text,start,duration
I think we're ready to begin,0.079,4.961
um he saw me look at the camera used to,3.419,2.821
be over there now you're over here okay,5.04,6.0
so I ran a poll uh 17 hours ago I asked,6.24,7.319
if you all wanted me to do a video on,11.04,6.3
the euai ACT man got names gotta get,13.559,5.461
better with names I'm good with acronyms,17.34,3.599
come to me for acronyms,19.02,5.88
gato Gaia Aegis okay anyways so y'all,20.939,5.881
were confused about what I meant by Haas,24.9,3.18
which is something from here in the,26.82,2.879
South basically Haas means just,28.08,5.1
something big right it's short it's a,29.699,6.421
southern dialect for a horse or in,33.18,5.16
particular a big horse,36.12,5.22
um so this document is a Haas because,38.34,5.46
look how freaking long it is,41.34,4.8
it just keeps going,43.8,4.739
so I plugged it in over here,46.14,6.419
and it is uh 70 000 tokens long so even,48.539,7.86
if I had access to uh GPT 432k which I,52.559,5.881
did briefly and then they revoked it I,56.399,4.021
think they got overwhelmed uh because I,58.44,5.52
used it I tried to use uh GPT uh 432k on,60.42,5.16
a couple things and it was just like,63.96,3.96
server is overloaded servers are anyways,65.58,4.68
so hopefully I get that back but the,67.92,4.5
idea was there would be I could read it,70.26,3.66
and you know chop it up into maybe four,72.42,3.96
sections uh three or four sections and,73.92,4.08
read the whole thing but we don't have,76.38,5.48
that and I do have uh GPT 3.5 turbo,78.0,7.02
16k0613 so this model is actually super,81.86,6.46
super steerable it is actually better,85.02,8.04
than original gpt4 at a lot of things,88.32,6.36
um it's still kind of dumb especially if,93.06,3.72
the instructions are too complex but as,94.68,5.72
long as you give it just one task uh GPT,96.78,8.879
3.3.5 turbo 0613 the June 13th model,100.4,7.24
um is more than adequate for most things,105.659,4.32
now that being said because I'm,107.64,4.439
interpreting legalese and I'm doing,109.979,5.901
summarization I'm going to use 04,112.079,6.54
613. So the instructions that I gave it,115.88,4.96
just now and I know that I'm ready to,118.619,4.561
start the video primary purpose,120.84,5.0
zoom in a little bit the primary purpose,123.18,4.799
or the following is a chunk of,125.84,3.88
legislation your job is to summarize,127.979,4.501
this legislative legislation with two,129.72,5.46
primary goals one drastically reduce the,132.48,4.8
word count and two retain enough context,135.18,3.9
that it will still make sense the reason,137.28,3.48
for this is the current AI technology,139.08,3.06
can only read chunks that are so big,140.76,3.36
method the user will give you a chunk of,142.14,4.2
legislation via chat your response must,144.12,3.9
only be the summarized version do not,146.34,3.24
chat with the user engage with them or,148.02,3.299
offer any other commentary summarize,149.58,4.879
only so I copied the first hundred lines,151.319,7.041
of the legislation into here,154.459,7.301
and this is what it spit out so it took,158.36,5.26
all that and just said okay this is what,161.76,4.38
it means so the first hundred lines,163.62,5.1
which you can see here,166.14,6.92
so we've got 344 000 characters,168.72,7.5
3665 lines of stuff to go through I'm,173.06,4.6
not going to read the whole thing I'm,176.22,5.34
not AI explained uh that dude just like,177.66,5.82
y'all are impressed with how much work I,181.56,3.24
can do I'm impressed with how much that,183.48,4.02
dude can read just honestly I'm maybe he,184.8,6.62
uses chat GPT I don't know anyways,187.5,7.14
if you're watching it's just a joke,191.42,6.36
um okay so anyways uh so ultimately,194.64,6.86
basically that that first hundred lines,197.78,6.819
uh was summarized and condensed down to,201.5,5.14
the European commission proposes a,204.599,4.261
regulation for harmonized rules on,206.64,3.72
artificial intelligence aiming to,208.86,4.26
balance the socioeconomic benefits of AI,210.36,5.28
with potential risks The Proposal seeks,213.12,4.8
to ensure AI systems are safe respect,215.64,4.08
fundamental rights provide legal,217.92,3.36
certainty for investment in Innovation,219.72,3.96
enhance governance and prevent Market,221.28,4.739
fragmentation so it has says nothing,223.68,5.1
about existential risks or prevent AI,226.019,4.44
from killing everyone it's sort of,228.78,3.78
implicit because like if AI kills,230.459,3.481
everyone you can't have investment in,232.56,3.42
Innovation but I would rather see,233.94,5.28
legislation that is explicit it proposes,235.98,4.74
a risk-based approach prohibiting,239.22,3.599
certain harmful AI practices and setting,240.72,4.32
requirements for high-risk AI systems,242.819,4.14
okay cool we'll see how they Define that,245.04,3.18
later,246.959,3.301
The Proposal also establishes a,248.22,3.719
governance system at member state level,250.26,3.36
and a European artificial intelligence,251.939,4.02
board at Union level Okay cool so this,253.62,3.54
is this is going to be an international,255.959,4.441
regulator great it is consistent with,257.16,5.34
existing Union legislation and policies,260.4,3.96
including data protection consumer,262.5,3.36
protection non-discrimination and gender,264.36,3.839
equality so these are all very low level,265.86,4.559
risks none of this is actually talking,268.199,4.56
like when they say high risk they're,270.419,3.84
probably meaning like the social credit,272.759,4.861
system uh that is used in China which,274.259,5.22
that is a high risk but it is not the,277.62,4.5
highest risk The Proposal is based on,279.479,4.261
article 114 of the Treaty of the,282.12,3.18
functioning of the European Union Union,283.74,5.519
the tfeu which ensures The Establishment,285.3,5.459
and functioning of the internal Market,289.259,4.38
okay so this is mostly a market-oriented,290.759,5.041
piece of legislation great so we know,293.639,4.141
that this works so let's go ahead and,295.8,3.36
save this out I'm going to walk you,297.78,3.419
through the whole process,299.16,4.62
um and then I'll yeah so this is this is,301.199,4.141
how this is how we're going to proceed,303.78,3.24
for those of you that are newer to my,305.34,3.24
channel this is actually the original,307.02,3.72
format that I used where I would where I,308.58,3.3
would walk you through the entire,310.74,3.899
process so just watched it on 2X and,311.88,5.46
you'll see the whole process of how I,314.639,4.681
approach these things so we'll call this,317.34,6.78
system summarize yes dot text,319.32,7.26
and so basically I give it a system a,324.12,4.38
system message with very very clear,326.58,4.44
instructions uh and and what I expect,328.5,4.68
and and what it should expect and then,331.02,3.66
it gives you very clear consistent,333.18,5.76
results so then we come over to,334.68,5.48
um,338.94,4.259
Step One simplify so this is I'm just re,340.16,6.879
I'm borrowing code uh which I always do,343.199,5.761
um,347.039,3.901
so we've got this so this is what we're,348.96,4.38
going to use to do that,350.94,4.68
um this is this is this is a KB service,353.34,3.66
that I'm working on so I'll show you,355.62,4.26
that in an upcoming video,357.0,6.68
um but you don't get to see that yet,359.88,3.8
we don't need KB functions so delete,364.139,3.541
that,366.539,3.241
flask routes we're not going to do a,367.68,4.079
flask app so that's also fine actually I,369.78,4.8
guess I just delete most of it okay cool,371.759,4.5
so we get rid of flask because we're not,374.58,2.76
going to do that we're not going to do,376.259,3.121
threading We're Not Gonna we don't need,377.34,7.94
flask or logging or Json or yaml okay,379.38,5.9
clean it up I probably should have done,386.22,3.539
that before I started the video but it's,388.199,3.961
good to it's good for you to see how,389.759,4.201
messy my process is,392.16,3.3
um I am the first person to tell you I,393.96,3.0
am not a professional developer I never,395.46,4.079
was my uh professional background was I,396.96,4.739
was uh in virtualization and automation,399.539,4.38
so I'm an automation engineer which is,401.699,3.661
why I approach things the way that I do,403.919,4.56
which is creating discrete steps,405.36,4.679
um so you notice the FI the title of,408.479,4.141
this is uh step one simplify,410.039,4.561
um Okay cool so basically what we're,412.62,6.12
going to do is we're going to ask this,414.6,7.379
handy dandy uh coding chat bot that I,418.74,6.54
built which you can use this it is,421.979,6.421
available over,425.28,5.52
here,428.4,4.62
um I had a bunch of people submitting,430.8,6.239
um uh uh pull requests and they kept,433.02,6.06
trying to like change the way that this,437.039,3.301
thing works so I just set it to public,439.08,4.559
archive so here's the thing is I welcome,440.34,6.359
pull requests that are like,443.639,5.761
simple fixes or small improvements don't,446.699,4.741
refactor my code just Fork it right like,449.4,4.26
there's 19 Forks if you want to if you,451.44,3.84
want to fundamentally change how it,453.66,3.06
works fine,455.28,3.18
um but I got tired of people messing up,456.72,3.36
my code so I just set it to public,458.46,3.54
archive and guess what like people you,460.08,3.059
can still do whatever you want with it,462.0,4.46
anyways so we're going to use this as a,463.139,5.641
co-pilot tool because I got frustrated,466.46,5.56
with GitHub because they're co-pilot uh,468.78,5.58
customer service whatever documentation,472.02,4.619
is garbage okay,474.36,5.339
uh let's see I have um here let's just,476.639,6.481
write a function that accepts,479.699,4.021
um,483.12,4.68
a gigantic string and returns,483.72,5.0
um,487.8,4.32
let's see and let's see splits it by,488.72,4.72
lines,492.12,3.54
and then chunks,493.44,7.68
um 100 lines uh together and returns uh,495.66,7.319
those chunks,501.12,5.4
as strings so,502.979,4.321
um,506.52,3.079
the final result,507.3,4.56
will be,509.599,4.661
um a list,511.86,4.08
of strings,514.26,6.48
each string represents 100 lines of the,515.94,6.5
original,520.74,4.8
okay so this is this is my coding,522.44,5.98
chatbot assistant that um it uses it,525.54,5.16
also uses gpt4,528.42,3.96
um you I I tried it you can actually,530.7,4.259
switch it to um,532.38,4.74
oh it's using the old one,534.959,5.161
hold on now I need to update that,537.12,6.18
hold on hold on that's the wrong model,540.12,5.58
coding chatbot assistant see this is,543.3,3.96
what I mean is someone someone updated,545.7,3.66
it oh no my local one uses the right one,547.26,3.72
so someone updated it with the wrong,549.36,3.36
model and made a bunch of other,550.98,3.96
assumptions and I was just like no this,552.72,5.76
is not this is not what I wanted okay,554.94,7.079
so here we go here's the function python,558.48,6.359
chunk strings line lines equals split oh,562.019,5.641
wow I did it in two lines cool,564.839,4.56
um,567.66,4.76
this functions first splits the string,569.399,6.841
into lines using the delimiter then it,572.42,7.08
uses a list,576.24,5.099
comprehension to create chunks of 100,579.5,5.38
lines each the range yep the result is a,581.339,5.161
list of strings cool,584.88,3.66
excellent excellent excellent so I'm,586.5,4.38
going to go ahead and copy this out,588.54,4.26
oh and another good thing is that uh,590.88,5.639
like well yeah you see how it works all,592.8,7.08
right so we'll go ahead and,596.519,6.601
do that so chunk strings so first we,599.88,5.04
want to just um,603.12,4.38
act equals,604.92,5.039
open file and I've got it named as,607.5,4.56
act.txt,609.959,4.94
just right here,612.06,2.839
um yep,614.94,5.76
oh and also this uh excuse me this this,616.2,7.079
uses a scratch Pad so what I can do is I,620.7,4.139
can actually populate the code that I'm,623.279,5.041
working on uh,624.839,5.041
over here,628.32,3.66
so now it's aware of what I'm working on,629.88,4.74
so this scratch Pad is a holdout so,631.98,4.32
rather than giving it the code in,634.62,4.68
sequence the the coding chat bot,636.3,4.44
actually has,639.3,4.68
a uh it has here let me show you the,640.74,4.86
system message so here's the,643.98,2.94
instructions your python coding,645.6,3.84
assistant let me turn on word wrap your,646.92,4.8
python coding assistant etc etc scratch,649.44,4.26
Pad so basically tells you the scratch,651.72,4.559
Pad is usable you know so that you know,653.7,4.259
what the user is working on,656.279,3.06
um so that you can talk about this and,657.959,3.781
this is passed to the chat bot every,659.339,5.101
time so whatever the conversation is,661.74,5.94
like that that is in the past,664.44,4.44
um I don't know if I'm explaining that,667.68,3.54
well anyways point being is that it see,668.88,3.959
it can see the code that you're working,671.22,5.359
on as long as you keep this updated,672.839,3.74
um okay,677.1,3.96
so that's the first hundred lines,678.0,5.22
um let's go ahead and,681.06,5.1
see where did it go step one simplify,683.22,5.58
um cool so act equals that and then,686.16,6.239
we'll say chunks equals uh let's see,688.8,5.159
chunk string,692.399,5.161
chunk string act all right so this this,693.959,5.701
basically creates the chunks that we're,697.56,4.2
going to be working on,699.66,2.76
um,701.76,3.48
and that is that,702.42,5.22
let's see I'm gonna need to watch this,705.24,4.26
because,707.64,3.24
um well here let me just show you,709.5,2.88
something real quick so the first,710.88,3.48
hundred lines,712.38,5.579
if we take that over to,714.36,5.76
playground and,717.959,4.681
paste it in here you see that 100 lines,720.12,5.82
is about four thousand tokens so that,722.64,4.98
would obviously be too big for text,725.94,6.54
davinci03 but with with gpt4 we have 8,727.62,7.92
000 tokens and then with 3.5 16k we have,732.48,5.22
16 000 tokens so we could actually,735.54,4.979
probably do more but the thing is is,737.7,6.54
that again 3.5 turbo is actually you,740.519,5.641
know what let's try it real quick heck,744.24,4.86
with it heck with it while we're here,746.16,5.64
let's go back over to chat,749.1,4.799
let's give it,751.8,3.659
this chunk,753.899,3.481
because if it can if I can do it faster,755.459,5.461
and cheaper why not why not,757.38,5.399
um and then we need this is our system,760.92,4.08
message we need to make sure that we're,762.779,7.141
going to use 3.5 turbo 16k so that it's,765.0,6.3
got enough space,769.92,3.719
okay turn the temperature down we do not,771.3,4.14
want you to be creative thank you very,773.639,3.241
much,775.44,3.48
um all right and then let's see how well,776.88,5.0
it summarizes it,778.92,2.96
um yeah so it's not this summarization,782.1,7.4
is not nearly as elegant or as concise,784.8,4.7
um but just glancing at it it's okay,789.839,6.24
I think I'm gonna stick with with four,794.579,4.081
but I'm glad I did this side by side so,796.079,5.281
the the the gpt4 it's a much more,798.66,5.7
elegant summary and it's also less wordy,801.36,4.86
even though so it's not quite as fast,804.36,4.62
and it's not quite as cheap because we,806.22,4.14
need to capture the essence of something,808.98,3.299
that's really important let's use the,810.36,3.3
smarter model,812.279,4.321
uh now that being said again 3.5 turbo,813.66,7.14
uh the the June 13th one plenty plenty,816.6,6.479
good for a lot of things,820.8,4.86
Okay cool so I'm glad I ran that test,823.079,4.861
where are we what am I doing,825.66,4.2
I don't know I'm lost,827.94,3.78
I'm always lost,829.86,4.56
there's this meme it's like a pug in a,831.72,5.7
field with a wizard hat and it says the,834.42,5.159
the the meme is like not all who wander,837.42,6.06
are lost except Dave is lost as and,839.579,5.94
that's me I am the pug with a wizard hat,843.48,4.26
does that make any sense I need more,845.519,3.961
coffee,847.74,3.899
oh,849.48,3.78
okay all right I hope you find this,851.639,3.121
entertaining because I just woke up and,853.26,2.94
I was like my brain's like hey we should,854.76,3.36
work on this,856.2,4.02
all right now we're back here all right,858.12,4.14
we've got our chunks so now what we need,860.22,3.9
to do so this is this is where it gets,862.26,4.5
fun so uh,864.12,6.06
four Chunk in chunks this is super,866.76,5.94
straightforward uh let's see,870.18,4.5
um messages,872.7,6.42
equals uh this is going to be a list and,874.68,6.24
then the,879.12,5.64
um roll will be whoops,880.92,5.34
user,884.76,4.879
and the content,886.26,3.379
wow I know how to type I promise it's,890.399,5.521
just getting it from my head to the to,893.459,4.081
the hands on the keyboard to the screen,895.92,4.44
that's too many steps right now,897.54,6.18
content equals chunk okay cool,900.36,5.279
so basically what we're doing right here,903.72,4.32
is we're setting it up to look exactly,905.639,5.341
like this so I'm I'm putting the Chunk,908.04,4.739
in here and then we also need to load,910.98,3.659
the system message so what we're going,912.779,3.3
to do is we're going to do,914.639,5.521
um system equals open file uh system,916.079,6.901
underscore summarize so this is this is,920.16,4.619
the instructions,922.98,3.299
um dot text,924.779,3.36
so system message is the new prompt,926.279,3.18
engineering,928.139,4.26
um all right cool so then we'll do uh,929.459,5.301
roll,932.399,2.361
system,935.16,5.119
and content,938.04,5.039
is system,940.279,4.481
all right so now we're ready to pass it,943.079,4.56
up to our model right here so then we'll,944.76,6.78
just do response equals chat bot,947.639,5.221
messages,951.54,3.9
ta-da as that simple,952.86,4.74
um so it's ready and so this response is,955.44,4.38
going to be the summary,957.6,4.2
and so now I need to figure out how I,959.82,4.139
want to save this save these summaries,961.8,5.099
out in a way that makes sense,963.959,4.44
um,966.899,4.401
let's see,968.399,2.901
let me ask my chatbot assistant okay,973.079,4.021
cool,975.24,5.94
great uh now I want to pass a list of,977.1,8.099
strings to a function and have that,981.18,9.839
function save them to a specified folder,985.199,7.58
as a parameter,991.019,8.541
and I want it to serialize those files,992.779,10.06
like so,999.56,5.219
um so it'll be like,1002.839,6.901
file001 dot text file zero zero zero two,1004.779,7.901
dot text and so on,1009.74,3.719
um,1012.68,5.459
let's see you should see the script I'm,1013.459,8.481
working on all right cool,1018.139,3.801
so then we're gonna we're gonna have um,1022.339,6.84
summaries equals list and then summaries,1025.339,6.141
dot append,1029.179,4.981
response so now basically we're gonna,1031.48,5.26
have the summaries all in order and what,1034.16,3.72
I'm going to do is I'm going to copy and,1036.74,3.599
paste this into my scratch Pad so that,1037.88,3.959
the chat bot knows what I'm working on,1040.339,3.781
and I'll show you how that it says oh,1041.839,5.58
okay cool I probably should have told it,1044.12,4.5
that I'm going to do that with the,1047.419,2.76
summaries it's usually pretty good at,1048.62,4.2
making the inference okay cool,1050.179,7.201
um let's see def save strings if OS path,1052.82,6.599
exists or if it doesn't exist make it,1057.38,3.96
otherwise enumerate strings,1059.419,4.021
so as long as I've been using python I'm,1061.34,4.5
not entirely sure how enumerate Works,1063.44,5.18
maybe I should ask,1065.84,2.78
all right but let's go over here,1068.72,6.86
let's go back to my script,1072.14,3.44
save strings cool,1077.66,4.08
um,1080.6,6.06
file I 05 interesting,1081.74,6.299
yeah see I'm not even sure how that,1086.66,2.759
works,1088.039,3.061
fantastic let's see if it let's see if,1089.419,3.721
it works though,1091.1,4.199
um the function first checks that the,1093.14,4.62
specified folder exists excellent then,1095.299,5.521
it uses it to make it then then iterates,1097.76,4.62
over the list of strings using enumerate,1100.82,3.599
which provides both the index and the,1102.38,3.659
value oh okay so that's what enumerate,1104.419,3.781
does the index is used to generate the,1106.039,6.181
file name with I colon 05 in the F,1108.2,6.06
string ensuring that the index is zero,1112.22,6.0
padded to five digits well that was easy,1114.26,5.52
cool,1118.22,3.3
um right and then it uses the right,1119.78,2.72
method,1121.52,6.0
uh yeah so instead,1122.5,6.82
um I want I was hoping it would use my,1127.52,4.86
save file but that's fine,1129.32,6.719
um so let's go to here and we're gonna,1132.38,6.84
do summaries,1136.039,5.161
and so instead,1139.22,4.5
we're going to assume that that is,1141.2,4.5
already there,1143.72,5.959
and then we are going to do,1145.7,8.599
file path equals,1149.679,4.62
um,1156.08,2.12
OS path join folder,1158.24,6.679
yeah that works,1161.66,3.259
and then,1173.12,4.74
we are going to do,1175.88,4.44
save file because this so my save file,1177.86,3.72
function,1180.32,4.68
um encodes it in utf-8 so just reuse,1181.58,5.219
something rather than do it here because,1185.0,4.32
you see this doesn't do it in utf-8 you,1186.799,3.541
want to make sure that you have a nice,1189.32,2.219
standardized thing that's what I'm doing,1190.34,2.52
here,1191.539,5.581
file path and S so that should be good,1192.86,7.86
fine all right so save strings,1197.12,7.679
um so then we do save strings uh we'll,1200.72,5.1
do,1204.799,2.76
summaries,1205.82,5.04
no I need to do yes summaries,1207.559,5.881
and then summaries,1210.86,5.1
and that should work uh now that being,1213.44,4.92
said I want to print the response as I,1215.96,4.76
get them,1218.36,2.36
um there we go and then,1222.74,3.26
response,1226.22,4.62
all right let's give this a whirl and,1228.2,5.58
see if it works,1230.84,6.9
CD what is this EU AI Act,1233.78,6.54
let's zoom in a little bit,1237.74,6.26
python stepo one,1240.32,3.68
so this will probably take a while,1245.72,4.86
because we've got three thousand line 3,1247.1,7.04
300 lines there we go cool,1250.58,3.56
hmm,1254.36,4.86
hold on oh right right right,1255.679,5.941
forgot about this okay so this is an,1259.22,3.72
important thing,1261.62,3.12
um what I started doing and I forgot to,1262.94,3.9
integrate this so what I started doing,1264.74,4.92
is I when I whenever I called a chat bot,1266.84,5.76
I return the text and the token use,1269.66,4.08
um so because this is actually really,1272.6,2.699
helpful if you just go ahead and return,1273.74,4.14
the tokens so that you know when to,1275.299,5.401
summarize it so instead what I need to,1277.88,5.64
do is response tokens,1280.7,4.08
um because I don't need to save I don't,1283.52,4.26
need to save a tuple out here,1284.78,5.639
but this otherwise looks good,1287.78,5.7
and then what I'm going to do is I'm,1290.419,5.301
going to go grab,1293.48,4.86
uh something to make the the printout,1295.72,5.199
look a little bit better,1298.34,4.64
um,1300.919,2.061
reflective journaling tool this one has,1303.5,5.88
this one has good pretty output,1306.02,7.58
so we're going to do formatted lines,1309.38,4.22
yep,1316.64,6.32
so take the formatted lines here,1318.86,4.1
and then we'll do,1328.58,5.339
rent do I knew my new line,1330.32,7.82
and formatted lines,1333.919,4.221
all right so that will that'll make it,1339.44,4.02
that'll make it prettier and then we'll,1341.299,3.961
also use,1343.46,3.3
um so this this is this is why I,1345.26,3.36
appreciate when people,1346.76,5.399
um uh add feedback is we added uh,1348.62,6.059
someone added the Halo which is a little,1352.159,5.161
like thing that shows that it's thinking,1354.679,4.201
so I'm going to add that real quick as,1357.32,3.739
well,1358.88,2.179
and the way that you use that is down,1362.96,5.12
here,1365.659,2.421
and so once once this is running I'll,1374.72,2.88
pause the video so you don't have to,1376.52,2.159
watch the whole thing so you're actually,1377.6,3.54
probably closer to the end if I had to,1378.679,4.581
guess,1381.14,2.12
oops spinner start,1386.12,3.32
so we'll say summarizing next chunk,1391.4,4.94
and then spinner stop,1398.36,3.02
okay so let's try this make sure it,1405.14,4.34
works,1407.12,2.36
summarizing next chunk hey look now it,1412.46,4.44
gives us some output better user,1414.74,4.74
experience it tells me what it's doing,1416.9,6.32
uh text wrap is not defined whoops,1419.48,5.579
what do you mean I need to import,1423.22,5.74
everything that you use import text wrap,1425.059,5.821
let's try that again,1428.96,3.959
third time's a charm all right and if,1430.88,3.48
this works then I'll go ahead and pause,1432.919,2.88
it and then we'll come back once the,1434.36,3.42
whole video is or once the whole thing,1435.799,3.061
is done,1437.78,2.779
uh why is it,1438.86,6.0
in hold on hold on that's not right,1440.559,7.061
why are you giving me this garbage,1444.86,6.02
that's not correct,1447.62,3.26
oh right I need to print out formatted,1452.6,5.9
text my bad,1455.059,3.441
don't give me this nonsense,1461.179,3.5
because basically what you want to do is,1464.96,3.599
you want to have it formatted in such a,1466.88,3.48
way there we go see so now we have this,1468.559,4.321
nice little Block it's easier to read,1470.36,3.72
um all right cool so this is working,1472.88,2.7
we'll watch it just a couple times you,1474.08,4.44
see how fast it is uh the the June 13th,1475.58,6.24
update is much much faster,1478.52,5.279
um let's see,1481.82,5.219
excellent excellent excellent so we will,1483.799,4.981
I'm gonna pause the video and we'll come,1487.039,4.921
back and uh and take a look at the final,1488.78,5.96
result cheers,1491.96,2.78
okay and we're back,1495.32,5.4
so after uh running it all here let me,1497.6,4.079
just show you what the results were,1500.72,2.939
we've got,1501.679,4.74
37 files,1503.659,5.64
um each with a summary and I wrote a,1506.419,4.26
really quick I hacked together a really,1509.299,4.321
quick script that uh merged them all and,1510.679,5.161
so the merged result is thirty four,1513.62,4.559
thousand uh characters long so we,1515.84,5.52
reduced it by a factor of ten to one uh,1518.179,5.521
just just shy of ten to one which when,1521.36,4.679
you plug that in over here that's six,1523.7,5.28
thousand tokens so we went from from,1526.039,4.561
sixty thousand or seventy thousand,1528.98,3.24
tokens I don't remember exactly to six,1530.6,3.66
thousand tokens one of the reasons that,1532.22,4.74
we had a more drastic reduction of token,1534.26,5.159
count is because we removed all the,1536.96,4.319
serial numbers and stuff because that's,1539.419,4.441
kind of Superfluous if you if you remove,1541.279,4.561
symbols and stuff like this these are,1543.86,4.62
high token uh things if it's ordinary,1545.84,5.64
words like Ai and each system these are,1548.48,6.12
easy for the model to tokenize so that,1551.48,5.46
means what we can do now is we can just,1554.6,3.959
take this whole thing,1556.94,5.82
and come over to chat and say um let's,1558.559,5.22
see,1562.76,6.72
uh uh main purpose uh here darn it,1563.779,9.421
believe that you are a uh chat bot,1569.48,7.22
tasked with uh discussing,1573.2,8.459
uh the following uh legislation with the,1576.7,7.24
user uh,1581.659,5.581
make sure to well hell that I mean,1583.94,5.839
that's all it is uh,1587.24,5.84
use the legislation,1589.779,7.061
summary below to conduct the,1593.08,4.839
conversation,1596.84,4.219
and that's it and so then we say uh,1597.919,6.0
legislation summary,1601.059,5.261
and we'll copy paste it here,1603.919,5.221
and so then we move over to we could do,1606.32,5.7
16k because that'll be that'll be faster,1609.14,4.08
um actually here let's try that real,1612.02,3.18
quick let's try that temperature there,1613.22,4.559
okay,1615.2,6.479
um give me a high level overview of this,1617.779,7.581
uh legislation,1621.679,3.681
after five years,1628.9,5.92
um how is it funded,1631.4,4.519
foreign,1634.82,3.0
let's see the funding and the,1635.919,3.581
implementation of this legislation comes,1637.82,2.7
from the multi-annual financial,1639.5,3.48
framework uh blah blah blah blah the,1640.52,4.019
financial impact includes allocation of,1642.98,3.42
resources,1644.539,4.561
um and having watch that go through how,1646.4,3.48
many,1649.1,1.64
um,1649.88,2.58
employees,1650.74,4.74
will it cost,1652.46,3.02
there we go 10 full-time equivalent,1656.299,3.781
employees,1658.22,5.939
all right so we can clear that out,1660.08,6.18
uh just because it ultimately is,1664.159,4.861
Superfluous so some of the questions,1666.26,5.899
um uh let's see will this legislation,1669.02,8.539
hamper research and innovation,1672.159,5.4
let's see the legislation aims to,1678.44,3.78
balance the benefits of AI research and,1680.659,2.88
Innovation with the need for safety and,1682.22,2.939
respect for fundamental rights it,1683.539,3.12
includes measures to support Innovation,1685.159,4.14
such as AI regulatory sandboxes and,1686.659,4.741
reducing the regulatory burden on smes,1689.299,4.5
and startups so that smes are going to,1691.4,4.08
be subject matter experts like you and I,1693.799,3.541
as well as startups,1695.48,3.78
the goal is to provide legal certainty,1697.34,3.959
and prevent Market fragmentation which,1699.26,4.26
can actually Foster Innovation by,1701.299,3.721
creating a Level Playing Field for AI,1703.52,3.0
developers however it imposes certain,1705.02,3.48
requirements and obligations on,1706.52,3.72
high-risk AI,1708.5,5.659
um how does this legislation Define,1710.24,7.88
higher risk AI,1714.159,3.961
uh let's see the this legislation,1719.62,5.38
categorizes AI systems into three risk,1722.179,4.74
levels unacceptable high and low or,1725.0,4.559
minimal high risk AI is based on their,1726.919,4.801
intended purpose and usage while the,1729.559,4.141
specific criteria may vary high risk AI,1731.72,4.079
generally include those use the safety,1733.7,3.599
components of products covered by Union,1735.799,4.38
harmonization legislation as well as AI,1737.299,4.141
used in critical infrastructure,1740.179,3.0
management education vocational training,1741.44,4.38
employment and workers management these,1743.179,4.201
high-risk AI systems are subject to,1745.82,3.0
requirements,1747.38,2.34
um,1748.82,2.88
what are unacceptable,1749.72,4.98
types,1751.7,3.0
the legislation prohibits certain AI,1755.12,3.299
practices that are considered,1757.34,3.059
unacceptable due to their potential to,1758.419,4.441
violate fundamental rights while the,1760.399,4.16
specific types may vary some examples,1762.86,4.439
include manipulate human behavior,1764.559,4.6
exploit vulnerabilities or enable social,1767.299,4.021
scoring by public authorities right so,1769.159,4.5
as I mentioned uh social credit system,1771.32,4.94
these practices are deemed harmful,1773.659,5.64
Freedom dignity democracy quality and,1776.26,4.919
fundamental rights okay cool,1779.299,5.901
does this legislation address,1781.179,7.961
existential risk at all,1785.2,9.54
um does it address autonomous AI,1789.14,5.6
the legislation does not explicitly,1795.02,4.32
address existential risk or autonomous,1797.299,6.781
AI systems wow okay now one thing that I,1799.34,6.42
will say is that if you look at the,1804.08,3.92
legislation it does talk about like,1805.76,5.46
basically human supervision,1808.0,6.94
um yeah so like AI regulatory sandbox by,1811.22,7.92
one or more member stages no like here,1814.94,6.839
um it where was it it talked about not,1819.14,5.1
supervision of the sandboxes but like AI,1821.779,4.38
systems basically need to be steerable,1824.24,3.9
it's in there somewhere,1826.159,5.88
um okay but let's see it does not,1828.14,5.34
specifically address a potential risk,1832.039,2.88
associated with highly autonomous or,1833.48,3.24
super intelligent AI systems that could,1834.919,3.36
pose existential threats however the,1836.72,3.059
legislation's emphasis on risk,1838.279,3.301
management human oversight oh that's,1839.779,3.181
what it was oversight not supervision,1841.58,3.0
and robustness requirements for,1842.96,3.5
high-risk AI systems indirectly,1844.58,3.839
contributes to addressing the safety,1846.46,3.88
concerns associated with AI Technologies,1848.419,4.14
it is important to note that addressing,1850.34,3.959
existential risks related to AI may,1852.559,3.6
require further research discussions on,1854.299,3.6
potentially additional regular,1856.159,4.921
regulations in the future okay cool,1857.899,6.541
so there you have it this file is saved,1861.08,6.24
as merged.txt,1864.44,4.88
um,1867.32,2.0
let me go ahead and send that up to git,1869.96,5.4
so you can all use it uh let's see get,1872.6,8.059
status get add get commit am all done,1875.36,8.28
get push all right so there you have it,1880.659,6.88
you have a uh a much much condensed,1883.64,7.38
version of of the EU act here and merged,1887.539,6.301
you can plug it into the playground and,1891.02,5.519
ask questions to your heart's content if,1893.84,4.5
you have access to it but otherwise you,1896.539,4.02
can at least just read it here,1898.34,3.42
um yeah I think we're all done thanks,1900.559,3.681
for watching,1901.76,2.48