►
Description
Join CS50's Professor, David J. Malan of Harvard University, and Copilot Product Leader, Shuyin Zhao, for a fun chat about GitHub Copilot, including how it works, how it can be used in today’s classrooms, and how it might shape tomorrow’s classrooms as well.
As always, feel free to leave us a comment below and don't forget to subscribe: http://bit.ly/subgithub
Thanks!
Connect with us.
Facebook: http://fb.com/github
Twitter: http://twitter.com/github
LinkedIn: http://linkedin.com/company/github
About GitHub
GitHub is the best place to share code with friends, co-workers, classmates, and complete strangers. Millions of people use GitHub to build amazing things together. For more info, go to http://github.com
A
Hello,
world
and
Hello
Universe,
my
name
is
David
Malin
and
I
teach
a
course
called
cs50
at
Harvard
University,
which
is
our
introductory
course
in
computer
science
for
majors
and
non-majors.
It's
a
course
in
which
we
use
c
as
our
first
programming
language,
followed
by
python,
followed
by
a
bit
of
SQL
and
Then,
followed
ultimately
by
a
bit
of
web
programming,
which
is
to
say
that
students
in
this
class,
most
of
whom
have
never
programmed
before
ultimately
learn
a
little
something
new
about
all
of
those
languages
and
all
of
the
associated
Technologies.
A
B
Hi
everyone,
I'm
shuyan
I'm,
leading
the
co-pilot
product
team,
very
excited
to
be
on
the
same
stage
with
David
to
talk
about
the
co-pilot
in
the
classroom.
To
be
honest,
it's
been
over
20
years,
since
my
work
is
being
evaluated
by
professor
and
you
know
not
as
much
as
I'm
excited
I'm,
also
a
little
bit
nervous,
especially
knowing
David
I,
actually
banned
our
work
right.
B
The
essentially
you
can
install
copilot,
as
extension
for
your
favorite
code,
editor
like
vs
code,
Visual,
Studio,
Japanese,
family
of
IDE
and
new
Vim.
It
can
provide
a
real-time
suggestions
based
on
the
comment
and
the
code.
Significant
you
are
typing
the
it
is
the
work,
a
result
of
work
of
collaboration
between
Microsoft
GitHub
and
open
AI.
A
B
So
the
principal
building
block
of
co-pilot
is
the
openai
Codex
model
and
it
is
very
powerful
pre-trained
the
large
language
model.
Essentially
you
fit
into
the
training
data.
In
this
case,
it's
your
source
code
and
natural
language
text
and
the
model
teaches
itself
to
recognize
those
complex
patterns
like
programming,
syntax
and
language
semantics,
and
it
does
this
to
the
point
that
it
has
enough
knowledge,
so
we
could
start
making
predictions
itself.
So,
in
other
words,
you
give
it
a
partial
text
and
a
partial
code.
It
starts.
B
You
know
to
predict
output
that
fit
in
the
missing
pieces
and
blanks
with
learn
patterns.
On
top
of
what
codex
does.
Copilot
also
draws
contacts
from
your
surrounding
text
in
your
open
tab,
open
file
or
neighboring
tabs
and
send
them
to
the
model.
Along
with
the
code
comment,
you
are
typing,
so
the
model
predictions
the
model
makes
it
actually
much
more
relevant.
That
is
actually
the
secret
sauce
of
co-pilot
because
it
feels
very
personalized.
It
feels
and
understand
you
it's
made
just
for
you,
so.
A
Okay,
good,
and
so
when
it
comes
to
when
it
comes
to
actually
using
a
tool
like
this,
it
sounds
similar
to
a
feature
that
I
think
a
lot
of
us
who
use
Ides
are
familiar
with
like
intellisense,
where
I'm
already
getting
autocomplete
even
before
copilot
existed
yeah.
So
what
really
is
the
line
between
something
like
intellisense.
B
In
this
yeah,
so
absolutely
I
think
the
interaction
mode
of
copilot
is
very
similar
to
intellisense
and
other
Auto
completion
features,
but
it
actually
does
much
more
than
that.
So
the
wall
factor
that
we
heard
from
our
users
is
essentially
you
can
just
describe
a
problem
in
natural
language
comment.
A
B
The
co-pilot
actually
generates
the
entire
solutions
for
you
right,
so
it's
like
bringing
the
collective
Knowledge
from
developers
over
around
the
world
just
right
at
your
fingertips,
so
it
saves
you
time
to
go
to
the
internet,
to
search
for
Solutions
and
keep
developers
right
in
the
flow.
Besides
that
co-pilot,
we
can
also
generate
boilerplate
and
repetitive
code
for
you.
It's
actually
free
up
developers
from
focusing,
so
they
can
focus
on
the
high
value
tasks
like
complex
problem
solving.
That
is
actually
one
of
the
key
reasons
like
why
developers,
love
copilot.
A
B
A
And
I
do
think,
that's
persuasive,
certainly
in
the
context
of
productivity
and
writing
boilerplate
code
and
solving
problems
that
would
otherwise
just
take
more
time
than
it
needs
to.
But
I
feel
like
as
an
educator
on
behalf
of
the
audience
here
like
exactly
what
it
does.
Well
is
perhaps
exactly
what
someone
like
me
wouldn't
want
a
tool
to
do,
because
it
almost
sounds
like
an
overly
helpful
teacher,
an
overly
helpful
student,
an
overtly,
helpful,
classmate
and
so
I
know.
There's
been
chatter
about
this
in
the
Cs
education
Community
already
online
and
in
conferences
already.
B
So
we
definitely
hear
the
concern
from
the
learning
communities
that
co-pilot
could
impact,
especially
beginner
students,
learning
experiences
so
David,
as
you
know,
like
at
GitHub.
First
and
foremost,
we
are
a
developer
first
and
Community
Based
company,
so
deeply
engaging
and
contributing
back
to
the
community
has
always
been
our
top
priority.
So
co-pilot
as
a
very
novel
AI
assistant
too,
does
have
the
potential
to
change
how
software
is
being
developed
and
the
AI
generative
model
technology
that
we
use
to
power
copilot
is
also
in
its
infancy.
B
So,
as
such,
it
became
extremely
important
for
us
to
engage
with
our
learning
communities
early
on
and
listen
to
feedback.
So
one
of
the
first
steps
we
did
is
to
make
a
co-pilot
for
free
to
both
teachers
and
students,
and
our
hope
is
for
our
CS
Educators
like
you
to
develop
an
understanding
of
co-pilot
and
its
educational
potential
and
eventually
integrated
into
the
sales
curricular.
You
know
to
enhance
the
classroom,
experiences
for
both
teachers
and
students.
You
know
David
you
and
I.
We
discuss
right.
We
live
in
this
new
world
of
aicc
the
software
development.
B
A
A
B
A
A
That
I
probably
should
write
here,
perhaps
asking
more
inquisitively.
Perhaps
do
you
want
to
write
a
for
Loop
here,
or
are
you
sure
you
want
to
Nest
the
code
that
many
times
or
to
provide
a
sort
of
observant
eye
that
a
good
teacher
or
even
a
stronger
classmate
could
provide
without
just
actually
handing
the
solution
and
I?
A
Think
that's
how
we
think
about
it
right
now,
these
Technologies
co-pilot
and
others
are
sort
of
really
good
at
not
only
finishing
your
current
thought,
but
also
giving
you
indeed,
your
second
and
your
third
thought,
and
that's
too
helpful
right
now,
but
I
think
once
there's
API
hooks
or
just
configurations
where
you
can
throttle
it
back
and
forth
to
attenuate
it
for
your
particular
students.
Your
particularly
class,
your
particular
assignments,
I
mean
then
I
think
it
becomes
a
really
positive
uplift,
especially
for
larger
online
courses
too
yeah.
B
Absolutely
I
think
David.
We
have
heard
your
concern.
You
know
about
this
impact
so
as
the
technology
evolves
like
I,
we
firmly
believe
there
is
a
potential
for
co-pilot
to
become
the
teaching
assistant
for
students
for
beginner
student.
Another
concern,
like
we
have
been
hearing
from
the
community
is
about
plagiarism
like
because
of
the
code,
the
reused
nature,
you
know
what
constitute
a
plagiarism
has
always
been
a
debate
and
detective
plagiarism
is
even
harder
a
problem.
So
do
you
think
a
co-pilot
make
the
problem?
Worse.
No.
A
I,
don't
think
it
necessarily
makes
the
problem
worse.
I
think
the
reality
is
in
particularly
introductory
courses
like
ours.
We
do
look
for
patterns
of
dishonesty
as
by
cross-comparing
all
student
submissions,
maybe
even
looking
at
past
submissions
as
well
and
common
patterns
sort
of
pop
up,
and
so
insofar
as
co-pilot
is
presumably
going
to
generate
similar
patterns,
at
least
for
similar
problems.
I
think
we're
still
going
to
detect
that
kind
of
behavior
through
our
same
channels.
A
I
think
if
you've
only
heard
about
co-pilot,
thus
far
in
scene
and
heard
of
tools
like
it
I
think,
first
and
foremost
just
try
it
out
I
think
the
most
enlightening.
If
not
unsettling
thing,
we
did
a
few
weeks
back
thanks
to
a
colleague
of
mine,
was
run
co-pilot
on
cs50's,
first
homework
assignment,
which
or
second
homework
assignment
just
passed
hello
world
and
pretty
much
by
way
of
a
sequence
of
tab.
A
So
that's,
admittedly,
a
bit
anomalous,
but
we
then
fast
forwarded
toward
the
end
of
our
semester,
where
we
have
students
build
a
full-fledged
web
application
using
Python
and
more,
and
even
that
with
pretty
much
some
tab
completions.
Maybe
a
couple
of
English
comments
which
copilot
also
amazingly,
were
recently
able
to
answer
those
problems
as
well,
and
so
that
just
put
into
perspective
for
us
and
I
think
for
introductory
courses.
A
The
Cs
ones,
the
cs2s
of
the
world,
probably
a
little
too
soon
to
maybe
be
a
net
positive
in
those
early
weeks,
especially
but
for
higher
level
courses
where
students
already
have
some
mental
model
for
and
some
mechanics
of
programming
under
their
belt.
Then
I
think
it
does
genuinely
offer
an
opportunity
to
help
onboard
them
to
boilerplate
code
to
apis,
to
avoiding
having
to
troll
through
documentation
or
Worse
undocumented
code,
to
figure
out
common
usage
patterns.
So
that
I
think
is
a
win
and
I.