think
is
there
is
the
same
libraries
being
used
in
pub
sub
room?
Does
anyone
know
yeah.
F
So
it
should
be
like
so
Pierre
pad
also
uses
pops
of
room.
So
what
might
be
happening
is
one
of
two
things
or
you
don't
have
the
latest
Safari
version
in
your
in
your
machine
or
you
have
some
old
dependency
of
the
leap,
peer-to-peer
stack
of
modules
and
because
you
might
be
installing,
with
NPM
five
year
likes
package
lock
or
if
you
are
using
yarn
yeah
like
this
lock
file,
and
so
my
advice
to
for
you
to
try
just
remove
that
lock.
Well,
remove
the
and
no
more
just
Folger
and
install
everything
again.
A
D
A
A
B
F
There
is
like
some
things
that
are
still
rough,
especially
like
if
you
open,
multiple
tabs,
so
that
that's
like
a
known
bug
and
I
can
heart
into
and
all
if
you
open,
multiple
tabs,
we
had
two
options
or
we
would
create
multiple
ipfs
nodes
or
we
will
try
to
use
one
nodes
and,
as
the
other
tabs
synchronized
through
the
indexdb
itself,
and
so
we
are
trying
to
do
that.
So
I
can
instead
of
a
key
opening,
multiple
nodes,
so
multiple
notes
creating
a
tht
presence.
F
D
F
Just
storing
your
browser
and
all
the
other
browsers
have
to
share
that
URL.
We
don't
have
any
pinning
service
attached
pfl
right
now
that
stores
a
little
for
you
also,
if
we
end
up
doing
that,
we
will
only
store
the
included
box
so
like
no
other
service
will
actually
be
able
to
like
see
the
data
that
you're
storing
everything
will
be
encrypted,
and
so
it
will
always
require
you
to
provide
a
key
in
order
to
like
see
what
was
the
content
of
the
pad,
which.
G
Okay,
yep
call
yeah,
so
we've
been
working
a
little
bit
over
a
together
on
some
of
this
archival
work
and
I'd,
say
I
wanted
to
demo
something
pretty
quick.
It's
calling
this
an
alpha
would
be
a
stretch.
It's
a
proof
of
concept
at
most,
but
trying
to
get
the
work
specification
to
work
properly,
with
sort
of
IPS
hash
structures
and
so
yeah
I
feel.
Why
not
sure
my
screen
will
see
what
what's
going
on
here
so
yeah?
Can
everybody
see
this?
Okay,
that's
working.
H
G
Quick
sort
of
reference,
the
archive,
so
we
get
the
actual
date
of
ingest
and
then
the
URLs
that
we
archives
and
clicking
this
works
properly.
You
can
actually
see
the
style
sheets
and
then
same
thing
with
this
thing
things
like
videos-
this
is
mr.
Jim.
Well,
it's
wonderful.
The
code
for
live
talk
is
obviously
not
included.
This
is
a
like
a
very
early
version,
but
just
focusing
on
sort
of
getting
some
basics
in
place.
Yeah,
it's
kind
of
quick
demo.
G
What's
going
on,
so
we're
using
this
a
bunch
in
the
context
of
trying
to
archive
government
data
and
try
to
get
clean,
consistent
archives
of
URLs
and
to
be
able
to
sort
of
duplicate
them.
There's
a
whole
bunch
of
stuff
in
here,
but
I'm
hoping
to
sort
of
just
drive
everybody's
attention
to
this,
and
if
you
have
any
interests,
you
can
turn
this
over
unto.
Did
the
other
repos
yeah
stop
traitor.
G
G
Have
the
rewrite
of
those
URLs
sort
of
match
so
that
multiple
iterations,
multiple
archives
of
the
same
thing
will
result
in
the
same
hash
and
so
rewriting
the
URL,
so
I'll
be
relative
to
the
root
of
the
hash,
but
also
using
the
domain
name
as
sort
of
the
folder.
The
top-level
folder
seems
to
be
a
good
starting
point,
but
I'm
not
totally
sure
that
that
will
get
clean,
consistent,
deduplicating
records.
There's
a
lot
of
work
to
be
done.
G
D
D
D
What's
relevant
to
the
few
pet
demo,
is
that
with,
when
you
have
content
on
ITSs,
then
you
need
to
figure
out
which
nodes
should
be
holding
the
content,
and
so
when
you
apply
that
with
an
idea
of
a
collective
asset,
so
whether
that
a
government
data
set
or
is
it
pure
pads
that
was
collaboratively
edited
by
three
people
from
the
same
company?
You
need
to
figure
out
where
you're
going
to
pin
that
stuff
and
who
gets
to
decide
what
gets
pinned
and
where,
and
so.
This
is
where
data
together
is.
D
This
kind
of
side
project
is
trying
to
tackle
that
more
in
general
and
tackle
those
problems,
that's
how
they
apply
to
the
defense
like
web
in
general,
not
just
icsf,
so
we
happen
to
be
building
tooling,
like
what
Brendan
showing
that
is
doing
this
with
our
TFS
tool.
But
it's
also
something
that
just
applies
in
general
with
any
Content
that
story
using
distributed
system
potential
item.
A
G
Yeah
so
currently
it's
all
being
written
into
a
work
file
which
actually
exists
at
the
of
the
archive
which
keeps
all
in
that's
Britney,
in
accordance
with
the
work
spec.
The
last.
The
next
thing
to
ship
is
to
actually
pull
the
body
of
the
workout
and
just
reference.
The
hash
of
the
actual
block
on
that
give
us
so
as
of
right
now,
you're
getting
duplicate,
writing
so
I,
wouldn't
recommend.
This
is
like
a
used
today
tool
in
the
future,
we'll
sort
of
make
it
so
just
references
the
hash
and.
C
A
G
Work
as
though,
as
our
partner,
it's
an
ISO
specification.
It's
really
with
the
old
back
from
the
table.
They
use
character.
Turn
newline
is
like
a
way
of
delimiting
things
yeah.
So
it's
the
nice
thing
is
it
does
give
you
everything
you
need
to
understand,
sort
of
what
came
from
that.
There
is
a
bit
of
a
privacy
concern
there,
though,
as
well,
because
the
work
spec
does
request
that
you
record
your
IP
address
when
you
do.
F
Just
added
some
stuff
and
Jeremy
also
added
one
item.
There
are
quick
things
just
like
announcements,
making
sure
everyone
is
on
a
little
I
go
yeah,
good
cool,
slow,
three
things.
One
of
them
is
so
you
saw
the
demo
of
peer
pads,
as
I
said,
like
we
wear,
this
conference
was
fast
in
London
for
a
weekend
and
we
spent
like
two
days
there
I
create
a
bunch
of
sessions.
F
There
were
a
bunch
of
discussions
which
of
those
will
be
converted
to
blog
posts
that
we
wouldn't
publish
for
everyone
to
see,
but
one
of
those
sessions
was
actually
a
workshop,
and
for
that
workshop
was
explaining
teaching
out
to
UCR.
It
is
an
idea
fest
to
create
apps
like
peer
pad
and
what
we
did
was
we
created
a
repo
called
peer
to
peer
flip
chart
which
helps
you
create
a
thing
with
the
same
charity
library
that
PayPal
is
using,
but
to
create
a
flip
chart
on
a
web
page.
F
So
you
can
like
at
your
participatory
meetings
and
everyone
can
click
chart,
make
drawings
and
take
notes
instead
of
using
text,
and
so
there
is
a
flyback
that
I
linked
here
on
the
notes
which
adds
like
what
was
the
workshop
step
by
step.
We
want
to
record
it
as
a
video
as
well,
so
it's
easier
easier
to
follow,
but
you
can
check
the
code
today
on
my
PFS
shipyard,
pew
pew
flip
chart
and
see
how
it
works.
F
It's
it's
kind
of
like
a
super
version
of
peer
pad
because
it
doesn't
have
the
crypto
and
just
as
the
sinking.
So
it's
good
to
get
started.
The
next
thing
is,
there
has
been
a
lot
of
development
on
the
windows
support
for
just
because
it
is
something
that
the
community,
just
like,
appeared
and
started
reading,
so
Richard,
no
trophies
here
well
he's
from
New
Zealand,
so
different
time
zone.
F
As
being
you
know,
like
going
module
by
module
and
leveling
up
failure
and
like
running
the
tests
on
Windows
and
like
fixing
the
the
things
that
we're
missing,
especially
on
a
lattice
or
layer,
to
make
sure
that
Jessica's
can
run
on
Windows.
So
I
expect
to
have
Detective
s
running
on
Windows
by
the
next
release,
0:27
and
last
I.
Don't
the
least
is
something
that
I'm
really
excited
about.
So
if
you're
used
like
by
the
way,
if
you
have
questions
raise
your
hand,
the
last
one
is
a
new
streaming
and
buffered
API
for
Jessup
EFS.
F
The
problem
is
the
readable
stream
shim
for
the
browser
is
kind
of
like
very
memory
intensive,
which
is
a
problem,
so
we
started
seeing
users
reporting
that
they
were
having
trouble
troubles
using
our
API,
still
pull
of
files
on
300
Mike's
and
more
and
yes,
there
were
some
performance
issues.
There
are
some
things
on
the
crypto
lands
but
like
after
fixing.
F
But
if
you
are
familiar
with
pou
streams,
you
can
use
books
frames
directly,
which
is
a
thing
that
we
use
internally
in
just
a
few
cents
or
if
you
don't
want
to
think
about
streams
at
all.
If
you
are
just
like
handling
with
files
like
small
files,
less
than
a
gig,
you
can
just
allow
us
to
give
you
a
for
with
the
file
and
you
get
the
file
and
you
can
use
it,
and
so
a
lot
of
developers.
Streams
are
still
like
thing.
F
That's
not
natural,
so
I
expect
that
a
lot
of
people
will
be
able
to
pick
up
on.
These
API
is
to
like
create
more
interesting
locations
are
like
piphus
because
it
removes
extremists
from
the
equation.
If
you
don't
want
them,
and
that
is
something
that
I'm
going
to
release
on
the
next
version
as
well
3:37,
it
is
there's
already
an
implementation
on
Jessica's
API,
so
that
will
ask
tradition.
We
implement
it
always
for
API,
and
so
that's
like
one
single
interface
for
both
plans,
and
there
is
a
pull
request
with
a
spec
already
be
fine.
F
F
F
It
will
be
also
a
specific
branch
that
well,
if
you
are
unfamiliar
with
the
process
of
developing
things
online,
we
always
have
like
work-in-progress
pull
requests
where
we
just
like
from
the
beginning.
We
open
the
pull
request
so
that
everyone
is
aware,
like
that
fort
is
happening
and
when
it's
time
to
review
using
quite
the
reviewers
and
then
merge
it,
and
so
let
me
lean
here
the
pull
request
for
the
interface,
which
is
where
I
would
love
to
get
your
feedback.
F
E
New
or
these
kind
of
it
up
for
over
12
dozen
few
fixes
front
that
were
not
present
enough.
For
that
really
spill,
rubies
can
I
get
one,
notably
the
Badger
alternative
data
store.
Experiment
works
a
lot
better
now
and
actually
build
platforms.
So,
theoretically
that's
the
one
to
know,
and
you
can
use
badger
and
shouldn't
worry
about
it,
but
we're.
E
Yeah
so
right
now,
ipfs
uses
a
combination
of
level
BB
and
files
on
this
and
some
tree
to
store
all
the
data
turns
out.
Storing
files
on
this
is
actually
not
that's,
especially
given
all
the
defeats
or
spells
you
have
to
make
for
different
directories
and
syncing
change-making
directories
and
storing
it
a
file
for
each
block.
The
overhead
free
trial
is
actually
nonzero,
so
badger
gives
us
a
single
M
map.
Sorry.
E
F
I,
like
I,
see
there
is
a
question
from
Sam
Holmes
here
on
materials,
about
charities,
just
wanted
to
point
out
that
we
have
a
repo
research
RG
and
it's
just
like
a
collection
of
materials,
about
charity,
stalks
papers,
blog
posts
and
also
issues
of
things
that
people
are
like
experimenting
with
right
now
on
top
of
ipfs.
So
if
you
want
like
to
learn
about
charities,
there's
a
lot
of
materials
get
started
there
or
if
you
want
to
follow
those
discussions,
and
they
are
typically
a
good
point
to
start.
A
H
Wanted
to
give
just
a
quick
introduction
to
the
project
that
isn't
fully
released
yet,
but
it
seemed
like
a
good
opportunity
to
tell
people
about
well
what
was
up
with
it,
and
then
we
can
go
into
more
detail
if
people
are
interested
either
offline
or
in
another.
One
of
these
fellows
so
for
anyone
not
familiar
with
tensorflow
I
mean
turn
on
screen,
shooting.
H
Okay,
so
can
can
folks
see
my
my
desktop
yep,
okay,
great,
so
tags
are
flow
ease
and
is
a
those
are
not
familiar.
It's
a
machine
learning
framework
and
toolkit
for
the
research
community
and
for
folks
to
be
running
machine
learning
in
production.
It's
a
very
official
Google
project.
They
they
have
a
lot
of
resources
on
it.
It's
extremely
extensible.
H
They
will
publish
their
results
and
know
sometimes
publish
source
code
and
so
they're
running
these
models
and
these
systems
across
terabytes
of
data
across
either
one
machine
make
tens
of
$10,000
or
dozens
or
hundreds
or
in
some
cases,
thousands
of
machines.
Turning
through
terabytes
of
data
trying
to
train
these
models.
H
So
tensorflow
has
an
interface
that
you
can
add
file,
system,
implementations,
and
so
they
have
actually
five
implementations
of
how
to
read
and
write
data
into
tensorflow
POSIX
Hadoop,
Google,
Cloud
Storage,
our
memory
Matt's
file
system
and
not
written
here,
but
there's
also
an
s3
file
system.
So
the
interface
for
adding
your
own
file
system
is
C++,
so
I
spent
a
little
while
combining
ipfs
and
tensorflow
in
making
IPPF
that
is
interplanetary
tensorflow.
H
This
implements
that
C++
interface
and
then
provides
a
new
URL
scheme
which
I've
gotten
a
little
bit
of
feedback
from
the
ipfs
team.
We
haven't
finished
that
discussion
yet,
but
the
premise
is
anything
that
you
start
with
IETF
colon,
slash,
slash
intensive
flow,
actually
jumps
out
to
the
like
ipfs,
Damon
and
or
network,
as
appropriate.
So
I
can
offer
a
quick
demo
I
just
downloaded
this
random
I.
H
H
So
if
I
for
Speed
I'm,
actually
gonna
tell
it
to
run
offline,
if
you
don't
about
to
note,
it
starts
up
its
online,
and
so
this
is
actually
running
the
daemon
in
process
and
using
a
sequence,
we'd
base
bridge
to
shuttle
data,
some
C++
in
to
go,
etc.
So
this
thing
is
loading
pre-press
files.
This
is
the
model
running
and
just
a
moment
we'll
see
you
start
doing
some
tensor
tensorflow,
like
normal
logging
operations.
That's
good
does
that.
Oh
thanks,
Allah
and
I'll
show
you
guys
the
other
half
of
this.
H
H
H
These
are
that
all
the
files
files
that
it's
writing
or
appending
to
you
et
cetera,
the
API
that
tensorflow
uses
to
write
to
file
systems
is
actually
a
surprisingly
good
fit
for
ipfs.
You
cannot
randomly
modify
a
file.
You
can
only
append
or
three
files,
so
it's
really
nice
time
shot
in
the
relatively
high
performance.
So
you
can
see
this
thing
is
running.
It's
running
really
really
slowly,
if
they're
running
on
GPU
etc,
much
faster,
which
is
running
in
the
cloud
or
much
much
faster.
Okay,
so
I'll.
Stop
that
right.
H
H
A
H
Thank
you,
I
can
try
to
reduce
the
visual
complexity
of
here.
We
go
okay,
so
this
is
the
tensor
board
interface
and
if
you
see
down
here
in
the
bottom
left
corner,
it
might
be
a
bit
small.
It's
reading
from
the
top
level
IP
TF
logs
directory-
and
you
can
see
a
couple
of
the
test-
runs
I
did
earlier
this
morning
and
you
can
see
that
this
is
the
actual
training
operation.
So
one
of
the
things
that
makes.
C
H
Special
is
this
is
publishing
to
an
by
default,
an
IPS
namespace,
which
means
that
anybody
that
runs
ten
port
on
their
machine
can
pull
logs
off
of
your
machine.
While
your
model
is
training
and
there's
more
advanced
things,
you
can
do
all
the
things
you
can
do.
A
tensor
board
work
with
this
because
it
looks
just
like
a
normal
file
system.
H
This
is
extremely
sensitive
to
high
performance.
Like
you
know,
you
want
to
be
able
to
read
from
these
things
and
as
quickly
as
we
possibly
can
so
one
of
the
big
things
to
do
for
this
is
making
sure
that
this
is
just
a
local
file
system
where,
as
close
as
we
can
get
to
it,
which
is
why
I'm
actually
running
a
daemon
in
process
rather
than
via
the
HTTP
eight,
because
researchers
and
folks
running
production
will
just
reject
something
that
wants
to
slogan
so.
H
H
Ipfs
overlay,
Network
and
if
I
were
to
click
on
logs
or
save
you'd,
be
able
to
actually
browse
the
actual
loans
that
were
generated
by
by
this
in
the
normal
I.
To
do
this
so
again,
this
is
a
call
ITTF,
it's
something
that
a
lot
of
researchers
are
extremely
excited
about
because
of
a
rough
and
flex.
It
is
to
share
data
sets
today,
sort
of
the
best
way
to
get
data.
H
Sex
right
now
is
something
called
the
best
way
to
get
them
quickly
is
something
call
academic
Florence
and
it's
an
out-of-band
way
to
download
data
ahead
of
time.
You
still
need
to
unpack
etcetera
and
with
IPTS,
you
read
a
you
know
a
particular
file
deep
in
a
data
set
and
you'll
fetch
it
just
in
time
to
the
for
the
machine,
learning
workers
that
are
doing
this.
So
that's
the
demo
that
I
have.
If
there
any
questions,
I
can
answer
or
we
can
talk
about
stuff.
That's.
H
Sway
back
LAN
at
MIT
I've
been
out
in
San,
Francisco
were
working
on
startups
and
programming
languages
and
shuted
systems
and
I've
spent
the
last
year
working
on
building
a
suite
of
tools
for
people,
doing
machine
learning
and
people
trying
to
apply
machine
learning
to
software
businesses
so
yeah.
That's
it
that's
a
super-high
level.
H
This
is
a
part
of
a
large,
comprehensive
suite
of
tools
for
making
it
possible
to
just
download
datasets,
like
you
would
watch
movies
on
Netflix
to
be
able
to
download
models
and
apply
them
just
like
you
would
download
an
app
from
an
app
store
or
a
Python
library
from
my
past.
So
that's
just
a
bit
of
background,
but
my
audio
quality
is
all
right.
A
D
A
H
So
the
IPT
f
provides
access
to
both
the
IP
NS
and
ipfs
namespaces,
so
datasets
will
be
published
using
IP
IP
FS
hashes
for
stability,
because
reproducibility
is
extremely
important,
but
when
publishing
logs
across
machines,
I'm
using
IP
NS
and
so
the
resolution
time
isn't
great
I'm
figuring
that
hospital
with
rising
type
will
raise
all
the
boats
here.
So
for
some
of
their
working
folks,
a
bit
that.
D
D
Yeah,
so
the
thing
I
wonder
in
the
Germans
Dawid
around
here
is
a
wonder
if
it
would
actually
be
worth
you
looking
into
using.
Instead,
if
you
have
a
constraint
number
of
machines
who
are
wanting
to
follow
those
updates
and
that
feed
of
that
information
propagating
is
important,
you
might
get
much
better
performance
reviews.
Awesome
I
might
be
wrong
there.
Jeremy
am
I
right
there.
F
So
you
publish
those
records
to
the
HD
and
then
other
peers
kinda,
like
fetch
those
records,
but
there
is
no
limitation
to
like
publish
those
records
to
topsail
as
well,
and
there
is
something
that
I
believe
visa
is
actively
working
on
the
IP
offense.
So
what
this
mean
is
like
soon,
you
will
be
able
to
use
IP
NS
and
get
fast
updates
across
multiple
nodes
without
even
to
think
about.
F
If
they
go
through
the
DHT
or
if
they
go
through
pops
up,
you
can
just
like
I
wish
your
name
as
you
always
would,
and
if
your
node
is
connected
is
online.
When
other
nodes
are
subscribing
to
that
IP
NS
name,
then
they
will
get
faster
updates
because
the
nodes
can
find
each
other
and
have
this
direct
connection.
D
D
J
H
Okay
yeah,
that
sounds
that
sounds
that
sounds
great
I
mean
the.
There
are
some
other
things
that
that
are
part
of
tensorflow
that
are
a
fantastic
fit
for
some
of
the
peer-to-peer
channels
in
ipfs,
and
so
I'm
looking
forward
to
expanding,
making
it
possible
to
sort
of
elastically
grow
the
cluster
of
workers
working
on
a
problem
and
doing
all
that
coordination
over
over
at
EFS
right
now,
I
just
model.
H
If
you
seem
to
disc
I,
published
the
IP
NS
handle
so
lower
latency
updates,
when
you
do,
that
would
be
maybe
really
magical
and
a
lot
of
the
research
community.
That
I've
shown
this
to
has
been
really
excited
about
being
able
to
apply
this.
So
hopefully
it's
a
much
larger
and
enthusiastic
and
well-funded
audience
for
ipfs
to
benefit
from
exposure
to
very
cool.