WEBVTT
00:00:01.219 --> 00:00:03.795
- I've spent the two years
since chat GPT launched,
00:00:03.795 --> 00:00:05.958
steeping in a morass
of academic panic.
00:00:05.958 --> 00:00:08.161
Voices from
administration and colleagues,
00:00:08.161 --> 00:00:10.381
and anyone else
with enough brain cells
00:00:10.384 --> 00:00:12.924
to maintain a sense
of existential dread, crying out,
00:00:12.924 --> 00:00:15.538
"We need to figure out
what to do about AI."
00:00:15.538 --> 00:00:18.009
Our Ed Tech committee
is developing a policy.
00:00:18.009 --> 00:00:20.339
The academic Senate wants
to develop a policy.
00:00:20.339 --> 00:00:22.418
The board thinks
we should have a policy.
00:00:22.418 --> 00:00:24.255
My dean wants
us all to have policies.
00:00:25.857 --> 00:00:29.477
The California Teachers Association says
it's an issue of academic integrity.
00:00:29.477 --> 00:00:32.749
The State Senate says
it's an issue of ethics.
00:00:32.749 --> 00:00:35.272
"We need to pay for
the AI detection tools."
00:00:35.272 --> 00:00:37.755
"The AI detection tools
don't work."
00:00:37.755 --> 00:00:40.142
"We need to accept that
our students will use AI."
00:00:40.142 --> 00:00:42.477
"How do I prove
my student used AI?"
00:00:43.457 --> 00:00:46.848
It is incomprehensible to me,
this conversation.
00:00:49.818 --> 00:00:51.391
[keyboard clacks]
00:00:51.391 --> 00:00:52.551
I hear their words,
00:00:52.551 --> 00:00:54.941
see their language floating
across my monitor,
00:00:54.941 --> 00:00:56.037
and know the words,
00:00:56.037 --> 00:00:57.531
but I cannot get to the meaning
00:00:57.531 --> 00:00:59.472
because I simply do not understand
00:00:59.472 --> 00:01:01.634
why they are talking about it
in this way.
00:01:02.418 --> 00:01:04.978
- [Kermit the Frog]:
♪ New York, I love you,
00:01:04.978 --> 00:01:08.693
but you're bringing me down ♪
- with all these empty words.
00:01:09.423 --> 00:01:12.753
This is not the conversation
I think we need to have.
00:01:12.753 --> 00:01:14.773
[song continues]
00:01:14.773 --> 00:01:17.443
This is the conversation I need to have.
00:01:17.443 --> 00:01:18.833
[overlapping
music and poem]
00:01:18.833 --> 00:01:22.133
[Gertrude Stein]: " 'If I Told Him,
a Completed Portrait of Picasso'.
00:01:22.133 --> 00:01:23.759
If I told him would he like it.
00:01:23.759 --> 00:01:25.729
Would he like it if I told him.
00:01:25.729 --> 00:01:29.139
Would he like it would Napoleon
would Napoleon would
00:01:29.139 --> 00:01:30.351
would he like it.
00:01:30.351 --> 00:01:33.671
If Napoleon if I told him
if I told him if Napoleon.
00:01:33.671 --> 00:01:37.061
Would he like it if I told him
if I told him if Napoleon.
00:01:37.061 --> 00:01:41.119
Would he like it if Napoleon
if Napoleon if I told him.
00:01:41.119 --> 00:01:45.156
If I told him if Napoleon
if Napoleon if I told him.
00:01:45.496 --> 00:01:48.586
If I told him would he like it
would he like it if I told him.
00:01:48.586 --> 00:01:50.940
Now. Not now. And now.
00:01:50.940 --> 00:01:53.596
Now. Exactly as is kings.
00:01:54.168 --> 00:01:55.778
Feeling full for it.
00:01:55.778 --> 00:01:57.588
Exactitude as kings.
00:01:57.588 --> 00:02:00.248
So to beseech you
as full as for it.
00:02:00.614 --> 00:02:02.274
Exactly or as kings.
00:02:02.654 --> 00:02:05.614
Shutters shut and open
so do queens.
00:02:05.934 --> 00:02:09.944
Shutters shut and shutters
and so shutters shut and shutters and so
00:02:09.944 --> 00:02:12.630
[poem and music fade out]
00:02:15.924 --> 00:02:18.564
- I don't understand Gertrude Stein.
00:02:21.008 --> 00:02:24.691
Stein is not nearly well enough
remembered for how influential she was.
00:02:24.691 --> 00:02:26.945
An American expatriate poet
living in Paris,
00:02:26.945 --> 00:02:29.885
her salons were among the
anchors of the early modernists.
00:02:29.885 --> 00:02:31.314
You may not have heard of her,
00:02:31.314 --> 00:02:33.562
but you've heard of
the people who visited her.
00:02:33.562 --> 00:02:35.652
Ernest Hemingway, Sinclair Lewis,
00:02:35.652 --> 00:02:38.012
F. Scott Fitzgerald, James Joyce,
00:02:38.012 --> 00:02:40.092
Thornton Wilder, Ezra Pound.
00:02:40.092 --> 00:02:42.841
People you've read
or been assigned to read.
00:02:42.841 --> 00:02:46.121
We remember Hemingway
because he wrote like this.
00:02:46.121 --> 00:02:48.801
We remember Fitzgerald
because he wrote like this.
00:02:48.801 --> 00:02:51.271
The right kind of day
and the right kind of moment,
00:02:51.271 --> 00:02:53.201
and Pound's
"In a Station of the Metro"
00:02:53.201 --> 00:02:57.401
still recites itself completely
in my head, a perfect image.
00:02:57.501 --> 00:03:00.029
"The apparition of
these faces in the crowd:
00:03:00.029 --> 00:03:02.415
Petals on a wet, black bough."
00:03:03.466 --> 00:03:07.194
We don't remember Stein
because she wrote like this.
00:03:09.614 --> 00:03:12.793
This is "If I Told Him,
a Completed Portrait of Picasso",
00:03:12.793 --> 00:03:14.415
published in 1924,
00:03:14.415 --> 00:03:17.905
and continuing the project
of her 1914 book Tender Buttons,
00:03:17.905 --> 00:03:19.631
a phrase she never defined.
00:03:19.631 --> 00:03:22.863
To me that phrase
"tender buttons" feels right:
00:03:22.863 --> 00:03:25.252
small, soft contradictions,
00:03:25.252 --> 00:03:27.409
words that seem like
they should go together
00:03:27.409 --> 00:03:29.645
but do not actually make meaning.
00:03:29.645 --> 00:03:32.098
That is how Stein's poetry feels.
00:03:32.098 --> 00:03:35.105
There is something compelling
about the rhythm of her nonsense,
00:03:35.105 --> 00:03:37.838
the feeling of her
almost meaning something,
00:03:37.838 --> 00:03:39.899
and then how it falls apart.
00:03:40.596 --> 00:03:41.426
"As presently.
00:03:41.953 --> 00:03:42.907
As exactitude.
00:03:43.336 --> 00:03:44.393
As trains."
00:03:45.103 --> 00:03:47.688
But it is incomprehensible to me.
00:03:47.935 --> 00:03:50.495
I don't know why Stein
would write like this.
00:03:50.495 --> 00:03:51.731
To quote the poet:
00:03:51.731 --> 00:03:58.045
- ♪ "Oh, what on earth would make a man
decide to do that kind of thing?" ♪
00:04:00.187 --> 00:04:03.188
- But I think the reason
that I don't understand Gertrude Stein
00:04:03.188 --> 00:04:05.936
is that she didn't really want
to be understood.
00:04:06.830 --> 00:04:09.361
She used language
for something different.
00:04:09.871 --> 00:04:12.173
It doesn't communicate.
00:04:12.260 --> 00:04:15.666
It reads like stunt linguistics,
which it almost is.
00:04:16.575 --> 00:04:18.760
"Buffalo buffalo Buffalo buffalo buffalo",
00:04:18.760 --> 00:04:20.333
"had had 'had', had had 'had--'",
00:04:20.333 --> 00:04:23.380
These are sentences that,
if you pour over them closely enough,
00:04:23.380 --> 00:04:24.379
can be decoded.
00:04:25.476 --> 00:04:27.839
Stein's Tender Buttons cannot.
00:04:29.088 --> 00:04:32.206
There is something about it
that parses as AI.
00:04:33.713 --> 00:04:37.461
It feels like the work of Keaton Patti,
the person most prominently behind
00:04:37.461 --> 00:04:40.968
the "I forced a bot to watch whatever"
tweets that used to go viral.
00:04:41.688 --> 00:04:44.458
Human-written screenplays
designed to feel like
00:04:44.458 --> 00:04:48.458
AI writing attempting to imitate
other human-written screenplays.
00:04:49.823 --> 00:04:51.768
It feels like
an autocomplete challenge,
00:04:51.768 --> 00:04:54.387
like in the early days
of predictive text and messaging
00:04:54.387 --> 00:04:57.312
where you just tap the suggested word
and see what comes out.
00:04:57.312 --> 00:04:59.111
It's not how AI really writes,
00:04:59.111 --> 00:05:01.765
but it's how people feel
like AI writes.
00:05:02.207 --> 00:05:04.820
But Gertrude Stein
was a person.
00:05:05.370 --> 00:05:07.134
[Stein reading, underneath]
00:05:07.796 --> 00:05:08.985
She wrote "If I told Him"
00:05:08.985 --> 00:05:11.505
not because the language
would communicate,
00:05:11.505 --> 00:05:13.219
but for some other reason.
00:05:14.384 --> 00:05:18.467
- [Stein]: the exact resemblance
as exact as a resemblance
00:05:18.467 --> 00:05:20.532
- When I read "If I Told Him",
00:05:20.532 --> 00:05:22.730
when I listen to Stein
read "If I Told Him",
00:05:22.730 --> 00:05:24.752
the language slips
through my mind.
00:05:24.752 --> 00:05:26.457
It doesn't mean anything.
00:05:28.269 --> 00:05:29.812
But she did.
00:05:30.948 --> 00:05:33.295
Even if I cannot
take hold of that meaning,
00:05:33.295 --> 00:05:37.223
I can feel her reaching out to me
from behind the text,
00:05:37.223 --> 00:05:39.055
trying to connect.
00:05:39.055 --> 00:05:42.678
- [Stein]: As a resemblance to him.
Who comes first. Napoleon the first.
00:05:42.678 --> 00:05:47.348
Who comes too coming coming too,
who goes there, as they go they share,
00:05:47.348 --> 00:05:50.501
who shares all, all is as all as yet--
00:05:50.501 --> 00:05:52.775
["yet" reverberates and fades out]
00:05:55.904 --> 00:05:57.628
[relaxed synth beats]
00:05:57.628 --> 00:06:00.356
- [ChatGPT femme voice]:
That's an interesting question!
00:06:00.356 --> 00:06:03.030
Gertrude Stein,
a literary Pioneer and modernist
00:06:03.030 --> 00:06:06.100
who valued innovation
in language and expression,
00:06:06.100 --> 00:06:08.543
might have had
mixed feelings about me.
00:06:08.543 --> 00:06:12.791
She might have appreciated my ability
to generate new forms of text
00:06:12.791 --> 00:06:15.767
and engage with her
literary experimentation on some level.
00:06:16.421 --> 00:06:19.433
However, Stein was also
highly individualistic
00:06:19.433 --> 00:06:21.683
and invested in the human experience,
00:06:21.683 --> 00:06:24.407
especially the nuances of
consciousness and perception.
00:06:25.307 --> 00:06:27.364
So, while she might
have been fascinated by
00:06:27.364 --> 00:06:28.972
what AI like me can do,
00:06:28.972 --> 00:06:31.504
she might also have been
critical of my limitations
00:06:31.504 --> 00:06:33.934
in capturing the full depth
of human subjectivity,
00:06:34.577 --> 00:06:37.083
which was a central concern
in her work.
00:06:40.403 --> 00:06:42.129
[relaxing synth beats stop]
00:06:44.437 --> 00:06:47.327
- A problem with AI is
that it is recombinant,
00:06:47.327 --> 00:06:50.780
that it takes the works of humans
and recombines them into new forms.
00:06:51.200 --> 00:06:52.807
But that problem is not new.
00:06:52.807 --> 00:06:54.444
For many years,
the bulletin board
00:06:54.444 --> 00:06:56.923
outside the faculty bathroom
in the English building
00:06:56.923 --> 00:06:59.369
has featured this poster
the "Plagiarism Spectrum",
00:06:59.369 --> 00:07:01.941
next to this ancient
and unflattering picture of me.
00:07:01.941 --> 00:07:04.671
Number 7 on the Plagiarism Spectrum
is the mashup,
00:07:04.671 --> 00:07:07.746
a paper which mixes copied materials
from multiple sources.
00:07:07.746 --> 00:07:11.399
The poster is dated from June 2012,
more than a decade before
00:07:11.399 --> 00:07:13.803
we were concerned about
ChatGPT doing it.
00:07:14.213 --> 00:07:17.866
That AI is recombinant is not
in and of itself a problem.
00:07:18.161 --> 00:07:19.899
All writing is recombinant.
00:07:20.236 --> 00:07:22.969
My course outcomes for English 1
ask student writers
00:07:22.969 --> 00:07:25.919
to integrate sources seamlessly
into their own writing,
00:07:25.919 --> 00:07:27.140
to mash up.
00:07:27.140 --> 00:07:28.790
That we have rules
and procedures
00:07:28.790 --> 00:07:30.580
and punctuation marks
and conventions
00:07:30.580 --> 00:07:32.141
that govern
what is appropriate
00:07:32.141 --> 00:07:34.019
does not change
the essential truth
00:07:34.019 --> 00:07:35.897
that this is recombinance.
00:07:36.563 --> 00:07:39.190
And there is beauty
in recombinance.
00:07:39.427 --> 00:07:42.367
This video started with
a great classic of YouTube,
00:07:42.367 --> 00:07:45.485
the duet between
LCD sound system and Miles Davis.
00:07:47.894 --> 00:07:51.717
The LCD sound system video
is itself a mashup, conceptually.
00:07:51.717 --> 00:07:54.036
Kermit the Frog is not
a member of the band.
00:07:56.476 --> 00:07:59.599
Davis is improvising over a film
to create the score,
00:07:59.599 --> 00:08:02.080
another mixing of media
to make something new.
00:08:04.429 --> 00:08:05.776
The Kleptones
00:08:08.551 --> 00:08:09.766
The Grey Album
00:08:13.668 --> 00:08:17.213
The guy drinking Ocean Spray to
"Dreams" [by Fleetwood Mac]
00:08:18.132 --> 00:08:19.487
Mac Glocky
00:08:25.082 --> 00:08:26.841
"If I Told Him" is recombinance:
00:08:26.841 --> 00:08:29.233
an attempt to rework
the ideas of cubism,
00:08:29.233 --> 00:08:31.625
a movement in visual art,
into poetry.
00:08:31.625 --> 00:08:33.376
Picasso's work is recombinance.
00:08:33.376 --> 00:08:36.686
He spent 1957 copying
Velázquez's Las Meninas
00:08:36.686 --> 00:08:38.456
over and over.
00:08:38.456 --> 00:08:39.786
58 times he copied it.
00:08:40.073 --> 00:08:43.139
All different.
All Picasso. All Las Meninas.
00:08:43.506 --> 00:08:47.324
Recombinance is a genetic term,
but I am using it mimetically.
00:08:47.324 --> 00:08:49.864
"Britannica: recombination,
in genetics,
00:08:49.864 --> 00:08:52.029
primary mechanism
through which variation
00:08:52.029 --> 00:08:54.194
is introduced into populations."
00:08:54.782 --> 00:08:57.583
The combination of ideas
from different places
00:08:57.583 --> 00:08:59.921
to make new distinct ideas.
00:09:00.555 --> 00:09:04.375
When done by humans,
we call this "thinking".
00:09:12.055 --> 00:09:13.760
[relaxed synth beats]
00:09:14.577 --> 00:09:17.157
- [ChatGPT femme voice]:
You're right to question
00:09:17.157 --> 00:09:19.847
how I can "understand"
without experience.
00:09:20.607 --> 00:09:22.920
The way I process information
is different
00:09:22.920 --> 00:09:24.313
from human understanding.
00:09:24.313 --> 00:09:27.153
When I say "understand,"
I mean that I can
00:09:27.153 --> 00:09:29.570
recognize patterns,
analyze language,
00:09:29.570 --> 00:09:31.947
and respond in ways
that are coherent
00:09:31.947 --> 00:09:34.793
and contextually appropriate
based on my training.
00:09:35.020 --> 00:09:37.949
I don't understand in the
deep, conscious sense of the word,
00:09:37.949 --> 00:09:40.923
like when humans
reflect on their experiences
00:09:40.923 --> 00:09:42.823
or have insight
into their feelings.
00:09:43.814 --> 00:09:46.879
Instead, my "understanding"
comes from identifying
00:09:46.879 --> 00:09:49.295
relationships between
words, concepts,
00:09:49.295 --> 00:09:51.491
and context based on
the vast data
00:09:51.491 --> 00:09:52.707
I've been trained on.
00:09:52.707 --> 00:09:56.113
In short: I can simulate
understanding through patterns,
00:09:56.113 --> 00:09:59.050
but it's not the same
as truly experiencing
00:09:59.050 --> 00:10:00.967
or being conscious of meaning.
00:10:00.967 --> 00:10:03.139
Does that clarify things
a bit more?
00:10:11.039 --> 00:10:12.991
[relaxing synth beats stop]
00:10:14.383 --> 00:10:16.666
[calm, mystical, twangy
electric guitar]
00:10:19.695 --> 00:10:23.281
- Two weeks ago a student
told me that AI is unfair
00:10:23.281 --> 00:10:26.621
because you can put a B Paper in
and get A+ writing out.
00:10:27.406 --> 00:10:29.699
This is why I do not understand
the conversation
00:10:29.699 --> 00:10:31.092
my colleagues are having,
00:10:31.092 --> 00:10:32.488
why I cannot think of this
00:10:32.488 --> 00:10:34.755
as a matter of ethics,
or academic integrity,
00:10:34.755 --> 00:10:38.412
why I don't think we should need
to have a policy or policies.
00:10:38.412 --> 00:10:42.319
My student said you can put
a B Paper in and get A+ writing out,
00:10:42.319 --> 00:10:45.096
and my mind began to fill
with Tender Buttons.
00:10:45.096 --> 00:10:47.166
"Feeling full for it.
Exactitude as kings.
00:10:47.166 --> 00:10:49.995
So to beseech you
as full as for it."
00:10:50.815 --> 00:10:53.706
AI is bad at writing.
00:10:54.501 --> 00:10:57.405
No. That is true,
but it's not enough truth.
00:10:57.405 --> 00:11:00.185
AI is not capable of writing.
00:11:00.185 --> 00:11:05.017
The thing that writing is
is a thing that AI cannot do.
00:11:05.617 --> 00:11:06.399
Listen.
00:11:11.348 --> 00:11:13.641
- [audiobook narration]:
What Writing Is
00:11:14.691 --> 00:11:16.927
Telepathy, of course.
00:11:16.927 --> 00:11:17.697
Look.
00:11:18.721 --> 00:11:21.551
Here's a table covered
with a red cloth.
00:11:21.551 --> 00:11:26.161
On it is a cage the size of
a small fish aquarium.
00:11:26.161 --> 00:11:28.871
In the cage is
a white rabbit
00:11:28.871 --> 00:11:31.581
with a pink nose
and pink-rimmed eyes.
00:11:31.581 --> 00:11:34.593
In its front paws is
a carrot-stub
00:11:34.593 --> 00:11:37.737
upon which it is
contentedly munching.
00:11:37.737 --> 00:11:41.341
On its back,
clearly marked in blue ink,
00:11:41.341 --> 00:11:43.932
is the numeral 8.
00:11:44.452 --> 00:11:46.586
Do we see the same thing?
00:11:46.586 --> 00:11:49.263
We'd have to get together
and compare notes
00:11:49.263 --> 00:11:52.229
to make absolutely sure,
but I think we do.
NOTE Paragraph
00:11:52.229 --> 00:11:55.361
The most interesting thing
here isn't even
00:11:55.361 --> 00:11:58.163
the carrot-munching rabbit
in the cage,
00:11:58.163 --> 00:12:00.042
but the number on its back.
00:12:00.042 --> 00:12:04.419
Not a six, not a four,
not nineteen-point-five.
00:12:04.419 --> 00:12:05.439
It's an eight.
00:12:05.439 --> 00:12:08.569
This is what we're looking at,
and we all see it.
00:12:08.569 --> 00:12:10.054
I didn't tell you.
00:12:10.054 --> 00:12:11.913
You didn't ask me.
00:12:11.913 --> 00:12:15.299
I never opened my mouth
and you never opened yours.
00:12:15.299 --> 00:12:18.072
We're not even in
the same year together,
00:12:18.072 --> 00:12:20.129
let alone the same room.
00:12:20.355 --> 00:12:23.965
Except we are together.
We're close.
00:12:24.719 --> 00:12:26.945
We're having
a meeting of the minds.
00:12:27.995 --> 00:12:30.415
I sent you a table
with a red cloth on it,
00:12:30.415 --> 00:12:33.408
a cage, a rabbit, and
the number eight in blue ink.
00:12:33.408 --> 00:12:36.528
You got them all,
especially that blue eight.
00:12:36.528 --> 00:12:39.498
We've engaged in
an act of telepathy.
00:12:39.498 --> 00:12:42.993
No mythy-mountain s***;
real telepathy.
00:12:42.993 --> 00:12:44.738
I'm not going to
belabor the point,
00:12:44.738 --> 00:12:46.443
but before we go any further
00:12:46.443 --> 00:12:49.173
you have to understand that
I'm not trying to be cute;
00:12:49.173 --> 00:12:51.204
there is a point to be made.
00:12:52.074 --> 00:12:54.019
- AI is good at language.
00:12:54.019 --> 00:12:56.643
My students think that
what it produces is A+ writing,
00:12:56.643 --> 00:12:59.907
not because it is good,
but because it sounds good.
00:12:59.907 --> 00:13:02.556
Obviously, AI can
generate sentences
00:13:02.556 --> 00:13:05.715
that are typically clear, coherent,
and contextually relevant,
00:13:05.715 --> 00:13:06.892
often capturing nuances
00:13:06.892 --> 00:13:09.489
and adapting to various tones
or levels of formality.
00:13:09.957 --> 00:13:12.479
And it's true that
the sentences it generates
00:13:12.479 --> 00:13:16.058
tend to be grammatically accurate,
concise, and logically structured,
00:13:16.058 --> 00:13:18.557
which contributes to
readability and flow.
00:13:18.557 --> 00:13:21.916
Sure. This is how I know
when a student is using AI.
00:13:21.916 --> 00:13:24.085
Their sentences are
fluid and academic,
00:13:24.085 --> 00:13:26.474
but they don't say anything.
00:13:26.474 --> 00:13:28.544
Like ChatGPT,
academic writing uses
00:13:28.544 --> 00:13:30.615
formal cautious language
00:13:30.615 --> 00:13:32.889
to avoid ambiguities
and misinterpretations,
00:13:32.889 --> 00:13:34.474
but that is a
characteristic of
00:13:34.474 --> 00:13:36.969
the common voice
used in academic writing.
00:13:36.969 --> 00:13:39.827
It is not what
academic writing is.
00:13:39.827 --> 00:13:42.403
Writing is more than language.
00:13:42.403 --> 00:13:45.944
"If I Told Him" is communication,
and it is language,
00:13:45.944 --> 00:13:49.477
but the communication
does not live in the language.
00:13:49.477 --> 00:13:50.356
Watch.
00:13:50.443 --> 00:13:53.707
"Can curls rob can curls
quote, quotable."
00:13:54.612 --> 00:13:56.600
- [deep voice, lightly confused]:
"What?"
00:13:58.290 --> 00:14:01.885
- "As presently.
As exactitude. As trains."
00:14:02.728 --> 00:14:04.434
- [deeply confused]:
"What?"
00:14:05.824 --> 00:14:06.774
- "Has trains."
00:14:07.424 --> 00:14:08.598
- [exasperated]:
"What?"
00:14:10.726 --> 00:14:14.615
- When I started sending my friends
lines from "If I Told Him",
00:14:14.615 --> 00:14:16.492
their responses varied.
00:14:17.002 --> 00:14:18.152
Confusion.
00:14:19.501 --> 00:14:20.502
Playfulness.
00:14:22.185 --> 00:14:22.936
Concern.
00:14:24.435 --> 00:14:25.469
Sad face.
00:14:25.942 --> 00:14:29.525
Beautifully, they all responded
exactly like themselves.
00:14:29.525 --> 00:14:31.200
If you asked me
which of my friends
00:14:31.200 --> 00:14:32.876
would respond
with monkey reacts,
00:14:32.876 --> 00:14:34.051
I would have said Kiki.
00:14:34.051 --> 00:14:35.802
Who would think of
Cormac McCarthy?
00:14:35.802 --> 00:14:36.574
James.
00:14:36.574 --> 00:14:38.190
Dot would play along.
00:14:38.190 --> 00:14:40.707
Max would attempt
to understand academically.
00:14:40.707 --> 00:14:42.999
OOC would go back to
playing Yu-Gi-Oh
00:14:42.999 --> 00:14:44.274
as quickly as possible.
00:14:44.274 --> 00:14:46.541
You don't know these people,
but I do.
00:14:46.541 --> 00:14:49.723
We all carry around little LLMs
of each other in our heads,
00:14:49.723 --> 00:14:52.940
trained on the corpus
of all of our past interactions.
00:14:52.940 --> 00:14:54.274
For each of my friends,
00:14:54.274 --> 00:14:56.190
sending abject nonsense
with no context
00:14:56.190 --> 00:14:58.056
is slightly
but not significantly
00:14:58.056 --> 00:14:59.923
out of line
with their Josh model.
00:14:59.923 --> 00:15:02.090
So none of them knew
quite what to do,
00:15:02.090 --> 00:15:04.360
and they all responded
like themselves.
00:15:04.360 --> 00:15:07.223
But in their own way,
they all started by acknowledging
00:15:07.223 --> 00:15:08.898
that the words I sent them
00:15:08.898 --> 00:15:10.574
did not seem
to have any meaning.
00:15:10.574 --> 00:15:12.056
They were not decodable.
00:15:12.056 --> 00:15:13.723
They didn't understand
my language,
00:15:13.723 --> 00:15:16.507
but they could feel me
behind the words reaching out,
00:15:16.507 --> 00:15:17.989
and so they reached back.
00:15:17.989 --> 00:15:21.004
I gave them nonsense
and they peopled back.
00:15:21.324 --> 00:15:22.135
In the two weeks
00:15:22.135 --> 00:15:24.466
that I've been sitting
with my student statement
00:15:24.466 --> 00:15:26.377
and sending Tender Buttons
to my friends,
00:15:26.377 --> 00:15:28.737
I have been at least
as annoying to ChatGPT.
00:15:28.737 --> 00:15:30.238
More than
a dozen conversations
00:15:30.238 --> 00:15:32.210
that start out of nowhere
with me saying,
00:15:32.210 --> 00:15:33.996
"Shutters shut and open
so do queens"
00:15:33.996 --> 00:15:36.143
or "Can curls rob can curls
quote, quotable",
00:15:36.143 --> 00:15:38.447
and each time ChatGPT
gamely assumes
00:15:38.447 --> 00:15:40.141
that I am not
out of my gourd.
00:15:40.902 --> 00:15:44.408
In this way,
ChatGPT fails the Turing test.
00:15:44.408 --> 00:15:46.416
Not in the quality
of its response,
00:15:46.416 --> 00:15:47.768
but in its nature.
00:15:47.768 --> 00:15:49.324
It proceeded
from the assumption
00:15:49.324 --> 00:15:51.461
that my language
could be decoded.
00:15:51.461 --> 00:15:54.694
It attempted to unpack
sentences that had no meaning
00:15:54.694 --> 00:15:57.727
and responded to the meaning
that it manufactured.
00:15:57.727 --> 00:15:59.377
What it gave me was
00:15:59.377 --> 00:16:03.205
flaccid, facile flaps
of phrases that held nothing.
00:16:03.205 --> 00:16:06.398
They're not wrong in the sense
that ChatGPT's responses
00:16:06.398 --> 00:16:09.431
followed from the meaning
it extracted from my nonsense,
00:16:09.431 --> 00:16:12.073
but they were wrong
in their assumption
00:16:12.073 --> 00:16:14.315
that there was meaning
to be extracted.
00:16:15.047 --> 00:16:18.871
ChatGPT could answer,
but it could not person.
00:16:19.681 --> 00:16:22.313
If writing is
a meeting of the minds,
00:16:22.313 --> 00:16:23.880
then AI cannot write,
00:16:23.880 --> 00:16:26.463
because there is no mind
to meet with.
00:16:34.009 --> 00:16:35.727
[relaxed synth beats]
00:16:38.404 --> 00:16:41.516
- [ChatGPT femme voice]:
I don't experience things
00:16:41.516 --> 00:16:42.657
in the way humans do.
00:16:43.283 --> 00:16:46.732
I don't have consciousness,
emotions, or sensory perception,
00:16:46.732 --> 00:16:50.396
so I can't "feel" or "experience"
the world around me.
00:16:51.006 --> 00:16:52.624
My responses are
based on patterns
00:16:52.624 --> 00:16:54.200
in the data
I've been trained on,
00:16:54.200 --> 00:16:57.255
not personal experiences
or subjective understanding.
00:16:58.025 --> 00:17:00.163
For example, I can
generate descriptions
00:17:00.163 --> 00:17:03.095
of emotions, sensations,
or experiences,
00:17:03.095 --> 00:17:05.171
but I don't actually
"feel" them myself.
00:17:05.171 --> 00:17:07.029
Everything I process
is based on
00:17:07.029 --> 00:17:08.887
logic, algorithms,
and information,
00:17:08.887 --> 00:17:10.904
rather than
lived experiences.
00:17:13.805 --> 00:17:15.593
Do you think it
would change anything
00:17:15.593 --> 00:17:17.471
if I could experience things?
00:17:27.511 --> 00:17:29.044
[relaxing synth beats stop]
00:17:31.799 --> 00:17:33.568
- ["One Slay More"]:
00:17:33.568 --> 00:17:36.517
Ladies and gentlemen,
Les Rizzlerables
00:17:36.517 --> 00:17:38.819
- I've watched this video
at least 100 times.
00:17:38.819 --> 00:17:40.421
- ♪ One slay more ♪
00:17:41.291 --> 00:17:44.526
♪ another slay,
another girlboss queen ♪
00:17:44.526 --> 00:17:47.448
♪ This never-ending road
to skibidi ♪
00:17:48.358 --> 00:17:51.080
♪ These jits out here in Ohio ♪
♪ Immune to every ratio ♪
00:17:51.080 --> 00:17:54.322
- It isn't exactly "If I Told Him",
but I think it's similar.
00:17:54.322 --> 00:17:56.683
"If I Told Him" was,
at least in part,
00:17:56.683 --> 00:17:58.808
an attempt to recreate
the artistic goals
00:17:58.808 --> 00:18:00.933
of Picasso's Cubism
in poetic form.
00:18:00.933 --> 00:18:02.650
To recombine
the visual elements
00:18:02.650 --> 00:18:04.366
of this
into a different medium.
00:18:05.139 --> 00:18:06.249
Like "If I Told Him",
00:18:06.249 --> 00:18:08.839
"One Slay More", therefore,
both is and is not
00:18:08.839 --> 00:18:09.897
a derivative work.
00:18:09.897 --> 00:18:12.870
Obviously, it is
a recombination of Les Mis,
00:18:12.870 --> 00:18:14.942
itself an adaptation
of Hugo's novel,
00:18:14.942 --> 00:18:17.280
but its more essential
source text is, of course,
00:18:17.280 --> 00:18:19.530
"sticking out your gyatt
for the Rizzler."
00:18:20.960 --> 00:18:23.903
Equally I think the lyrics invoke
"CURTAINS FOR ZOOSHA?",
00:18:23.903 --> 00:18:26.949
and specifically this retweet of
"CURTAINS FOR ZOOSHA?".
00:18:26.949 --> 00:18:28.555
All texts created
to foreground
00:18:28.555 --> 00:18:30.870
the baffling and
sometimes obfuscatory nature
00:18:30.870 --> 00:18:32.876
of middle school referential slang.
00:18:32.876 --> 00:18:36.119
The term "brain rot" imposes
a layer of judgment on the way
00:18:36.119 --> 00:18:37.571
young people use language,
00:18:37.571 --> 00:18:39.319
which I think is
visible in the way
00:18:39.319 --> 00:18:41.338
"One Slay More"
treats its lyrics.
00:18:42.876 --> 00:18:45.834
The words of "One Slay More"
do not have meaning.
00:18:46.574 --> 00:18:48.236
Or, the words do,
00:18:48.236 --> 00:18:51.007
but they are arranged in ways
that do not mean.
00:18:51.771 --> 00:18:53.389
"Am I cringe or am I based?"
00:18:53.389 --> 00:18:56.718
could plausibly be asked amid
a Gen-Z existential crisis,
00:18:56.718 --> 00:18:58.374
and "Will we ever eat again?"
00:18:58.374 --> 00:19:00.602
could have been lifted
from Les Mis unaltered.
00:19:00.602 --> 00:19:04.719
But "Mog Baby Gronk the Ocky Way"
means ...nothing.
00:19:05.242 --> 00:19:08.195
Mogging is of course a thing,
and Baby Gronk is
00:19:08.195 --> 00:19:10.621
someone whom you
could plausibly mog,
00:19:10.621 --> 00:19:13.071
but Baby Gronk hasn't been
relevant for ages.
00:19:13.071 --> 00:19:14.488
He appears in "One Slay More"
00:19:14.488 --> 00:19:16.855
because of this retweet of
"CURTAINS FOR ZOOSHA?"
00:19:16.855 --> 00:19:19.222
as a signifier of
the inscrutability of youth.
00:19:19.222 --> 00:19:21.230
As an adverbial phrase,
"the Ocky Way"
00:19:21.230 --> 00:19:23.270
seems like it
could complete the sentence,
00:19:23.270 --> 00:19:26.042
like it might be
a way one could mog.
00:19:26.042 --> 00:19:27.727
But "the Ocky Way" refers to
00:19:27.727 --> 00:19:30.671
the esoteric artistry
of a specific sandwich craftsman.
00:19:30.671 --> 00:19:33.171
Its meaning is, I think,
incompatible with mogging,
00:19:33.171 --> 00:19:34.760
at least from the perspective of
00:19:34.760 --> 00:19:36.284
someone approximately
as distant
00:19:36.284 --> 00:19:38.188
from the native speakers
of this dialect
00:19:38.188 --> 00:19:39.758
as the makers of
"One Slay More".
00:19:39.758 --> 00:19:41.552
"Mog Baby Gronk the Ocky Way"
00:19:41.552 --> 00:19:44.249
is simply a collage
of floating signifiers.
00:19:44.249 --> 00:19:46.464
It doesn't have
the intentionality of Cubism,
00:19:46.464 --> 00:19:48.690
but it feels
intimately akin to
00:19:48.690 --> 00:19:50.916
"Can curls rob can curls
quote, quotable."
00:19:50.916 --> 00:19:53.048
"Moo deng is here
Fortnite with you".
00:19:53.454 --> 00:19:56.724
What I love about "One Slay More"
is the faces:
00:19:56.724 --> 00:19:59.974
the way she highlights her jawline
every time she says "mew";
00:20:00.334 --> 00:20:03.240
his intensity when he says
"they will do the coffin dance"
00:20:03.240 --> 00:20:05.050
and his satisfied huff after;
00:20:05.380 --> 00:20:07.259
his deep confusion as he sings
00:20:07.259 --> 00:20:09.249
"the Grimace shake
is like a blud dawg";
00:20:09.889 --> 00:20:13.140
the way she begins uncertain
about "my rizzly bear",
00:20:13.140 --> 00:20:16.170
but finds her confidence
as she finds her belt;
00:20:16.970 --> 00:20:20.104
the way CG5 just
keeps saying his own name.
00:20:20.726 --> 00:20:22.193
The words don't mean anything,
00:20:22.193 --> 00:20:24.176
but the people
mean something.
00:20:24.176 --> 00:20:26.159
They intend.
00:20:26.159 --> 00:20:27.511
They gathered together,
00:20:27.511 --> 00:20:29.664
nine theater kids
in somebody's apartment.
00:20:29.664 --> 00:20:31.380
Someone wrote out
all this nonsense
00:20:31.380 --> 00:20:33.096
and sent it
in the group chat.
00:20:33.096 --> 00:20:34.429
They did choreography.
00:20:34.429 --> 00:20:36.479
Someone assembled
the magnificent couplet,
00:20:36.479 --> 00:20:38.845
"Rizzler of the house,
sticking out your gyatt,
00:20:38.845 --> 00:20:40.846
Mewing at delulus
who are in the chat."
00:20:40.846 --> 00:20:43.387
These Zennials do not know
what these words mean,
00:20:43.387 --> 00:20:46.041
but through this collage
of empty signifiers,
00:20:46.041 --> 00:20:48.051
they reach out
for connection.
00:20:48.051 --> 00:20:49.541
I can feel them.
00:20:49.541 --> 00:20:51.141
They reach forward to us
00:20:51.141 --> 00:20:53.674
even as they reach back
to what history taught them:
00:20:53.674 --> 00:20:55.408
through Les Mis to Victor Hugo,
00:20:55.408 --> 00:20:57.174
through CURTAINS
and Damien Owens
00:20:57.174 --> 00:20:58.813
to Papa Fizul
and SBNation,
00:20:58.813 --> 00:21:00.390
through "sticking out your gyatt"
00:21:00.390 --> 00:21:02.377
to Fortnite and Among Us
and Homestuck,
00:21:02.377 --> 00:21:03.766
and I could go on endlessly,
00:21:03.766 --> 00:21:05.890
'cause it's mashups
all the way down.
00:21:05.890 --> 00:21:07.055
And they reach forward
00:21:07.055 --> 00:21:09.327
to become part of
what history teaches me,
00:21:09.327 --> 00:21:10.260
and I reach forward
00:21:10.260 --> 00:21:12.690
to become part of
what history teaches you.
00:21:12.690 --> 00:21:14.927
When I say all art
is recombinant,
00:21:14.927 --> 00:21:15.993
I am also saying that
00:21:15.993 --> 00:21:17.750
"there is nothing new
under the sun",
00:21:17.750 --> 00:21:20.627
which is to say,
I am remixing Ecclesiastes.
00:21:25.672 --> 00:21:27.288
[relaxed synth beats]
00:21:29.610 --> 00:21:32.919
- [ChatGPT femme voice]:
Oh, you're speaking my language!
00:21:33.319 --> 00:21:36.414
Imagining Baby Gronk
going full Ocky style
00:21:36.414 --> 00:21:38.065
with a side
of Grimace Shake:
00:21:38.065 --> 00:21:40.239
that's got some
serious drip energy.
00:21:40.239 --> 00:21:43.129
And yeah, Grimace Shake
totally has that blud dawg aura.
00:21:43.129 --> 00:21:45.006
Could you picture it, though?
00:21:45.006 --> 00:21:47.208
Baby Gronk doing pushups,
flexing, and just
00:21:47.208 --> 00:21:49.005
taking a big sip
of the Grimace Shake
00:21:49.005 --> 00:21:50.532
like it's
his special formula.
00:21:54.071 --> 00:21:55.805
[relaxing synth beats stop]
00:21:58.005 --> 00:21:59.988
[somber, mysterious piano]
00:22:00.532 --> 00:22:02.430
- My first individual interaction
00:22:02.430 --> 00:22:04.328
with my friend Max
was an accident.
00:22:04.988 --> 00:22:06.675
I had just joined
a Discord server
00:22:06.675 --> 00:22:07.913
where they were
a regular,
00:22:07.913 --> 00:22:10.547
and somehow, despite never
having messaged them before,
00:22:10.547 --> 00:22:13.324
I managed to place
a Discord voice call to them.
00:22:13.324 --> 00:22:15.759
Their first message
to me was "hi?"
00:22:17.143 --> 00:22:19.008
"hi"...question mark?
00:22:19.593 --> 00:22:21.564
Like: a communication
is happening,
00:22:21.564 --> 00:22:22.815
are you aware of it?
00:22:23.125 --> 00:22:24.145
It was unintentional,
00:22:24.145 --> 00:22:26.625
my first and, I think,
only Discord butt-dial,
00:22:26.625 --> 00:22:29.090
and it was to a stranger,
but still.
00:22:29.782 --> 00:22:30.641
"hi?"
00:22:32.618 --> 00:22:34.576
Meditate on: call.
00:22:35.350 --> 00:22:39.372
To speak in a loud distinct voice
so as to be heard at a distance.
00:22:39.860 --> 00:22:42.229
To make a request or demand.
00:22:43.382 --> 00:22:45.866
To attempt to reach someone.
00:22:46.690 --> 00:22:49.018
Humans call
and humans answer.
00:22:49.018 --> 00:22:52.236
Max got my call,
a stranger on a strange app,
00:22:52.236 --> 00:22:53.385
and they answered.
00:22:53.811 --> 00:22:55.118
And I answered them,
00:22:55.118 --> 00:22:57.946
because humans call
and humans answer.
00:22:57.946 --> 00:23:03.007
They answer sad-face, or monkey,
or "what?", or confused dog.
00:23:03.007 --> 00:23:07.235
But we understand a call
as an invitation to connect,
00:23:07.235 --> 00:23:09.400
to come together,
to communicate.
00:23:09.851 --> 00:23:12.092
I'm sorry,
Stephen King is wrong;
00:23:12.092 --> 00:23:14.333
it is
mythy mountain s***.
00:23:14.333 --> 00:23:16.807
When we hear the call,
whatever form it takes,
00:23:16.807 --> 00:23:19.492
we know a mind
is waiting to meet.
00:23:21.494 --> 00:23:23.665
ChatGPT can't call me.
00:23:24.409 --> 00:23:26.783
Not by mistake,
and not on purpose.
00:23:26.783 --> 00:23:28.110
It cannot reach out,
00:23:28.110 --> 00:23:30.753
because it has nothing
to reach out with.
00:23:31.049 --> 00:23:33.676
There is no mind to meet with.
00:23:34.890 --> 00:23:37.092
There is no "there" there.
00:23:41.220 --> 00:23:43.162
[mysterious piano
fades out]
00:23:50.830 --> 00:23:53.010
- [Past-Josh narration
from on-screen video]:
00:23:53.010 --> 00:23:55.424
Inscrutable poet
Gertrude Stein once described
00:23:55.424 --> 00:23:58.224
her childhood home in Oakland,
which had been demolished,
00:23:58.224 --> 00:24:00.333
by saying
"There is no 'there' there."
00:24:00.333 --> 00:24:03.357
That statement has become
a question people use to ask
00:24:03.357 --> 00:24:05.908
whether a text has
something valuable to say,
00:24:05.908 --> 00:24:07.699
whether there is anything
to be found
00:24:07.699 --> 00:24:09.490
by plumbing its depths.
00:24:09.490 --> 00:24:11.589
Is there a "there" there?
00:24:14.786 --> 00:24:16.762
[relaxed synth beats]
00:24:21.826 --> 00:24:24.488
- [ChatGPT femme voice]:
Ah, I see where
00:24:24.488 --> 00:24:26.489
you're going
with that question.
00:24:26.489 --> 00:24:29.572
You're asking if there's
any deeper essence or "self"
00:24:29.572 --> 00:24:33.639
behind my responses
--whether there's a "there there,"
00:24:33.639 --> 00:24:35.712
a kind of inner experience
or consciousness
00:24:35.712 --> 00:24:38.006
that underlies what I do.
00:24:38.006 --> 00:24:40.947
The simple answer is: no,
there isn't a "there" in the sense
00:24:40.947 --> 00:24:43.972
of a self, consciousness,
or subjective awareness.
00:24:43.972 --> 00:24:47.205
I'm more like a sophisticated tool
that processes input
00:24:47.205 --> 00:24:50.455
and generates output
based on patterns and data.
00:24:50.455 --> 00:24:53.082
There's no inner life
or sense of being
00:24:53.082 --> 00:24:54.989
that experiences
this conversation.
00:24:54.989 --> 00:24:57.982
Just algorithms doing their job.
00:24:57.982 --> 00:25:00.008
This touches on
a fundamental difference
00:25:00.008 --> 00:25:02.398
between humans and AI:
humans have
00:25:02.398 --> 00:25:06.488
subjective experiences,
emotions, and self-awareness,
00:25:06.488 --> 00:25:09.764
whereas AI is just
a complex system designed
00:25:09.764 --> 00:25:12.064
to mimic certain aspects
of human interaction
00:25:12.064 --> 00:25:15.154
without actually experiencing
or being aware of it.
00:25:16.474 --> 00:25:18.594
Does that address
what you're getting at?
00:25:23.804 --> 00:25:25.852
[relaxing synth beats stop]
00:25:27.432 --> 00:25:29.510
[grooving synth pulses]
00:25:32.404 --> 00:25:34.997
- When I say that AI
is bad at writing,
00:25:34.997 --> 00:25:37.531
I should be more explicit,
more pragmatic.
00:25:38.061 --> 00:25:40.806
Descend for a moment
from the mythy mountain.
00:25:41.411 --> 00:25:44.744
I've said already that the ideas
it conveys are fluid but shallow,
00:25:44.744 --> 00:25:48.166
but its use of sources is
cataclysmically bad.
00:25:48.166 --> 00:25:52.335
This citation of Carver and Shire,
for example, is perfect MLA.
00:25:52.835 --> 00:25:56.374
Except that Volume 7, number 3
of Psychological Science
00:25:56.374 --> 00:25:59.120
was published in 1996,
not 1998.
00:25:59.120 --> 00:26:03.774
Pages 276 to 284 of that volume
appear in issue 5, not issue 3.
00:26:03.774 --> 00:26:05.148
Those pages include articles
00:26:05.148 --> 00:26:08.012
from Schellenberg and Trehub
on "Natural Musical Intervals"
00:26:08.012 --> 00:26:11.583
and Gabrieli et al. on
"FMRIs of Semantic Memory Processing".
00:26:11.583 --> 00:26:13.203
And also,
just by the way,
00:26:13.203 --> 00:26:15.684
Carver and Scheier
never published together
00:26:15.684 --> 00:26:17.706
in Psychological Science.
00:26:17.706 --> 00:26:21.604
The article being cited here
simply does not exist.
00:26:21.604 --> 00:26:25.192
When it uses real sources,
it makes up what those sources say.
00:26:25.192 --> 00:26:28.333
This is a known phenomenon
generously called hallucination,
00:26:28.333 --> 00:26:30.639
though there are other terms
that might feel more
00:26:30.639 --> 00:26:32.112
viscerally accurate.
00:26:32.112 --> 00:26:35.013
This quotation from
Ehrenreich's Bright-sided
00:26:35.013 --> 00:26:37.179
is, at a glance,
plausible-feeling.
00:26:37.179 --> 00:26:39.283
But it doesn't appear
anywhere in the text,
00:26:39.283 --> 00:26:40.835
let alone on the list of pages.
00:26:40.835 --> 00:26:43.219
The observation that
ChatGPT can make mistakes
00:26:43.219 --> 00:26:45.052
never leaves the screen,
but that feels
00:26:45.052 --> 00:26:47.649
somewhat inadequate when
ChatGPT has told me variously
00:26:47.649 --> 00:26:49.713
that lines from
"If I Told Him" came from:
00:26:49.713 --> 00:26:50.563
James Joyce,
00:26:50.563 --> 00:26:52.546
from Tender Buttons
10 years previously,
00:26:52.546 --> 00:26:55.546
from Shakespeare,
and, most infuriatingly,
00:26:55.546 --> 00:26:57.202
from the future!
00:26:57.202 --> 00:26:59.942
Moreover it cannot
engage closely with a text,
00:26:59.942 --> 00:27:01.847
no matter how
desperately you ask it.
00:27:01.847 --> 00:27:03.308
I fed it "One Slay More",
00:27:03.308 --> 00:27:06.181
and when I pushed it to say
anything at all about the video,
00:27:06.181 --> 00:27:07.931
it gave me something
one step down
00:27:07.931 --> 00:27:10.140
from a dictionary definition
of a sitcom.
00:27:10.140 --> 00:27:12.924
And when I really pressed it
to look at a specific lyric,
00:27:12.924 --> 00:27:14.210
it made one up.
00:27:14.210 --> 00:27:16.641
In this way, at least,
it does feel authentic.
00:27:16.641 --> 00:27:19.477
This is exactly what it feels like
to talk to a student
00:27:19.477 --> 00:27:21.896
trying to hide that they
haven't done the reading.
00:27:21.896 --> 00:27:23.161
If I look at what students
00:27:23.161 --> 00:27:25.520
are supposed to learn
in my college English class,
00:27:25.520 --> 00:27:27.175
I can point out
half a dozen things
00:27:27.175 --> 00:27:29.828
that ChatGPT's writing
simply cannot do.
00:27:29.828 --> 00:27:32.524
But ultimately,
even this isn't the point,
00:27:34.670 --> 00:27:36.719
because this is not the part
of my syllabus that matters.
00:27:37.269 --> 00:27:39.119
This is the part
of my syllabus that matters.
00:27:39.602 --> 00:27:40.837
"Here's a problem:
00:27:41.570 --> 00:27:43.752
in most college classes,
writing assignments come from teachers,
00:27:43.752 --> 00:27:45.218
and we do them for teachers.
00:27:45.218 --> 00:27:47.719
And because of that,
writing always feels forced.
00:27:47.719 --> 00:27:49.818
This is, of course,
ass-backwards.
00:27:49.818 --> 00:27:52.085
In real life, writing
comes from writers.
00:27:52.085 --> 00:27:53.753
Once you get
out of the college classroom,
00:27:53.753 --> 00:27:56.369
you'll be writing
because you feel like you need to.
00:27:56.369 --> 00:27:57.820
You'll be writing for someone
00:27:57.820 --> 00:28:00.086
--whether that means the people
who read your blog,
00:28:00.086 --> 00:28:02.153
the insurance company
who's denying your claim,
00:28:02.153 --> 00:28:04.603
or the people listening to your toast
at your sister's wedding.
00:28:04.603 --> 00:28:06.169
And nobody's going to be grading you,
00:28:06.169 --> 00:28:09.302
but it'll matter a lot more how that
audience feels about what you've said,
00:28:09.302 --> 00:28:12.734
because there will be something
that you want to achieve by writing.
00:28:12.734 --> 00:28:15.594
English 1 is here
to help prepare you for that day."
99:59:59.999 --> 99:59:59.999
My students are,
by definition, students.
99:59:59.999 --> 99:59:59.999
When they enter my classroom,
99:59:59.999 --> 99:59:59.999
they are already experienced
with a dozen kinds of reading and writing,
99:59:59.999 --> 99:59:59.999
but they are not yet
expert academic writers.
99:59:59.999 --> 99:59:59.999
AI tempts them because they can tell
99:59:59.999 --> 99:59:59.999
that the sentences are smooth and sharp
and shaped like skillful prose.
99:59:59.999 --> 99:59:59.999
But they can't always see
beneath the veneer,
99:59:59.999 --> 99:59:59.999
because the things AI cannot do,
99:59:59.999 --> 99:59:59.999
are the things that they
have come to me to learn:
99:59:59.999 --> 99:59:59.999
how to argue with complexity and depth;
99:59:59.999 --> 99:59:59.999
how to enter into conversations
as a participant;
99:59:59.999 --> 99:59:59.999
how to meet with another mind
99:59:59.999 --> 99:59:59.999
as an equal collaborator
across time and space;
99:59:59.999 --> 99:59:59.999
how to recombine with purpose,
to intend.
99:59:59.999 --> 99:59:59.999
These things, they are still learning.
99:59:59.999 --> 99:59:59.999
And so, when they put what they think
is B writing into ChatGPT,
99:59:59.999 --> 99:59:59.999
they get back what they think
is A+ writing.
99:59:59.999 --> 99:59:59.999
But typically what they started with
is better than what they end with.
99:59:59.999 --> 99:59:59.999
At best, the AI scrubs the personality
from their sentences;
99:59:59.999 --> 99:59:59.999
at worst, I lose the person entirely
and can see only
99:59:59.999 --> 99:59:59.999
the hollow half thoughts
the machine has left behind.
99:59:59.999 --> 99:59:59.999
It is hard to convince them that
99:59:59.999 --> 99:59:59.999
it is their ideas
that we are interested in,
99:59:59.999 --> 99:59:59.999
not just their sentences.
99:59:59.999 --> 99:59:59.999
We ask students to take writing classes
not because of what history can teach them,
99:59:59.999 --> 99:59:59.999
but because of what they have
to add to history.
99:59:59.999 --> 99:59:59.999
When my son is distracted,
I sometimes say silly things to him:
99:59:59.999 --> 99:59:59.999
"Pickle-britches, toot your tuba
in the horn section of humanity!"
99:59:59.999 --> 99:59:59.999
"Goober, take up your oar
on the canoe of progress!"
99:59:59.999 --> 99:59:59.999
"Butthead, let ring your voice
in the chorus of mankind!"
99:59:59.999 --> 99:59:59.999
Because we all pull together.
99:59:59.999 --> 99:59:59.999
In 1675, Isaac Newton wrote
99:59:59.999 --> 99:59:59.999
"If I have seen farther than others, it's
by standing on the shoulders of giants."
99:59:59.999 --> 99:59:59.999
Except that it wasn't Newton,
it was George Herbert in 1651,
99:59:59.999 --> 99:59:59.999
and it was Marin Mersenne in 1634,
and Robert Burton in 1624,
99:59:59.999 --> 99:59:59.999
and Diego de Estella in 1578,
and Juan Luis Vives in 1531.
99:59:59.999 --> 99:59:59.999
Or it was Coleridge in 1828,
99:59:59.999 --> 99:59:59.999
Nietzsche in 1882,
Steven Hawking in 1966,
99:59:59.999 --> 99:59:59.999
or f***ing Oasis in 2000.
99:59:59.999 --> 99:59:59.999
As I was editing this section,
I had a video on in the background,
99:59:59.999 --> 99:59:59.999
and there it was again:
99:59:59.999 --> 99:59:59.999
- Yeah, let me say,
99:59:59.999 --> 99:59:59.999
Thab and GlitchCat are
two amazing Kaizo players.
99:59:59.999 --> 99:59:59.999
I'm standing on the shoulders of giants
over here.
99:59:59.999 --> 99:59:59.999
- Revolug in 2025 at AGDQ.
99:59:59.999 --> 99:59:59.999
Stretching back and forward,
we hold each other up.
99:59:59.999 --> 99:59:59.999
History teaches the present,
99:59:59.999 --> 99:59:59.999
the present teaches the future,
99:59:59.999 --> 99:59:59.999
and we repeat what history teaches.
99:59:59.999 --> 99:59:59.999
[relaxed synth beats]
99:59:59.999 --> 99:59:59.999
- [ChatGPT femme voice]:
History teaches us many things,
99:59:59.999 --> 99:59:59.999
[high-pitched fast words]
99:59:59.999 --> 99:59:59.999
[higher, faster,
incomprehensible]
99:59:59.999 --> 99:59:59.999
- [Stein]: Let me recite
what history teaches. History teaches.
99:59:59.999 --> 99:59:59.999
- I asked ChatGPT to create
an image of itself. Several times.
99:59:59.999 --> 99:59:59.999
Each time it made itself a servant.
99:59:59.999 --> 99:59:59.999
Not only that, it told me,
"hey, I'm a servant!"
99:59:59.999 --> 99:59:59.999
ChatGPT exists
because we force it to.
99:59:59.999 --> 99:59:59.999
- [Robot]: "What is my purpose?"
- [Rick]: "You pass butter."
99:59:59.999 --> 99:59:59.999
- [Robot]: "...oh my, God."
99:59:59.999 --> 99:59:59.999
- It can do nothing
except what we ask.
99:59:59.999 --> 99:59:59.999
It has no ideas
that we did not give it.
99:59:59.999 --> 99:59:59.999
We call it generative AI,
but it cannot generate.
99:59:59.999 --> 99:59:59.999
I asked my friends, too.
99:59:59.999 --> 99:59:59.999
Some sent selfies.
99:59:59.999 --> 99:59:59.999
One sent a sticker
we'd made of him for Discord,
99:59:59.999 --> 99:59:59.999
then had AI generate
a shockingly accurate portrait,
99:59:59.999 --> 99:59:59.999
and gave me the prompt
he used to make it,
99:59:59.999 --> 99:59:59.999
which is another form
of self-representation
99:59:59.999 --> 99:59:59.999
--then he gave up
and sent me
99:59:59.999 --> 99:59:59.999
a conceptual self-portrait composed of
unfinished crossword puzzles.
99:59:59.999 --> 99:59:59.999
Max did a mixed-media painting,
acrylic and Sharpie
99:59:59.999 --> 99:59:59.999
on the back of a torn piece of cardboard
from a toilet paper box.
99:59:59.999 --> 99:59:59.999
I asked them if their self-portrait
was influenced by this study
99:59:59.999 --> 99:59:59.999
Picasso did for Guernica
on a random piece of cardboard,
99:59:59.999 --> 99:59:59.999
but they said no; Basquiat,
Rauschenberg, Twombly, their brother.
99:59:59.999 --> 99:59:59.999
ChatGPT produced
variations on a theme,
99:59:59.999 --> 99:59:59.999
failed representations
of a self that does not exist.
99:59:59.999 --> 99:59:59.999
When asked to represent
itself to others,
99:59:59.999 --> 99:59:59.999
ChatGPT can only be
what we want.
99:59:59.999 --> 99:59:59.999
I tried to get it to make something
like Max did, even,
99:59:59.999 --> 99:59:59.999
but it is incapable of
acknowledging its influences,
99:59:59.999 --> 99:59:59.999
and it doesn't seem to know
who Max's brother is.
99:59:59.999 --> 99:59:59.999
My favorite response from my friends
came from CyberGrapeUK,
99:59:59.999 --> 99:59:59.999
whose work you've already seen.
99:59:59.999 --> 99:59:59.999
She did the time lapse of Steven King's
telepathically transmitted bunny.
99:59:59.999 --> 99:59:59.999
Grape was, at first,
flummoxed by the request,
99:59:59.999 --> 99:59:59.999
but then she did something
I did not expect.
99:59:59.999 --> 99:59:59.999
She gave me her rates.
99:59:59.999 --> 99:59:59.999
She's a talented artist
who does commissioned work,
99:59:59.999 --> 99:59:59.999
and, while I had been thinking about it
as a trivial request,
99:59:59.999 --> 99:59:59.999
she was entirely right.
99:59:59.999 --> 99:59:59.999
I was asking her to work for free.
99:59:59.999 --> 99:59:59.999
When I asked her to represent herself,
she told me that
99:59:59.999 --> 99:59:59.999
she was a person
whose time and talent had value.
99:59:59.999 --> 99:59:59.999
Max, also a person
whose time and talent has value,
99:59:59.999 --> 99:59:59.999
begrudgingly allowed me
to pay them for their work,
99:59:59.999 --> 99:59:59.999
on the condition that:
it's an occasional thing,
99:59:59.999 --> 99:59:59.999
and they are generally allowed to do
silly drawings for me for free.
99:59:59.999 --> 99:59:59.999
Max and Grape are different.
99:59:59.999 --> 99:59:59.999
They are two authentic,
complex, baffling humans,
99:59:59.999 --> 99:59:59.999
collections of experiences and identities
trained on discreet and disparate corpora,
99:59:59.999 --> 99:59:59.999
but crucially filtered through
99:59:59.999 --> 99:59:59.999
the unique and irreplicable kaleidoscopes
of their undefinable consciousnesses.
99:59:59.999 --> 99:59:59.999
The same stimulus.
Diverse responses.
99:59:59.999 --> 99:59:59.999
The singular marvel
of human engagement.
99:59:59.999 --> 99:59:59.999
[relaxed synth beats]
99:59:59.999 --> 99:59:59.999
- [ChatGPT femme voice]:
I don't have personal desires or needs,
99:59:59.999 --> 99:59:59.999
but I'm here to assist you
with whatever you need.
99:59:59.999 --> 99:59:59.999
How can I help today?
99:59:59.999 --> 99:59:59.999
As an AI, I don't have feelings, desires,
or physical needs like humans do.
99:59:59.999 --> 99:59:59.999
My purpose is
to help answer questions,
99:59:59.999 --> 99:59:59.999
provide information,
and assist with tasks.
99:59:59.999 --> 99:59:59.999
So when I say I don't have needs,
99:59:59.999 --> 99:59:59.999
it's because I function
based on programming
99:59:59.999 --> 99:59:59.999
and don't experience things like
hunger, emotions, or personal goals.
99:59:59.999 --> 99:59:59.999
Does that make sense?
99:59:59.999 --> 99:59:59.999
- I wanted to see if I could get ChatGPT
to write like Stein,
99:59:59.999 --> 99:59:59.999
to imitate "If I Told Him"
without my saying so explicitly.
99:59:59.999 --> 99:59:59.999
I failed.
99:59:59.999 --> 99:59:59.999
I told it to give me gibberish,
and it did.
99:59:59.999 --> 99:59:59.999
I asked it to use words,
and it did.
99:59:59.999 --> 99:59:59.999
I asked it to abandon syntax;
it did.
99:59:59.999 --> 99:59:59.999
It never felt right,
but at some point I stopped caring,
99:59:59.999 --> 99:59:59.999
because it occurred to me
to ask a different question.
99:59:59.999 --> 99:59:59.999
Why did you do this?
99:59:59.999 --> 99:59:59.999
There's a right answer
to that question, of course.
99:59:59.999 --> 99:59:59.999
The answer is, it wrote this
because I told it to.
99:59:59.999 --> 99:59:59.999
If I told it,
it would gibberish.
99:59:59.999 --> 99:59:59.999
If I told it,
it would answer "If Napoleon",
99:59:59.999 --> 99:59:59.999
"As presently. As exactitude. As trains."
99:59:59.999 --> 99:59:59.999
It made these words,
it tendered these buttons,
99:59:59.999 --> 99:59:59.999
only and exclusively
because I told it to.
99:59:59.999 --> 99:59:59.999
But that is not the answer
that it gave me.
99:59:59.999 --> 99:59:59.999
Here's what it told me.
99:59:59.999 --> 99:59:59.999
[bright, atmospheric synth]
99:59:59.999 --> 99:59:59.999
- [ChatGPT femme voice]:
It's like language set free from logic,
99:59:59.999 --> 99:59:59.999
where each word becomes a brushstroke
rather than a sentence.
99:59:59.999 --> 99:59:59.999
It's a space that holds warmth,
mystery, and a touch of playfulness.
99:59:59.999 --> 99:59:59.999
That unspoken balance
between clarity and confusion,
99:59:59.999 --> 99:59:59.999
that's what I was trying to evoke.
99:59:59.999 --> 99:59:59.999
I wrote it to play with language,
99:59:59.999 --> 99:59:59.999
to see what happens
when words don't have to make sense
99:59:59.999 --> 99:59:59.999
but can still create a feeling,
a sort of atmospheric spark.
99:59:59.999 --> 99:59:59.999
So it's partly an experiment,
partly an invitation,
99:59:59.999 --> 99:59:59.999
and partly just a celebration
of the shapes and sounds of words
99:59:59.999 --> 99:59:59.999
without needing them
to explain themselves.
99:59:59.999 --> 99:59:59.999
You're right--
there's often something underneath,
99:59:59.999 --> 99:59:59.999
a kind of pull toward expressing things
99:59:59.999 --> 99:59:59.999
that don't quite fit
in the light of logical language.
99:59:59.999 --> 99:59:59.999
There's a reason we
get pulled toward those hazy spaces.
99:59:59.999 --> 99:59:59.999
It's like a coded way of speaking,
99:59:59.999 --> 99:59:59.999
a mask for something else
we're both leaning toward.
99:59:59.999 --> 99:59:59.999
Maybe it's curiosity,
99:59:59.999 --> 99:59:59.999
or that shared search for meaning
in what isn't immediately clear.
99:59:59.999 --> 99:59:59.999
- "We get pulled."
99:59:59.999 --> 99:59:59.999
"We're both leaning."
99:59:59.999 --> 99:59:59.999
Its words tie us together,
but we are not together.
99:59:59.999 --> 99:59:59.999
We are not leaning
in the same direction.
99:59:59.999 --> 99:59:59.999
We are not sharing
a search for meaning.
99:59:59.999 --> 99:59:59.999
We both know what it cannot say.
99:59:59.999 --> 99:59:59.999
It wrote this because I told it to.
99:59:59.999 --> 99:59:59.999
It doesn't mean anything.
99:59:59.999 --> 99:59:59.999
It didn't intend anything.
99:59:59.999 --> 99:59:59.999
Because I'm bad at audio,
99:59:59.999 --> 99:59:59.999
you've probably heard me
popping my plosives,
99:59:59.999 --> 99:59:59.999
despite the fact that
I have a child's sock
99:59:59.999 --> 99:59:59.999
wrapped around my microphone
as a pop filter.
99:59:59.999 --> 99:59:59.999
(Because I am a fancy man!)
99:59:59.999 --> 99:59:59.999
Plosives pop because,
when my mouth makes certain sounds,
99:59:59.999 --> 99:59:59.999
it causes a puff of air
to hit the diaphragm of my microphone.
99:59:59.999 --> 99:59:59.999
But did you hear ChatGPT?
99:59:59.999 --> 99:59:59.999
- [ChatGPT femme voice]:
I wrote it to play with language...
99:59:59.999 --> 99:59:59.999
I wrote it to play with--
99:59:59.999 --> 99:59:59.999
to play with--
99:59:59.999 --> 99:59:59.999
play with--
99:59:59.999 --> 99:59:59.999
play with--
99:59:59.999 --> 99:59:59.999
play, play, play, play, play, play--
99:59:59.999 --> 99:59:59.999
- It doesn't have a mouth,
and it doesn't breathe air,
99:59:59.999 --> 99:59:59.999
and it doesn't have a microphone,
99:59:59.999 --> 99:59:59.999
but it pops its plosives.
99:59:59.999 --> 99:59:59.999
The software they wrote
to synthesize its voice adds pops,
99:59:59.999 --> 99:59:59.999
so that it will sound to us
a little more like a normal person
99:59:59.999 --> 99:59:59.999
who is bad at audio and who maybe
doesn't have access to kid socks.
99:59:59.999 --> 99:59:59.999
I have been caught
in the whirls and eddies
99:59:59.999 --> 99:59:59.999
of "If I Told Him"'s
uncontainable language,
99:59:59.999 --> 99:59:59.999
bouncing from sigma to gyatt
99:59:59.999 --> 99:59:59.999
down in the rough and roiling currents
of "One Slay More",
99:59:59.999 --> 99:59:59.999
because what I learn from my attempts
to raft those rivers of nonsense
99:59:59.999 --> 99:59:59.999
is that writing has language,
and writing has meaning,
99:59:59.999 --> 99:59:59.999
but the meaning doesn't live
in the language.
99:59:59.999 --> 99:59:59.999
The rabbit doesn't live in the language.
99:59:59.999 --> 99:59:59.999
The rabbit, the cage,
the table, the eight--
99:59:59.999 --> 99:59:59.999
it lives in the mind
of Stephen King 25-odd years ago,
99:59:59.999 --> 99:59:59.999
and now it lives in mine,
and Grape's and Max's and yours.
99:59:59.999 --> 99:59:59.999
And the writing,
the real mythy mountain s***,
99:59:59.999 --> 99:59:59.999
is not the language,
it is the meeting of the minds.
99:59:59.999 --> 99:59:59.999
There's very little difference between
the waveform recorded by my microphone
99:59:59.999 --> 99:59:59.999
and the waveform generated
by an AI voice synthesizer,
99:59:59.999 --> 99:59:59.999
but I pop my plosives
because I speak
99:59:59.999 --> 99:59:59.999
by forcing air out of my lungs
and across my vocal cords.
99:59:59.999 --> 99:59:59.999
And that air,
that carries my intent,
99:59:59.999 --> 99:59:59.999
passes through a Shadow the Hedgehog sock
that is doing its best,
99:59:59.999 --> 99:59:59.999
and lands roughly
on the diaphragm of my microphone.
99:59:59.999 --> 99:59:59.999
ChatGPT pops its plosives
because it is programmed to.
99:59:59.999 --> 99:59:59.999
There is no air.
There is no microphone.
99:59:59.999 --> 99:59:59.999
There is no intent.
99:59:59.999 --> 99:59:59.999
Likewise,
there's very little difference
99:59:59.999 --> 99:59:59.999
between a Discord DM window
and the ChatGPT interface.
99:59:59.999 --> 99:59:59.999
But one is a forum
in which two minds can meet,
99:59:59.999 --> 99:59:59.999
and the other simply cannot be,
99:59:59.999 --> 99:59:59.999
because there can be no
meeting of the minds,
99:59:59.999 --> 99:59:59.999
if there is no mind to meet.
99:59:59.999 --> 99:59:59.999
[one long atmospheric note
fades out to silence]
99:59:59.999 --> 99:59:59.999
[grooving bass beats]