text,start,duration
hey everyone David Shapiro here,0.42,3.18
um,3.06,1.98
who I don't know when I'm gonna post,3.6,5.16
this video but gpt4 just dropped and um,5.04,7.58
so I'm already experimenting with it,8.76,3.86
um now the biggest thing that I noticed,12.84,6.48
in the live stream demo was that the,15.78,7.02
chat gpt4 has a token count of 32 000,19.32,4.92
tokens,22.8,4.5
and the ratio of tokens to words is,24.24,5.16
roughly two-thirds so that's about 20,27.3,3.18
000 words,29.4,5.339
twenty thousand words is a lot that is,30.48,6.96
about a hundred Pages uh just shy of 100,34.739,4.861
pages of text,37.44,5.52
um which is enough working memory,39.6,5.58
to do basically any human task almost,42.96,4.14
any human task especially when you,45.18,3.42
consider the amount of intrinsic,47.1,3.0
knowledge or background knowledge or,48.6,4.22
latent knowledge that the model has,50.1,7.979
so we just hit a new era of AI so the,52.82,8.079
the only way you can use gpt4 right now,58.079,6.54
is if you're on chat GPT plus,60.899,6.301
um and I already signed up for the API,64.619,5.581
of course but it looks like it'll be a,67.2,5.52
minute so anyways so I'm going back to,70.2,4.26
Ascension so here's the notes that I had,72.72,3.48
for Ascension,74.46,2.4
um,76.2,4.02
and then I also have the previous uh uh,76.86,5.64
one right here,80.22,4.8
um so what I'm doing is I'm adding in,82.5,4.799
some of the information that I was,85.02,5.0
brainstorming last time,87.299,5.881
Okay so,90.02,4.72
let's see what are the specific,93.18,4.5
Milestones there we go all right so,94.74,6.12
let me add some of these notes,97.68,6.24
and we're just gonna we're just gonna go,100.86,6.96
refresh Ascension part two uh great here,103.92,8.1
are some other notes we worked on in,107.82,7.46
another session,112.02,3.26
um here actually let me,117.72,5.31
just grab this as a whole,119.34,4.26
[Music],123.03,2.57
um,123.6,2.0
because the uh the context window is so,127.92,3.48
long,130.619,4.501
great here are some notes from,131.4,8.419
another session we worked on,135.12,4.699
um,141.0,4.56
wait there's more are you ready,142.14,5.34
so one thing I've noticed already is,145.56,3.899
that gpt4 is a little bit more hot on,147.48,4.02
the biscuit ready to go so you got to,149.459,4.5
tell it like wait,151.5,4.2
um,153.959,3.841
of course I am here to help take your,155.7,5.039
time okay blah blah blah I don't want to,157.8,4.92
interrupt it so you'll notice that gpt4,160.739,3.72
is a little bit slower,162.72,3.659
so that's fine all right so then we'll,164.459,3.541
come up here,166.379,6.541
um let's see I just got uh this one,168.0,5.879
um,172.92,5.099
here are some possible milestones,173.879,4.14
here actually let's just let's just grab,179.94,4.46
all this,182.04,2.36
great,185.7,2.48
um actually no that's fine,190.8,2.96
oops,204.48,2.539
let me check if there's more so,213.48,4.92
basically sorry for the silence I was,216.659,3.421
just kind of thinking,218.4,2.339
um,220.08,3.48
so I'm giving it everything and this is,220.739,4.56
let's just oops,223.56,6.78
let's just copy paste everything,225.299,5.041
into a new word document and see how,230.459,4.261
many words this is,232.56,5.759
so this is only 2300 words so this is,234.72,7.56
about a tenth of the context window,238.319,6.301
so we're we're already planning the,242.28,4.379
entire story,244.62,3.0
um,246.659,3.121
basically copy pasting everything that,247.62,3.6
I've already got,249.78,5.7
let's see there it is okay,251.22,6.9
there we go okay great so now that we've,255.48,3.9
got this okay,258.12,2.1
um,259.38,2.759
I think,260.22,5.28
that is the grand total of all the,262.139,7.741
information we've got about story so far,265.5,6.3
please,269.88,6.96
um recap everything in totality,271.8,7.16
um,276.84,2.12
uh,279.24,3.42
about this story,280.62,5.46
please capture all details,282.66,6.599
um in a complete report,286.08,3.96
um,289.259,3.201
I will be using,290.04,4.64
this document,292.46,6.88
to some uh uh let's see two uh what's,294.68,6.22
the word um,299.34,3.48
let's see as a,300.9,4.799
as the main uh let's see there's a,302.82,6.12
there's a term for it in in fiction,305.699,4.681
um where you use it like it's your main,308.94,2.52
Bible,310.38,2.94
what's the main Bible for,311.46,2.459
um,313.32,2.04
story,313.919,3.301
okay so while that's going let's talk,315.36,4.559
about gbt4,317.22,4.86
um so I watched the whole live stream,319.919,6.78
gpt4 is multimodal it can accept images,322.08,6.6
um if you look at,326.699,4.761
product,328.68,2.78
um they mention,332.88,6.02
I think they mentioned videos,335.46,3.44
um did they,339.66,6.18
no okay so basically it has it can,342.419,5.581
understand images which means that it,345.84,4.5
can look at a screen right it can look,348.0,4.139
at your code it can look at what you're,350.34,3.78
doing and understand what it's doing so,352.139,3.961
it's multimodal on the input I don't,354.12,3.48
think it's multimodal on the output,356.1,4.2
that's fine,357.6,4.319
um and you can see okay here it is it's,360.3,4.32
going a little bit slower,361.919,5.301
um there we go,364.62,2.6
um,368.039,5.16
yeah but you know so in terms of the,369.0,7.68
cognitive performance of gpt4 it looks,373.199,5.581
like it's an incremental,376.68,5.88
um improvement over GPT 3.5,378.78,6.78
um so that is that is the that is like,382.56,4.8
where we're starting,385.56,5.4
um but on top of the incremental,387.36,7.02
improvements there is uh there's this,390.96,5.1
the cognitive Improvement and so if you,394.38,3.3
look on this,396.06,2.16
um,397.68,2.88
it they talk about like how much it has,398.22,4.08
improved on benchmarks,400.56,4.56
so the uniform bar exam Chad gbt placed,402.3,4.86
in the 10th percentile so it passed the,405.12,4.799
bar gpt4 placed in the 90th percentile,407.16,4.92
so that means that,409.919,5.461
um gpt4 now outperforms 90 percent of,412.08,5.82
lawyers on the bar exam,415.38,6.48
and then the biology Olympiad it was in,417.9,5.82
the 31st percentile now it is in the,421.86,3.6
99th percentile,423.72,4.62
so uh,425.46,5.34
to say that this is a Quantum Leap,428.34,5.52
Forward is not an exaggeration so I just,430.8,5.1
posted a um,433.86,5.459
a uh uh a poll,435.9,5.94
a couple days ago,439.319,4.681
um I guess it was actually one day ago I,441.84,5.28
I said will we achieve AGI this year and,444.0,8.28
20 19 say yes AGI this year 73 so almost,447.12,7.32
three quarters of people say no AGI in,452.28,4.56
the future and then um nine percent say,454.44,5.28
AGI already happened I suspect this is,456.84,4.919
going to change after people see what,459.72,5.9
what gpt4 is capable of,461.759,3.861
okay cool,465.78,8.0
um disentangling from Power there we go,468.479,5.301
Moment of Truth,474.3,3.14
great,478.02,2.54
um okay this is not quite as detailed as,481.86,4.02
I had hoped but that's okay because all,484.259,3.66
the information is here,485.88,6.42
um let's see you seem to be missing Act,487.919,6.06
One,492.3,3.72
um plot Beats,493.979,5.06
so let's see what it says,496.02,3.019
okay so,502.08,5.239
um,505.259,2.06
I was just talking to my fiance about,508.139,6.181
the scale of 32 of a 32 000 token memory,510.08,7.36
so that's about 20 000 Words which when,514.32,4.68
you combine that twenty thousand words,517.44,3.12
with the intrinsic knowledge that the,519.0,3.18
model already has so that means that you,520.56,2.82
only need to give it the relevant,522.18,4.14
updated details like the documentation,523.38,5.22
that you're working on or,526.32,4.32
um some episodic memory or whatever and,528.6,4.14
it can kind of confabulate or,530.64,3.78
triangulate the rest of the of the,532.74,3.719
missing information so for instance I,534.42,4.2
don't need to describe Rome to the model,536.459,4.741
when I'm talking about Rome,538.62,3.8
um,541.2,5.4
so that is pretty interesting not only,542.42,7.08
that it is it surpasses human abilities,546.6,5.7
uh in a lot of things and so here I'm,549.5,4.66
just giving it a little bit of steering,552.3,4.44
and off it goes to the races and I'm,554.16,4.44
like okay great,556.74,2.64
um,558.6,2.04
great,559.38,1.92
um,560.64,2.34
what do you think,561.3,6.78
um should we try to write act one or uh,562.98,9.24
plot the rest of act two and three,568.08,6.24
so let's see what it says I'm as a,572.22,3.119
language model I'm happy to help you,574.32,3.72
with either option okay,575.339,5.641
um so that's actually good I did say at,578.04,4.14
the beginning that it seems like it's a,580.98,3.24
little bit hot on the biscuit so I would,582.18,4.08
have expected it to say let's rate,584.22,3.299
um but it seems like it's taking a,586.26,3.66
little bit more balanced uh uh proposal,587.519,4.88
right now,589.92,2.479
yeah,595.08,3.12
um you know what,596.339,3.661
we don't know exactly where the story is,598.2,3.24
going to end up,600.0,2.76
um so,601.44,3.839
I would say,602.76,5.699
Let's uh because 20 000 words that's,605.279,6.24
going to basically be the first act so,608.459,4.861
well let's see what it let's see what it,611.519,3.361
says okay,613.32,2.579
um,614.88,3.36
I think we should,615.899,4.38
go ahead and,618.24,5.46
plan and write act one in a bit more,620.279,7.56
detail uh most novels are 30 to 50,623.7,6.48
chapters,627.839,6.241
um so act one should be,630.18,8.159
um eight to uh what's that about 15.,634.08,10.22
8 to 15 chapters or so uh can you,638.339,9.56
take those main plot Beats,644.3,6.539
we discussed,647.899,8.261
and map those onto about a dozen,650.839,9.341
chapters or so please include,656.16,7.619
details about what happens in each,660.18,5.279
chapter,663.779,5.781
um be sure to include information about,665.459,9.56
plot advancement character development,669.56,5.459
um world building,675.3,8.34
um and uh and themes for each chapter so,678.3,6.779
one thing that I have found in using,683.64,4.74
chat GPT to help me write is you can,685.079,5.88
weave together all four of those pillars,688.38,5.94
of Storytelling in one go plot character,690.959,6.06
setting theme those are the four primary,694.32,4.44
ingredients you put those together you,697.019,4.621
tell chat GPT that's what you want and I,698.76,4.86
won't say nine times out of ten about,701.64,3.96
fifty percent of the time it does pretty,703.62,4.26
good A couple other times you have to,705.6,3.9
say let's look at this from a slightly,707.88,3.6
different perspective,709.5,3.959
um and that's where I use jungian,711.48,4.32
archetypes and Joseph Campbell so that,713.459,4.38
it's so that it basically adopts like oh,715.8,3.539
actually I know what I'm talking about,717.839,3.421
when we talk about constructing a story,719.339,4.201
you could also use Save the cat or any,721.26,3.06
other,723.54,3.96
um models uh for storytelling so you can,724.32,4.98
say like let's do a five act structure,727.5,4.519
like a Shakespeare or a or a four act TV,729.3,5.82
structure so anyways let's see what it,732.019,5.32
does with this so,735.12,4.5
one of the one of the common rules is,737.339,4.68
less is more,739.62,4.92
um and so this is becoming increasingly,742.019,4.921
true as these models become smarter you,744.54,4.919
give it just enough information,746.94,4.44
um to to know what your goal is and then,749.459,4.82
you kind of set it free,751.38,2.899
all right so introduction selling the,756.779,4.5
heirloom,758.76,4.8
there we go chapter three exploring Rome,761.279,3.721
experiencing the city's vibrant,763.56,2.719
atmosphere,765.0,3.779
Rome's Landmark social classes daily,766.279,4.481
life themes Discovery contrast between,768.779,3.421
rich and poor,770.76,3.54
chapter four a chance encounter I love,772.2,5.18
these not these titles,774.3,3.08
okay so because,779.94,3.98
let's see I don't think I think the,784.92,6.02
first Patron that's too soon,787.2,3.74
because also she's like a teenager,793.98,5.76
I don't know well I guess if she's 16 or,797.1,4.76
whatever,799.74,2.12
so,802.38,4.68
yeah explores the city a public event,805.2,3.84
meets an influential woman confronts her,807.06,3.18
mother,809.04,4.859
yeah that was act yeah hold on hold on,810.24,7.039
stop stop stop stop stop,813.899,3.38
hold on a second,818.04,5.88
so in this case it clearly lost the plot,819.839,7.081
um so because we we had five plot beats,823.92,4.979
here so,826.92,2.94
um,828.899,4.801
let's see uh let's see please expand the,829.86,6.18
five plot beats,833.7,7.439
um listed below into 12 chapters,836.04,6.84
um,841.139,3.661
because right now it's confabulating,842.88,4.28
stuff,844.8,2.36
so we can see that this this task,847.8,6.36
complexity is not quite up to Snuff it,850.62,6.68
needs a little bit more help,854.16,3.14
um yep there we go,862.139,3.5
all right so let's try again,866.339,3.841
here's a detailed breakdown of Act One,868.44,3.36
expanding the five plot beats into 12,870.18,4.32
chapters okay so it was off to an okay,871.8,4.5
start I probably should have saved that,874.5,3.839
but when I saw that it was It was kind,876.3,4.38
of rushing through the story and to be,878.339,4.68
fair rushing through the story like that,880.68,4.98
is pretty typical for amateur writers,883.019,5.701
actually is it takes a lot of discipline,885.66,6.359
to slow down and say we'll get there,888.72,5.9
um there we go,892.019,2.601
I like how we're kind of starting off,896.279,4.381
nice and slow the sale of the heirloom,898.62,6.54
Cornelia explores Rome the public event,900.66,6.66
so,905.16,6.859
public event is there a new mentor,907.32,4.699
Court Intrigue and social climbing,912.899,6.0
she's the first major Patron okay so,915.8,5.8
it's really hell-bent on just jumping,918.899,5.221
the gun okay so this this is an example,921.6,5.88
of what I mean by hot on the biscuit,924.12,4.5
um,927.48,4.74
hang on you're getting to the,928.62,8.64
um to the end game now uh let's uh let's,932.22,6.66
say too fast,937.26,3.48
um let's slow,938.88,3.48
this down,940.74,3.24
um,942.36,4.56
think through everything,943.98,8.4
that uh that happens in act one,946.92,10.46
of a story based on the three Act,952.38,6.54
structure,957.38,5.139
or Joseph Campbell's,958.92,4.5
um,962.519,4.041
hero's journey,963.42,3.14
now,970.139,3.26
based on that,971.519,4.56
let's brainstorm so telling it to,973.399,4.18
brainstorm rather than just go ahead and,976.079,4.38
write let's brainstorm,977.579,4.681
um,980.459,9.32
12 chapters that only adhere to act one,982.26,7.519
therefore act one must end with the,989.88,7.019
Crossing the Rubicon,994.459,5.8
moment of no return,996.899,7.5
maybe we need to discuss that before,1000.259,6.0
proceeding,1004.399,4.081
okay let's see what it says,1006.259,4.26
you're right I may have progressed too,1008.48,4.979
quickly through act one again pretty,1010.519,4.74
typical amateur,1013.459,5.101
um uh uh move since I know better I can,1015.259,4.921
keep it on track I was hoping that it,1018.56,3.3
would be able to figure that out but,1020.18,4.159
that's fine,1021.86,5.16
okay for the Crossing the Rubicon moment,1024.339,4.12
this could be when Cornelia makes a,1027.02,2.76
decision that commits her to the path of,1028.459,4.021
becoming a courtesan there we go so the,1029.78,4.799
refusal will actually probably be,1032.48,4.92
refusing the mentor the mysterious,1034.579,5.72
influential woman,1037.4,2.899
okay so since it's since I gave it a,1040.76,3.48
little bit of feedback and guidance,1043.04,2.7
let's see,1044.24,5.28
if it's able to figure out that okay so,1045.74,5.76
she's exploring Rome the sale of the,1049.52,5.22
family heirloom okay there we go there,1051.5,5.46
we go so it's slowing it down so she,1054.74,5.7
discovers the heirloom in chapter two,1056.96,6.0
um then she explores Rome she's she's,1060.44,4.08
exploring she doesn't know what's going,1062.96,5.099
on then she sells it she decides to sell,1064.52,7.2
it because she realizes oh here we go,1068.059,5.581
um and then okay so the mysterious,1071.72,4.44
figure gets there uh better life and,1073.64,5.22
loyalty to the family oh here we go a,1076.16,4.259
chance encounter she meets her future,1078.86,5.46
Mentor who introduces her to the War uh,1080.419,4.561
um,1084.32,2.64
world of courtesans,1084.98,4.559
she attends a public event where she'll,1086.96,4.079
make important connections allies and,1089.539,3.721
enemies,1091.039,4.981
um among the events guests ooh so a lot,1093.26,5.7
happens at this event the mentors offer,1096.02,4.32
there we go,1098.96,2.94
the mentor offers her guidance and,1100.34,2.94
assistance in becoming a successful,1101.9,2.7
courtesan there we go the mentor,1103.28,2.639
provides her with resources and,1104.6,3.3
connections to get started family,1105.919,4.021
tension she faces disapproval and,1107.9,4.639
concern yes that is the debate perfect,1109.94,4.979
weighing the options Cornelia considers,1112.539,4.661
the consequences and the values she,1114.919,3.901
crosses the Rubicon she accepts her,1117.2,3.24
mentor's offer,1118.82,5.58
there we go much better,1120.44,8.16
excellent much better okay let's take,1124.4,8.7
that exact outline of 12 chapters and,1128.6,6.42
expand on,1133.1,6.72
um the details uh for each one let's,1135.02,6.18
start,1139.82,2.52
um,1141.2,3.479
uh let's start by so this is what I,1142.34,5.82
found chat GPT has the best is once you,1144.679,5.281
have the chapter then you say give me,1148.16,4.019
the scenes in the chapter and honestly,1149.96,3.9
this is not a bad way for humans to,1152.179,3.141
outline it too,1153.86,4.679
let's start by expanding,1155.32,4.06
um,1158.539,5.701
uh the list uh with scenes so each,1159.38,9.419
chapter should be between one and uh,1164.24,9.08
let's say three or four uh scenes,1168.799,9.601
I think most chapters in novels are,1173.32,6.7
one to three,1178.4,3.6
so that's fine,1180.02,2.88
um,1182.0,2.58
so given,1182.9,3.3
this,1184.58,2.9
please,1186.2,6.18
restate the chapters but expand upon,1187.48,8.559
them by listing listing out and,1192.38,5.46
describing,1196.039,4.581
scenes,1197.84,2.78
um yeah all right let's see if it can,1201.26,3.74
handle that,1202.64,2.36
do yeah so again you can tell that there,1205.82,6.9
is that I as an expert am providing a,1209.299,5.341
lot of structure and guidance to this so,1212.72,4.459
that's fine,1214.64,2.539
Cornelia and her family are shown,1219.26,2.46
struggling in their daily life,1220.52,3.48
showcasing this blah blah okay so it's,1221.72,4.079
just defaulting to two scenes that's,1224.0,3.12
fine,1225.799,4.681
um certainly uh for for more amateur,1227.12,5.4
writers just following something more,1230.48,4.079
algorithmic like this is perfectly fine,1232.52,4.2
I mean you hear me talking about you,1234.559,3.541
know the three-act structure and Joseph,1236.72,3.78
Campbell and and that so on and so forth,1238.1,4.86
so I can I there there was someone had,1240.5,5.299
it wrote a comment like was it my my my,1242.96,5.839
University literature Professor,1245.799,5.38
complained that they have spent the last,1248.799,4.361
30 years trying to undo the damage that,1251.179,4.74
Joseph Campbell did and it's like you,1253.16,4.56
know what Joseph Campbell's work has,1255.919,4.741
produced many billion dollar movies so,1257.72,4.14
you know,1260.66,2.94
um if you watch the special features of,1261.86,4.5
Avengers end game actually there's a or,1263.6,4.8
maybe it was um Iron Man,1266.36,4.38
one of them anyways so in the special,1268.4,4.68
features of some of the MCU movies near,1270.74,4.5
the end of phase four,1273.08,4.32
um they had a round table with all of,1275.24,5.04
the directors and they kept alluding to,1277.4,5.34
the old book and I'm like hmm I wonder,1280.28,4.44
if that's the hero's journey,1282.74,3.78
because if you if you're familiar with,1284.72,3.36
the hero's journey and Joseph Campbell's,1286.52,3.659
work you watch,1288.08,4.32
um you know particularly,1290.179,4.561
um the Iron Man movies and the Avenger,1292.4,3.6
movies,1294.74,4.98
um they very very algorithmically follow,1296.0,6.0
um those kinds of things,1299.72,7.26
oh okay so there we go we're almost done,1302.0,7.32
we've got two scenes for each one that's,1306.98,3.78
fine,1309.32,3.239
she confides on a close friend or,1310.76,3.299
sibling so it's still it's still kind of,1312.559,3.061
waffling like it's it's being kind of,1314.059,6.181
generic but given how long this is and,1315.62,6.179
and the fact that it's it's staying on,1320.24,3.0
the plot,1321.799,4.081
um is is pretty impressive,1323.24,4.679
um I really like this,1325.88,4.88
okay cool,1327.919,2.841
um,1331.82,2.88
I'm gonna go ahead and copy this out,1333.02,6.18
actually because this is really good,1334.7,8.099
so we come back here so act one,1339.2,5.76
um so then we'll do uh let's see,1342.799,5.12
chapters,1344.96,2.959
uh let's see,1349.78,6.34
unformatted text there we go,1353.02,4.48
um let's see how long are we at we're,1356.12,3.24
only at 22 minutes we can keep going we,1357.5,3.6
can keep going,1359.36,3.96
um let's see where should we go next,1361.1,6.42
okay where should we go next,1363.32,8.04
um what do you think what typically goes,1367.52,7.74
into planning scenes all right so let's,1371.36,5.88
see let's see if it knows now that you,1375.26,3.419
have a detailed outline for act one you,1377.24,3.66
can consider moving on to planning Act 2,1378.679,3.781
and act three however if you prefer to,1380.9,3.54
focus on act one for now you can delve,1382.46,3.78
in deeper into the planning of the,1384.44,3.66
scenes,1386.24,3.36
um let's see when planning scenes,1388.1,3.059
consider the following aspect's purpose,1389.6,3.72
ensure each scene has a clear purpose in,1391.159,3.601
advancing the plot developing characters,1393.32,3.839
or building the world,1394.76,4.74
um setting okay yep choose an,1397.159,5.181
appropriate setting,1399.5,2.84
characters okay so this is this is,1403.58,5.219
pretty generic but it's good in order to,1406.1,5.4
remind itself because there's um so,1408.799,4.86
there's this concept in Neuroscience,1411.5,5.34
called priming I'm actually reading,1413.659,5.4
this book behave,1416.84,4.26
um it is a very dense book but it's,1419.059,4.141
actually hilarious the dude who wrote it,1421.1,4.199
is sarcastic as hell,1423.2,6.78
um anyways so even just reading words of,1425.299,7.441
a particular tone can really drastically,1429.98,4.62
change your cognition and decisions and,1432.74,4.2
behavior the same is also true of,1434.6,4.62
language models just by priming it with,1436.94,4.2
some ideas like when I said think about,1439.22,4.439
Joseph Campbell and what goes into an,1441.14,4.32
act one and you saw how much it improved,1443.659,5.221
so this is called priming so a big part,1445.46,7.079
of using language models is priming,1448.88,7.5
um great cool so,1452.539,3.841
um,1456.74,3.72
I like that,1458.419,3.541
uh,1460.46,5.64
let's start with chapter one please,1461.96,8.339
write out a much more detailed outline,1466.1,6.9
of chapter one,1470.299,4.821
um,1473.0,2.12
after one the desperate decision,1478.4,4.38
scene one introducing Cornelia the story,1480.62,3.539
opens with Cornelia and her family's,1482.78,2.519
modest home providing a brief,1484.159,2.88
description of her parents and,1485.299,4.921
personality again this is all kind of uh,1487.039,4.921
algorithmic,1490.22,4.5
um if you open it this way like so if,1491.96,5.4
you if you describe the main character's,1494.72,5.64
personality or show it right off the bat,1497.36,5.52
that's going to be a little bit like,1500.36,4.74
kind of on the nose,1502.88,4.56
um whereas like you you definitely,1505.1,4.52
should demonstrate your main character,1507.44,4.619
in the first chapter,1509.62,4.78
but if you just describe it with an info,1512.059,5.041
dump that's not the best,1514.4,5.3
um let's see,1517.1,2.6
I moved on to Scene Three,1521.9,6.92
deciding to sell the heirloom,1525.14,3.68
yeah we didn't we didn't go that far,1533.059,6.061
I like that it invented the name okay,1536.72,3.959
well searching for something valuable to,1539.12,3.419
sell she discovers the family's heirloom,1540.679,3.6
okay but that doesn't happen until,1542.539,4.52
chapter two,1544.279,2.78
um,1547.46,2.18
so it looks like it looks like we need,1555.86,4.88
to prime it again,1557.779,2.961
here so we'll do we'll come back here,1562.7,6.12
and say uh out after outline of chapter,1565.299,4.781
one,1568.82,3.18
um as a reminder,1570.08,5.76
here is the grand totality of chapter,1572.0,8.419
one that we've discussed so far,1575.84,4.579
please stay Within,1581.679,6.961
the bounds of chapter one,1584.6,4.04
okay now please expand this brief,1591.74,9.72
outline into a more detailed outline so,1596.779,6.301
you might have noticed that sometimes,1601.46,3.78
what I'm doing is I'm rewriting the last,1603.08,4.44
one because what you want is if it gets,1605.24,3.78
something catastrophically wrong like,1607.52,3.48
this you want to remove it from the chat,1609.02,3.659
history because otherwise you end up,1611.0,3.36
with what's called prompt contamination,1612.679,3.841
remember how I just talked about priming,1614.36,4.62
priming works both ways it can be,1616.52,4.38
positive in that you can prime the model,1618.98,4.02
to think in the way that you want it to,1620.9,4.5
but if it has wrong information in the,1623.0,4.08
history that's going to continue,1625.4,4.56
polluting it and human like humans can,1627.08,4.5
get stuck in a rut like this too right,1629.96,3.9
remember like if you're ever having a,1631.58,3.599
discussion or you're in a meeting or,1633.86,3.179
whatever and you get the wrong idea in,1635.179,3.181
your head it can be really hard to get,1637.039,3.481
out of that rut so in that case it's,1638.36,4.14
actually kind of similar to humans one,1640.52,3.539
advantage that these models have is you,1642.5,3.24
can just flat out erase the wrong,1644.059,4.941
information from its memory,1645.74,3.26
okay so let's see if it does a little,1649.4,6.5
bit better this time also I'm uh,1651.559,4.341
uh I'm still contending with burnout,1657.2,5.339
myself and so I um I probably shouldn't,1658.82,5.219
be doing this right now but it's just,1662.539,3.361
too exciting,1664.039,4.801
okay I like this description of the,1665.9,4.86
modest home with her family highlighting,1668.84,2.819
contrast between their current,1670.76,3.0
circumstances and their once prosperous,1671.659,4.5
past there we go Cornell helps her,1673.76,4.019
mother with chores revealing her dutiful,1676.159,2.941
nature and her desire to contribute to,1677.779,4.801
the family oh that's good so this is,1679.1,6.0
showing instead of telling much much,1682.58,5.28
better so remember when it when the last,1685.1,4.319
time it was like oh we're just gonna,1687.86,4.199
tell you about her personality no wrong,1689.419,5.221
you show their personality and so you,1692.059,4.321
you show that she's helping her mother,1694.64,3.72
with chores you show that she's,1696.38,4.62
bickering with her siblings,1698.36,5.04
um you show them talking about how,1701.0,3.659
things used to be,1703.4,2.7
great,1704.659,2.941
um Cornelius father returns looking,1706.1,3.0
weary and defeated the family gathers,1707.6,3.9
for a simple Frugal meal highlighting,1709.1,4.98
their financial difficulties this is,1711.5,4.14
phenomenal,1714.08,3.599
um I'm gonna call this a day because,1715.64,4.26
holy crap,1717.679,4.801
um this is this is exciting and just to,1719.9,5.46
think that chat GPT just came out a few,1722.48,5.04
months ago that was what four months ago,1725.36,5.46
and we're already what quadrupling going,1727.52,7.32
from 8 000 tokens to 32 000 tokens,1730.82,7.5
uh this this poll that I did AGI this,1734.84,6.18
year yeah tell me what you think now,1738.32,7.94
all right that's it wrapping it up bye,1741.02,5.24