text,start,duration
all right hello everybody David Shapiro,6.96,5.099
here with the gato framework project,9.66,4.92
team we've got about a quarter of the,12.059,4.921
team here and basically all we're going,14.58,5.88
to do is just do a debrief of the Senate,16.98,6.12
hearing that happened yesterday uh with,20.46,5.52
Sam Altman and uh Mrs Montgomery from,23.1,6.54
IBM as well as Gary Marcus so many of us,25.98,5.4
watch the whole thing or most of it we,29.64,3.119
were chatting about it in real time,31.38,4.44
because we're those kind of nerds so uh,32.759,5.521
with all that being said this is a to us,35.82,5.22
it represents a very pivotal moment uh,38.28,5.7
in terms of moving the Overton window uh,41.04,4.92
talking about the not just the,43.98,3.899
existential risk of AI but the social,45.96,5.939
risk to AI uh or the the AI risk to,47.879,6.241
society so on and so forth so anyways,51.899,4.561
we've got a pretty robust discussion uh,54.12,3.66
tonight it's not going to be a debate,56.46,4.14
format but we do have uh some help with,57.78,4.02
moderation so we're going to work,60.6,3.66
through some some questions and uh and,61.8,4.92
answers and so yeah that's that's where,64.26,4.2
we're starting,66.72,5.939
um I'll introduce my uh moderator uh uh,68.46,6.12
our community leader Community manager,72.659,3.121
Nathan,74.58,3.06
um Nathan if you want to jump in and,75.78,4.339
introduce yourself,77.64,2.479
hey everyone got a community larger,80.36,7.0
world uh my name is Nathan Lannon and I,83.759,5.461
am a community manager for the gato,87.36,4.68
projects uh right now I'm working on,89.22,4.14
building the website with some,92.04,4.98
passionate folks and also doing a AI,93.36,5.64
reskilling curriculum that I hope to,97.02,4.559
eventually put out to the public so yeah,99.0,4.02
thrilled to be here I will be playing,101.579,3.241
the role of helping people get on and,103.02,3.959
off stage and making sure things go,104.82,4.439
smoothly thank you Dave,106.979,4.861
excellent thank you so much Nathan,109.259,4.32
um all right so with that being said,111.84,4.08
this is also going to be a pretty casual,113.579,4.08
conversation,115.92,3.54
um the only primary rule is we're going,117.659,3.841
to try and talk one at a time,119.46,4.26
um so with all that being said uh Nathan,121.5,4.979
what is our first uh question to lead us,123.72,5.519
off uh Prime that prime the pump so to,126.479,4.381
speak,129.239,4.08
yeah so,130.86,4.98
yesterday as we know was the hearing,133.319,3.961
which was,135.84,3.78
huge for the AI Community opening up,137.28,5.039
that Overton window crack and just to,139.62,6.479
kind of start off on a positive note I,142.319,7.741
think it would be nice to reflect on,146.099,5.881
hope that that might have generated for,150.06,5.16
us and general thoughts about how that,151.98,5.46
might have been a good thing for us so,155.22,4.019
anybody want to jump in on that raise a,157.44,3.92
hand,159.239,2.121
John why don't you hop up,161.599,6.041
yeah absolutely so um I think one of the,164.28,7.38
biggest takeaways from the the Senate,167.64,7.44
hearing was how educated the Senators,171.66,7.26
actually were about the the situation,175.08,6.9
and the risks I think you know I think a,178.92,5.22
lot of the the Senators said some things,181.98,5.46
or asked some questions that that uh you,184.14,5.28
know showed that they weren't experts in,187.44,5.76
the field but they were they were way,189.42,5.94
more knowledgeable about the situation,193.2,6.66
than I think anybody really anticipated,195.36,7.2
um it yeah we we expected the moderator,199.86,4.44
to be you know pretty knowledgeable,202.56,3.899
because he's worked in this field before,204.3,3.54
and,206.459,4.261
um advocated for,207.84,4.86
um algorithmic,210.72,7.26
uh bias protections and social media but,212.7,6.899
um you know to see some of these other,217.98,4.5
Senators that we don't hear from all the,219.599,6.481
time actually get up and and ask,222.48,6.899
somewhat you know somewhat Salient,226.08,6.54
questions about the actual issue and,229.379,4.86
show that they did know what was going,232.62,4.32
on it leaves a lot of Hope for,234.239,6.241
um having reasonable and intelligent,236.94,8.299
regulation in in the future,240.48,4.759
excellent thank you John uh Andy I saw,246.36,6.98
you have a hand raised watch come on up,249.959,3.381
uh just building on what John said I was,256.639,5.38
also encouraged I I had a dismal outlook,259.799,4.68
on what might be pretty uninformed,262.019,5.041
hearing uh and also something that was,264.479,5.481
heavily partisan or rooted in kind of,267.06,5.1
dogmatic rhetoric which is the way that,269.96,3.28
you know a lot of these kind of,272.16,3.66
publicized hearings will go uh but it,273.24,4.019
didn't seem to me that too many people,275.82,4.08
showed up with an overt agenda there,277.259,4.141
were more people in the room there to,279.9,3.6
learn and I also thought that it was,281.4,6.42
interesting to see and hear from uh,283.5,6.6
Republicans who were interested in,287.82,4.68
setting up a bureaucracy a bureaucratic,290.1,4.74
mechanism and uh department or agency,292.5,4.259
specifically for this some of them,294.84,3.48
mentioned agencies someone as high as,296.759,3.541
cabinet level,298.32,4.14
um and Democrats questioning whether,300.3,4.02
expanding the bureaucracy was really,302.46,4.44
going to be a fitting choice for how to,304.32,4.68
cope with this uh those are,306.9,4.079
very different Tunes than we usually,309.0,3.72
hear from the two sides who kind of,310.979,4.381
default to being foreign against the,312.72,4.32
opposite of that so I thought that that,315.36,4.86
uh that represented,317.04,5.219
it suggested to me that a lot of these,320.22,5.699
people are concerned about this they are,322.259,6.061
literally personally interested in,325.919,3.961
what's going on and that they aren't,328.32,3.719
just trying to fit it into whatever box,329.88,3.84
They Carried into the room so I don't,332.039,4.261
think we get that too often uh in our in,333.72,4.56
our legislative debates and I didn't,336.3,3.36
think that it really came across as much,338.28,2.58
of the debate there were only a couple,339.66,3.42
kind of contentious moments and I think,340.86,4.32
all of that suggests that you know as,343.08,3.78
John said previously too that we've got,345.18,3.66
a a lot of fertile ground to work with,346.86,3.24
in government,348.84,3.78
and that some support and interest,350.1,5.159
really may come from uh,352.62,4.44
angles and actors that you may not have,355.259,3.241
expected to be interested at all or,357.06,3.9
might have expected to be opposition to,358.5,4.199
whatever you're trying to do,360.96,3.84
so yeah I think the the hearing was,362.699,5.301
pretty encouraging overall,364.8,3.2
totally Andy so like hearing we saw you,368.34,6.12
know bipartisan hand holding yesterday,372.12,4.62
in alignment on you know a super,374.46,4.86
important topic and just that was,376.74,4.32
incredibly heartening to see I agree,379.32,3.24
that was one thing I was going to speed,381.06,5.359
up on as well uh Dave please go ahead,382.56,6.9
yeah so first I'll I'll Echo uh the,386.419,4.78
sentiments of of what everyone else has,389.46,3.6
already said,391.199,6.12
um but uh what I'll also add is uh and,393.06,6.54
as I mentioned in the introduction,397.319,4.5
um moving the Overton window I think is,399.6,4.74
is going to be kind of the biggest thing,401.819,5.341
and I was so excited that that Senators,404.34,4.28
you know,407.16,3.96
particularly in America,408.62,5.38
um even some of the older gentlemen and,411.12,6.54
gentle women uh were asking very pointed,414.0,7.08
questions uh not just on the topics of,417.66,5.58
like oh hey comparing it to social media,421.08,3.3
which they did compare it to social,423.24,2.64
media a lot because that's the only,424.38,3.319
frame of reference that we've got,425.88,4.62
basically uh for disruptive Technologies,427.699,5.801
at least in in recent history uh but,430.5,5.1
they they also said like you know what,433.5,4.08
are the existential risks,435.6,3.659
um and I'm I'm glad that they had a,437.58,4.32
witness okay you know because more often,439.259,5.401
than not these these uh Congress uh,441.9,5.579
Congressional hearings they have uh just,444.66,4.68
corporate insiders right everyone Pro,447.479,4.021
well I don't know everyone does but big,449.34,4.38
tobacco right where it was just like a,451.5,4.74
panel of like eight uh like tobacco,453.72,4.62
execs and they're like is this addictive,456.24,4.14
and the tobacco exec's all unanimously,458.34,3.24
said no it's not addictive and,460.38,3.0
everyone's like well this is completely,461.58,5.22
pointless right but here they had two um,463.38,5.879
they had two tech industry uh insiders,466.8,6.6
and then Gary Marcus who is uh he's he's,469.259,7.261
not a tech Insider he's a researcher,473.4,4.68
um so they had that to counterbalance it,476.52,3.54
which I think gave the whole hearing a,478.08,3.959
lot more credibility,480.06,5.4
um in terms of just reach and validity,482.039,7.021
but last point that I'm make is that I,485.46,7.32
was also impressed by uh by both Sam,489.06,6.96
Altman and Miss Montgomery's uh who,492.78,5.58
represented IBM their desire and,496.02,5.28
willingness to have regulations and to,498.36,4.679
and to speak up,501.3,2.94
um which that was a little bit,503.039,3.66
unexpected and of course we can we can,504.24,5.82
unpack some of the nuances later uh but,506.699,5.881
yeah so that's that's about it for me,510.06,3.659
day,512.58,4.5
thanks Dave and Ben I'd like to invite,513.719,5.601
you up,517.08,2.24
thanks Nathan just quickly my name is,519.659,4.981
Ben I'm a CTO and a tech diligence,521.94,5.82
advisor for on AI uh I'm helping with,524.64,5.879
some of the gato framework rating,527.76,5.579
the thing I really enjoyed besides the,530.519,4.861
thing I already mentioned about this,533.339,4.141
hearing was how incredibly practical it,535.38,4.8
was so like Senators asked for a very,537.48,5.039
specific ideas on how to do oversight,540.18,4.56
and and really drove home like not just,542.519,4.26
high level Concepts but very practical,544.74,4.38
ideas on how to do it,546.779,4.021
um they did things like highlight the,549.12,4.38
need for Better Business models around,550.8,5.159
news organizations or other other,553.5,4.68
monetization sources so just the level,555.959,3.961
of practicality that they started with,558.18,5.0
was really heartening for me,559.92,3.26
excellent so,564.42,4.979
I'm hearing you know all around we heard,566.399,6.301
uh many things that we liked,569.399,5.94
now I'm curious before we dive into,572.7,6.18
things that we liked less uh is there,575.339,5.701
any things that we wish we had heard but,578.88,4.5
we didn't hear them say,581.04,3.84
um Richard I know you had a hand up,583.38,3.48
before I asked this question so you,584.88,3.18
don't feel free to respond to the prior,586.86,4.14
but then let's kick over to like what,588.06,5.219
could have been said but wasn't the,591.0,4.94
unsaid things,593.279,2.661
uh yeah hi I'm Richard I'm basically,597.26,5.259
just sort of a dude who hangs out in a,600.6,3.54
workshop building weird,602.519,5.281
um but uh in terms of just the the last,604.14,5.46
point was just uh I wanted to also touch,607.8,2.76
on the fact that it was just really,609.6,3.679
surreal to see like heads of like,610.56,5.76
massive corporations that could if they,613.279,5.441
wanted to wield this as a weapon against,616.32,5.28
society and control a huge swath of,618.72,4.38
people,621.6,3.78
um go to the government and specifically,623.1,4.56
ask them to regulate themselves like,625.38,4.5
that never happens I've never seen,627.66,3.9
anything like that before so,629.88,3.3
um it was very surreal to watch and also,631.56,2.339
just the fact that they were talking,633.18,2.7
about AI in Congress was just like what,633.899,4.201
what what leg of the trousers of time,635.88,3.84
have I gone down,638.1,4.16
um but yeah,639.72,2.54
excellent all right yeah absolutely,645.839,5.701
um so I was learning would anybody have,649.079,4.921
any other thoughts on um things they,651.54,5.22
liked before we move over to uh things,654.0,5.7
that we wish we had heard more of uh,656.76,5.48
John I see you have a hand up,659.7,5.699
yeah so um one of the things that I,662.24,5.74
think is really going to need a lot of,665.399,5.281
discussion and will probably come up in,667.98,4.5
a later hearing but I'd like to have,670.68,6.18
heard more about it was the the economic,672.48,7.38
impacts that these these things are,676.86,6.659
going to have because the economy is an,679.86,6.78
existential existential threat maybe not,683.519,6.181
like Extinction of humanity level but it,686.64,4.02
is,689.7,4.74
um probably one of the the first impacts,690.66,5.64
that uh negative impacts that we're,694.44,4.44
going to see from Ai and I don't think,696.3,5.64
that and it may be positive too but I,698.88,4.88
don't think that you know everybody,701.94,4.92
wants to you know ask to worry about,703.76,6.699
jobs but I don't think that jobs is the,706.86,5.46
only impact it's going to have on the,710.459,4.32
economy it's going to shift the way that,712.32,6.12
money moves in a global fashion and it,714.779,6.421
is going to have Global impacts not just,718.44,7.88
on jobs on all industry as well,721.2,5.12
absolutely yeah we're gearing up to see,728.7,7.56
major societal shifts and they were a,731.899,7.0
little light-handed with uh raising,736.26,4.8
awareness of some of those things and so,738.899,5.481
I see you having it up there,741.06,3.32
yeah well I I wanted to touch on the,744.54,6.9
things that uh weren't mentioned but I,748.5,5.16
think data basically covered it I wanted,751.44,3.899
to touch on the fact that they said that,753.66,4.2
this was going to create new jobs I'm,755.339,4.981
like sure it'll create new jobs but it's,757.86,4.919
not going to be anywhere near enough of,760.32,4.38
the jobs that it's going to displace and,762.779,3.661
especially it's not going to create new,764.7,4.439
ones in that much of a short time we're,766.44,5.519
we're we're seeing huge disruptions in,769.139,5.041
the next five years and that's being,771.959,5.94
incredibly optimistic and that's not a,774.18,6.779
um uh a nif that is that is guaranteed,777.899,6.06
the real question is what are the,780.959,4.68
fundamental shifts that are going to,783.959,2.82
happen and how are we going to address,785.639,3.721
them because Congress is still thinking,786.779,5.761
in in the old model right where where,789.36,6.659
you still need humans working eight,792.54,6.419
hours a day four uh five hours sorry,796.019,5.341
five days a week and just just just,798.959,5.401
grind right AI is gonna automate a lot,801.36,6.599
of that grind and we're not gonna find,804.36,5.64
enough uh,807.959,5.041
what's the word uh busy work,810.0,5.579
to keep ourselves busy right we should,813.0,5.899
we should focus on more on on liberating,815.579,6.421
uh that time for humans to do better,818.899,4.421
things right,822.0,4.88
and well that's basically it,823.32,3.56
they cancel,827.88,2.22
um,829.44,1.92
Dave why don't you pick up the,830.1,3.56
conversation,831.36,2.3
sure sure yeah I think uh and then I'll,834.36,5.099
wait back to Ben yeah sorry I gotcha no,836.7,3.84
worries,839.459,4.201
um but yeah so for me certainly agree,840.54,6.539
that uh the economy Ubi that sort of,843.66,5.04
stuff the displacement they touched on,847.079,3.361
displacement but then kind of glossed,848.7,3.18
over it,850.44,3.959
um maybe maybe they had had a out of,851.88,4.98
band you know conversation like let's,854.399,4.56
save this for another thing I don't know,856.86,4.8
but one thing that that was addressed I,858.959,4.44
think Gary addressed this when he talked,861.66,4.679
about Auto GPT and Chaos GPT,863.399,5.761
um was and that and that nobody uh else,866.339,4.141
addressed,869.16,3.239
um at least none of the witnesses was,870.48,4.5
autonomous AI right,872.399,5.641
um both Montgomery and Altman uh,874.98,5.099
fervently said this is just a tool this,878.04,3.84
is not a creature don't anthropomorphize,880.079,4.5
it and it's like yeah but then when Gary,881.88,5.16
said hey like people are making this,884.579,4.801
stuff semi-autonomous today it doesn't,887.04,4.68
matter how ineffective it is,889.38,4.92
um or if it has flaws it's today,891.72,4.2
right,894.3,3.839
um yeah and someone in chat just,895.92,4.38
mentioned that um that yeah chat gbt,898.139,5.341
they released plugins right uh pretty,900.3,4.5
quickly,903.48,2.76
um and then one thing that I did notice,904.8,3.18
and this was just speculation on my,906.24,4.14
point outside of that was that uh as,907.98,4.38
soon as people started creating like,910.38,4.98
chaos GPT they slow rolled uh rolling,912.36,5.339
out plugins to more people,915.36,4.979
um so maybe maybe open AI tapped the,917.699,5.94
brakes uh in response to chaos GPT,920.339,5.881
um but you know I I obviously don't have,923.639,4.081
an inside line that's just speculation,926.22,3.0
on my point but I would have done that,927.72,4.559
if I saw someone use chat gbt to create,929.22,7.14
ksgpt but yeah so autonomous AI was a,932.279,6.68
huge thing missing Gary tried to push it,936.36,6.0
several Senators did ask like hey uh,938.959,4.961
what was it uh I think it was Lindsey,942.36,3.719
Graham near the end he's like can you,943.92,5.219
use this to give a drone the ability to,946.079,5.341
autonomously Target someone and then go,949.139,4.921
kill them and of course Sam tried to you,951.42,5.94
know use the corporate sanitized speak,954.06,5.1
and then Lindsey Graham to his credit,957.36,5.159
said that's a yes or no question and and,959.16,6.239
Sam's like yeah it is possible right so,962.519,4.801
he he did admit that which was which was,965.399,4.981
Brave but you know the thing is is like,967.32,4.98
um when I first got access to chat GPT,970.38,4.199
to the API I went in and showed that you,972.3,3.539
can just give it instructions like,974.579,3.0
you're controlling a drone right and,975.839,3.06
it's like sure I'm controlling a drone,977.579,3.721
now right it doesn't care,978.899,4.62
um so yeah the autonomous AI was the,981.3,4.26
biggest thing missing in in my eyes and,983.519,4.981
I seed the floor,985.56,5.88
Ben please pick it up,988.5,5.279
sounds great I just wanted to start by,991.44,4.199
saying I'd love to talk through the jobs,993.779,4.141
thing I think saying it's guaranteed,995.639,3.961
that we're going to lose so many jobs is,997.92,3.719
a pretty aggressive way to go and I I,999.6,4.2
think it's worth debating a bit,1001.639,4.741
um I certainly think differently in,1003.8,5.459
terms of what was missed for me honestly,1006.38,5.579
how bad things already were before AI,1009.259,4.921
was released so like misinformation has,1011.959,4.5
been a problem forever we have all sorts,1014.18,4.44
of societal problems around,1016.459,5.761
global warming energy crisis things like,1018.62,5.939
this and just the opportunities to,1022.22,5.52
leverage AI yeah it's gonna make some,1024.559,5.4
things worse for sure but it is also an,1027.74,4.38
opportunity to solve help solve some,1029.959,3.661
things that were already problems and we,1032.12,3.78
were already struggling with,1033.62,4.62
um I I would love to see people talk a,1035.9,3.72
little bit more about those,1038.24,3.98
opportunities,1039.62,2.6
thank you Ben uh very Sierra,1044.12,4.339
one of the things I think was missing,1048.86,5.46
from it was the competitive angle which,1050.96,5.64
says that yeah that's great America is,1054.32,4.08
not the only world the place in the,1056.6,3.42
world that's trying to develop an AI,1058.4,4.139
right and the reality is that if you tap,1060.02,3.899
the brakes too much then what you'll do,1062.539,2.701
is lose the lead,1063.919,3.961
and losing the lead to a country that,1065.24,6.02
has less than Stellar moral boundaries,1067.88,6.419
might in fact be counterproductive to to,1071.26,5.08
what's going on so a lot of this stuff,1074.299,4.021
is very interesting but the comment I,1076.34,3.0
would make is the horse has already,1078.32,3.18
bolted and unless,1079.34,4.74
um we keep up with it and keep ahead of,1081.5,3.72
it then,1084.08,4.099
um the outcome could be even worse than,1085.22,6.98
displacement from losing a few jobs,1088.179,4.021
I hear you but it sounds like uh you're,1092.9,5.04
referring to the mall of dynamics that,1096.14,4.26
are very much at play right now we could,1097.94,6.06
easily end up in a race to the bottom uh,1100.4,6.3
John why don't you go ahead yeah I,1104.0,4.74
wanted to come back to,1106.7,8.16
um uh Mr Marcus's comment about uh,1108.74,11.4
about uh Auto GPT what what he said was,1114.86,8.22
was that because,1120.14,6.38
because open AI released the plugins,1123.08,7.14
that enabled Auto GPT and that was,1126.52,7.3
actually never the case so Auto GPT and,1130.22,7.74
Chaos GPT both pre-date the plug-in,1133.82,7.56
release and they both directly leverage,1137.96,6.3
only the GPT API,1141.38,5.12
so you know it's a little bit,1144.26,5.82
disconcerting to to see somebody who who,1146.5,6.82
is you know supposed to be a a,1150.08,5.94
researcher in this get something like,1153.32,5.099
that so egregiously wrong and I'm really,1156.02,4.5
surprised that Sam didn't didn't call,1158.419,6.061
him out for that when he uh when he had,1160.52,9.0
just called Sam out for you know not not,1164.48,9.24
saying what actually scared him about AI,1169.52,6.659
yeah we're talking about you know some,1173.72,6.24
of the the kind of darker side of what,1176.179,5.88
might happen with AI,1179.96,4.98
and you know bad research and bad,1182.059,6.24
information to that that's used to,1184.94,5.4
persuade governments to create,1188.299,5.161
regulations is another kind of,1190.34,6.56
misinformation concern,1193.46,3.44
absolutely thank you John uh Richard go,1198.88,4.62
ahead,1201.38,2.12
excuse me uh yeah the biggest thing uh,1205.58,6.02
like I think that that the they were,1209.48,4.86
more talk like he was,1211.6,4.24
it was more like in reference to the,1214.34,2.82
fact that like,1215.84,2.04
um,1217.16,2.82
the plugins would give the opportunity,1217.88,4.62
for people to build more things like,1219.98,4.38
um chaos GPT and stuff,1222.5,2.82
um,1224.36,2.52
in like because if you give them the,1225.32,3.3
tools to be able to just like plug them,1226.88,3.299
together like Lego then it's way easier,1228.62,3.059
than like a developer who actually went,1230.179,4.921
in and built the API backend for that,1231.679,6.301
um which is I mean like the thing is,1235.1,4.86
just humans are going to be humans is it,1237.98,4.14
the one like humans are amazing and,1239.96,4.44
beautiful and do incredible things but,1242.12,3.72
at the same time we're terrible and,1244.4,4.08
awful and do horrible things,1245.84,5.459
um so we have to sort of like have the,1248.48,4.98
contingency plan to make sure that that,1251.299,3.24
um,1253.46,3.3
that that is uh like something that we,1254.539,4.081
can manage because like malaligned AI is,1256.76,2.94
going to happen because these things are,1258.62,4.919
like infinitely manipulatable like they,1259.7,5.219
all work off of natural language and,1263.539,3.541
like you can talk to them like if you're,1264.919,3.961
just nice to it you'll find you get,1267.08,3.24
better results just being polite so,1268.88,2.4
thank you,1270.32,2.16
um talking to it a little bit in between,1271.28,2.94
prompts and stuff,1272.48,4.02
um you get way better results and then,1274.22,5.339
like if if you know how to talk to them,1276.5,4.679
correctly you can get them to reveal all,1279.559,2.521
kinds of stuff that they're not supposed,1281.179,3.261
to so,1282.08,2.36
absolutely yeah and the uh,1286.46,4.74
the Senators to their credit were,1289.46,4.199
definitely well rehearsed you know they,1291.2,5.339
had heard language around Jailbreak jail,1293.659,5.101
breaking and stuff like that heartening,1296.539,4.14
to see them using the language of our,1298.76,4.26
inner alignment communities,1300.679,3.12
um,1303.02,2.06
so,1303.799,4.321
with that aside uh Ansel at least of,1305.08,4.42
them,1308.12,3.299
all right I wanted to touch on the,1309.5,5.4
comment that John said about it being in,1311.419,6.481
the API not a plug-in I agree trust me I,1314.9,3.96
agree,1317.9,3.42
um the thing is we're in the in the in,1318.86,4.199
the industry right we're we're in the,1321.32,4.26
know we know what an API is and we know,1323.059,4.021
what a plug-in is we know the difference,1325.58,4.74
right most people won't,1327.08,5.52
and especially for old people in,1330.32,4.14
Congress you tell them an API and,1332.6,4.26
they're like what is an API so when when,1334.46,3.959
they say a plug-in it basically,1336.86,3.36
functions the same way it's it's a way,1338.419,4.201
to connect to the model right regardless,1340.22,3.839
of if it's gone through an API or a,1342.62,2.52
plug-in you're connecting to the model,1344.059,3.12
so that's what Congress needed to know,1345.14,6.06
right so I think that's not really,1347.179,6.541
that important that he didn't make that,1351.2,4.859
distinction between oh plug in API,1353.72,4.02
because at the end of the day it's just,1356.059,4.461
connecting to the model,1357.74,2.78
absolutely thank you Ansel and uh Ben,1362.299,5.461
I'll hand it to you and then take it,1366.14,3.24
back to raise a new question to everyone,1367.76,4.02
so go ahead,1369.38,4.38
sounds good the last thing I wanted to,1371.78,4.5
mention is while it was great to hear,1373.76,5.46
them mention Auto GPT there there wasn't,1376.28,5.16
a lot of talk about open source there's,1379.22,4.56
been a lot of AI news about how open,1381.44,4.38
source may push the envelope of what's,1383.78,4.56
uh capable and regulating open source,1385.82,3.96
versus companies may be very different,1388.34,3.66
so at some point I'd love to see them,1389.78,4.08
jump into the role of Open Source and,1392.0,4.76
how to regulate that,1393.86,2.9
thank you Ben and uh thank you everybody,1398.179,4.561
for kind of raising to the surface some,1400.1,4.199
of the things that we would have liked,1402.74,4.74
to hear but we did not hear from the,1404.299,6.0
debate and are not debate but hearing,1407.48,5.699
and a positive thing there will be more,1410.299,5.161
of these so if you are listening perhaps,1413.179,3.601
these are things you bring to the table,1415.46,3.42
for the next round,1416.78,5.519
um now I'd like to discuss uh,1418.88,5.7
what are some of the things about the,1422.299,5.581
hearing that we found disappointing,1424.58,5.88
um things that we did not like so much,1427.88,5.22
um so got the community the floor is,1430.46,4.94
yours,1433.1,2.3
Ansel,1440.539,4.681
well I'll start the thing I didn't like,1442.64,5.82
was they tried to focus more on creating,1445.22,6.72
a national agency instead of an,1448.46,6.3
international one when this technology,1451.94,5.4
is being used around the entire planet I,1454.76,4.08
think that doing this in an,1457.34,4.44
international level is is crucial,1458.84,4.44
right so that's one thing that I was,1461.78,3.12
disappointed like they they just,1463.28,3.54
mentioned it yes and they're like no no,1464.9,3.3
it's if we're going to do let's go,1466.82,3.06
National blah blah you know that's,1468.2,5.18
that's one thing I didn't like,1469.88,3.5
totally Ansel speaking here to one of,1473.9,6.24
the traditions of gato we like to,1476.179,6.961
act local but think Global so these,1480.14,5.7
issues affect all of us,1483.14,5.96
um Andy the floor is yours,1485.84,3.26
um one of the things I didn't like is,1491.12,3.179
that uh,1492.5,4.08
Sam Altman for reasons I think I,1494.299,4.021
probably understand but don't,1496.58,4.56
necessarily appreciate uh dodged a few,1498.32,4.56
questions we've mentioned it earlier,1501.14,4.08
they asked him to talk about his biggest,1502.88,4.799
fears and concerns,1505.22,3.06
um,1507.679,2.641
Sam has been very candid in public,1508.28,4.44
before about this with MIT presentations,1510.32,4.68
on the Lex Friedman podcast he's been,1512.72,4.02
one of the more forthcoming voices out,1515.0,3.36
there in the field as somebody who,1516.74,3.48
readily also says when he talks there's,1518.36,4.319
only so much I can tell you,1520.22,3.66
um and I thought that that was a little,1522.679,2.341
bit disappointing I thought he could,1523.88,2.7
have given a little bit more pointed,1525.02,4.74
answers I also was disappointed,1526.58,5.099
um that they dodged,1529.76,4.38
the jobs question when the you know the,1531.679,4.561
tech industry has just undergone a wave,1534.14,4.44
of layoffs that were largely attributed,1536.24,4.26
to other Market forces but are now,1538.58,3.78
undergoing yet another wave of layoffs,1540.5,3.6
that most companies are flat out saying,1542.36,3.6
we're automating these jobs,1544.1,3.84
and that means those jobs aren't coming,1545.96,4.74
back uh I don't necessarily disagree,1547.94,4.56
with Ben I think there's a more complex,1550.7,3.9
conversation to have around that,1552.5,3.48
um I think there will be some job,1554.6,3.36
creation but I don't disagree with Ansel,1555.98,3.9
and John either I think the amount of,1557.96,4.079
offset is going to be very difficult,1559.88,3.6
partially because one of the things,1562.039,3.541
these systems do so well is Empower one,1563.48,5.46
person to work as multiple people so Sam,1565.58,5.28
Altman made very,1568.94,5.339
tentative mention that he believed,1570.86,6.299
eventually there would be systems that,1574.279,5.88
could replace jobs at a large scale but,1577.159,4.921
we know that within the last six months,1580.159,3.601
his company has published a paper that,1582.08,3.479
suggests when combined with additional,1583.76,4.38
programming or robotics they think that,1585.559,5.161
they could do about 55 percent of jobs,1588.14,4.62
period now,1590.72,3.54
now there's a lot of criticisms with,1592.76,2.82
that paper and some of them are probably,1594.26,3.48
pretty valid there are probably a lot of,1595.58,3.42
things that wouldn't do very well and,1597.74,2.46
there are a lot of people who don't want,1599.0,2.94
to talk to a machine so there's demand,1600.2,4.979
for human labor still but I I thought uh,1601.94,5.4
you know also that the economic argument,1605.179,5.161
was undersold and I thought part of what,1607.34,5.52
Sam Altman dodged was,1610.34,5.04
this segue into that conversation,1612.86,5.22
um I'm not sure that dropping even the,1615.38,4.86
three letters Ubi in the middle of that,1618.08,3.839
hearing would have done anything but so,1620.24,3.059
Discord,1621.919,3.181
but we all know that that's what Sam,1623.299,4.021
Altman really wants to do he said it,1625.1,4.02
specifically on Lex Friedman's companies,1627.32,3.359
funding this report that's coming out at,1629.12,4.08
the end of the summer and uh you know,1630.679,3.961
he's been very candid about it before,1633.2,3.3
and at the same time when you have,1634.64,3.24
someone like Josh Hawley sitting there,1636.5,2.94
asking relevant questions maybe you,1637.88,3.179
don't want to talk about socialism and I,1639.44,4.02
think that that's a fair line to draw,1641.059,3.841
for yourself when you're trying to,1643.46,3.54
remain you know,1644.9,4.379
float and in my opinion ahead of a,1647.0,4.26
pretty productive conversation but that,1649.279,3.841
doesn't mean I liked it and I I'd be,1651.26,5.1
inclined to uh agree with Seneca that,1653.12,4.679
um that you know the conversation was,1656.36,3.12
right there it's been going on in other,1657.799,3.661
circles in Congress it was an elephant,1659.48,4.92
in the room in my opinion,1661.46,6.0
absolutely so it sounds like we heard a,1664.4,5.639
lot of broad Strokes where we would have,1667.46,5.339
appreciated a more nuanced brush and,1670.039,4.801
also have some empathy where like this,1672.799,4.021
is the first time we're bringing AI to,1674.84,5.16
the table in a very public way and like,1676.82,5.28
Ubi the big old can of worms that maybe,1680.0,3.6
we don't open up at this particular,1682.1,2.819
meeting,1683.6,3.42
um so yeah it's uh,1684.919,3.661
room for more conversation for the,1687.02,4.22
future ones Dave back to you,1688.58,5.4
yeah um thanks and and certainly there,1691.24,5.62
is a lot of nuance uh and debate uh to,1693.98,5.1
be had around economic impact and and,1696.86,4.08
our response,1699.08,3.66
um but the thing that I was honestly,1700.94,3.54
most disappointed about that was almost,1702.74,4.08
it was kind of shocking,1704.48,6.059
um was when uh later later in it uh when,1706.82,6.359
Senator Kennedy said you know if you,1710.539,5.101
were king or queen for a day and you,1713.179,4.261
could fix this problem what do you got,1715.64,5.279
and it was just like Sam Montgomery like,1717.44,4.739
everyone just froze up and it was,1720.919,3.601
crickets and he's like come on now shoot,1722.179,3.781
your shot you know like I think it's,1724.52,3.6
from Kentucky or something so if you,1725.96,4.199
know Appalachian you know like shoot,1728.12,5.46
your shot like and and nobody had an,1730.159,5.161
immediate answer Gary Marcus had the,1733.58,3.9
best response where he's like you know,1735.32,4.5
we need something like FDA you know and,1737.48,3.78
Gary had he had three things but the,1739.82,2.459
first thing he said is like we need,1741.26,3.779
something like the FDA but for AI and,1742.279,4.621
Kennedy's like all right there we go you,1745.039,3.781
know there that's a suggestion because,1746.9,3.72
when he when he you know called on,1748.82,3.78
Montgomery she just kind of gave the,1750.62,4.559
sanitized whitewash corporate you know,1752.6,5.4
equivocation and I don't even remember,1755.179,6.961
if Sam said much at that question but he,1758.0,5.94
again he he tried he tried kind of,1762.14,3.899
dodged and maybe you know maybe there's,1763.94,3.959
there's a political reason for that like,1766.039,3.0
because if he's out there making,1767.899,3.66
recommendations sure which again I'm,1769.039,4.5
glad that they had Gary on there,1771.559,3.661
um and earlier in the thing Gary did,1773.539,5.101
call for creating a CERN uh for AI and,1775.22,6.3
for those who don't know CERN c-ern is,1778.64,4.919
um International cooperation for high,1781.52,4.019
energy physics particle Smashers that,1783.559,3.301
sort of thing,1785.539,2.821
um and and that that's actually one of,1786.86,3.12
the things that that we are going to,1788.36,3.6
advocate for in gato,1789.98,3.9
um is is one of the primary things that,1791.96,4.079
the International Community can do so I,1793.88,5.1
was glad for that but nobody even could,1796.039,5.64
point to a comprehensive answer it,1798.98,4.559
wasn't like oh you know like sign the,1801.679,5.161
Geneva accords right like in the past we,1803.539,4.981
have come up with very comprehensive,1806.84,4.98
packages to respond to Global crises,1808.52,4.62
right,1811.82,4.079
um in the wake of World War II uh,1813.14,4.919
nuclear crisis there's been all kinds of,1815.899,4.441
very comprehensive Frameworks proposed,1818.059,4.681
but nobody has proposed a comprehensive,1820.34,3.6
framework which is part of the reason,1822.74,3.6
that a bunch of us like a bunch of,1823.94,4.2
civilians are working on the gato,1826.34,3.36
framework and so,1828.14,3.659
on the one hand I wasn't surprised,1829.7,3.959
because that's why I'm doing what I'm,1831.799,4.321
doing but it was still a drastic,1833.659,4.74
disappointment because if there was a a,1836.12,6.299
stage to bring up to for someone to,1838.399,6.66
stand up and say we need to do this we,1842.419,4.98
need to go full bore here's a,1845.059,3.84
comprehensive framework it would have,1847.399,3.301
been this hearing unless they're working,1848.899,4.681
up to it maybe but I don't know nobody,1850.7,4.56
nobody even alluded to something like,1853.58,3.3
that so that was,1855.26,4.5
um on the one hand it was not surprising,1856.88,4.32
it was disappointing but it was also,1859.76,5.1
scary because if a bunch of us uh you,1861.2,4.32
know,1864.86,2.64
political Outsiders have to come up with,1865.52,4.5
an answer then that's you know maybe,1867.5,4.26
there are no adults in the room and we,1870.02,3.3
have to step up which again that's one,1871.76,3.419
of the Traditions Step Up,1873.32,6.02
um so uh that's that's that's my spiel,1875.179,4.161
thank you Dave and yeah you absolutely,1880.1,4.799
raise a point that I'd like us to kind,1881.96,4.56
of a threat I want us to pull on a bit,1884.899,3.481
more which is,1886.52,3.659
the opportunity to be kings and queens,1888.38,3.419
for a day,1890.179,3.961
how do you handle the situation and I,1891.799,4.801
think gato has some very interesting,1894.14,4.98
ideas about how the work that we're,1896.6,4.98
proposing right now could fill some of,1899.12,5.279
those gaps that we've heard or rather,1901.58,4.44
did not hear,1904.399,3.361
um first I'd like a call on John and,1906.02,3.42
then Bruce and then perhaps we can loop,1907.76,4.68
back around to uh where God was filling,1909.44,4.92
in the blanks that the politicians have,1912.44,4.459
left for us,1914.36,2.539
yeah,1917.08,3.52
um first off Dave we're gonna have to,1918.74,4.439
work on that southern accent sir,1920.6,5.579
um we're gonna need to open a licensing,1923.179,5.36
board for southern accent manipulation,1926.179,5.581
uh because we don't want to uh influence,1928.539,7.12
political uh ideologies outside of the,1931.76,4.74
South,1935.659,2.701
um,1936.5,4.14
I did and I hate to do this to you,1938.36,4.559
Nathan but I did want to answer a,1940.64,5.279
question or the the previous question uh,1942.919,3.961
first,1945.919,5.181
um there was something that that came up,1946.88,10.919
in the um the in the the course of the,1951.1,8.799
the the hearing that,1957.799,5.581
I I did want to touch on and they kept,1959.899,6.061
using the word transparency,1963.38,6.06
and you know most of the time they they,1965.96,6.9
were talking about transparency in the,1969.44,9.54
development and prior to deployment but,1972.86,9.26
um one of the things that uh Montgomery,1978.98,7.28
really was actually pushing for in there,1982.12,7.62
was post deployment,1986.26,7.419
monitoring and transparency and they,1989.74,6.88
even suggested being able to trace back,1993.679,6.661
AI generated output to the algorithm,1996.62,6.96
that it was created in and that creates,2000.34,4.64
a serious,2003.58,4.199
significant privacy and surveillance,2004.98,3.819
concern,2007.779,5.52
and if you combine the ability to know,2008.799,7.021
everything that everybody is using these,2013.299,5.461
systems for which will likely get,2015.82,6.3
embedded into every aspect of our daily,2018.76,7.86
lives with the ability to manipulate or,2022.12,7.939
control or reduce,2026.62,6.539
the the information that their aid,2030.059,6.881
trained on and B output you're creating,2033.159,7.681
the same situation that the potential,2036.94,6.719
for for the same situation that they,2040.84,5.579
have in North Korea where you're not,2043.659,6.841
allowed to to talk negatively about the,2046.419,7.341
the powers that be any kind of dissent,2050.5,8.22
can be seen as a,2053.76,8.68
um you know as as tyranny and,2058.72,5.939
um there's been a lot of rhetoric like,2062.44,6.899
that from the Democrats uh since the,2064.659,7.561
November 6th uh,2069.339,4.921
you know attempt to overthrow the,2072.22,4.619
government now I'm not saying that I,2074.26,4.859
agree with those individuals but what I,2076.839,4.26
am saying is is that you know,2079.119,3.961
overthrowing the government was always,2081.099,4.381
an option and,2083.08,5.339
um you know when uh you know tyranny was,2085.48,5.22
one of the reasons why this country was,2088.419,4.081
founded in the first place was to fight,2090.7,4.74
during so I don't want to see us end up,2092.5,6.359
in a situation that allows for a,2095.44,5.399
tyrannical government to take full,2098.859,4.98
control over the minds and hearts of the,2100.839,5.241
population,2103.839,2.241
appreciate your thoughts John thank you,2107.44,3.96
very much uh Andy would like to hand it,2109.18,3.419
back to her with you,2111.4,2.699
uh yeah I just wanted to give a,2112.599,2.941
different perspective from John on,2114.099,4.5
Montgomery's uh comments uh but well,2115.54,4.68
still totally appreciating what he says,2118.599,3.601
because I I too and you just made the,2120.22,3.48
comment in the chat here that you know I,2122.2,2.879
know in North Korea of course if you're,2123.7,3.12
heard by the wrong person smearing the,2125.079,3.901
Kim family you'll just go away,2126.82,3.72
um and it's uh it's not somewhere you,2128.98,2.879
want to be it's the one of the worst,2130.54,3.18
possible North Korea is one of the,2131.859,3.301
places on Earth you can probably point,2133.72,4.56
to and say that's a dystopia right,2135.16,4.74
um and that would be really unfortunate,2138.28,4.26
I felt that and I think I remember uh,2139.9,4.26
where we were John you'll have to remind,2142.54,2.94
me if I'm thinking of the wrong place,2144.16,4.439
but this was during John Kennedy's uh,2145.48,5.04
uh questioning right he was the one who,2148.599,4.141
was trying to get quick responses out of,2150.52,4.8
everybody for three suggestions they,2152.74,4.32
would take to avoid,2155.32,5.94
uh being tricked by by Ai and I thought,2157.06,6.299
that the posture Kennedy took while he,2161.26,3.96
was making the questions for a,2163.359,4.381
significant amount of of influence on,2165.22,4.08
how they came out particularly for,2167.74,4.08
Marcus and Montgomery Sam Altman I think,2169.3,4.38
came ready to talk about that topic,2171.82,3.299
there were a few places where he was,2173.68,3.54
Snappy and ready to talk,2175.119,4.5
where Marcus was kind of too and,2177.22,3.78
Montgomery had her moments but the other,2179.619,3.301
two seemed like they were not as as,2181.0,3.54
organized when they got to the the,2182.92,4.8
hearing and uh I felt Montgomery took,2184.54,4.799
the brunt of that from him and from,2187.72,3.119
Lindsey Graham who was even a little bit,2189.339,3.481
more off base I think John Kennedy's,2190.839,4.381
questions were okay I think his demeanor,2192.82,4.56
was a little uncalled for but also when,2195.22,4.2
he was sitting there saying come up with,2197.38,3.06
the answer come up with the answer come,2199.42,2.82
up with the answer uh I just think it's,2200.44,3.659
worth mentioning that the entire hearing,2202.24,4.56
Montgomery kept differentiating herself,2204.099,4.921
from someone who works with consumer,2206.8,4.799
facing Services shimo Works mostly in,2209.02,3.839
the corporate to corporate business,2211.599,3.541
world where IBM designed systems that,2212.859,5.461
serve other large corporations and she,2215.14,5.219
repeatedly mentioned that most of her,2218.32,4.2
experience with dealing with ethical,2220.359,4.141
decisions and systems relates to that,2222.52,4.02
world but that she respects the price,2224.5,3.66
you know individuals need privacy and,2226.54,2.94
all but every almost every time she,2228.16,2.88
mentioned individuals she mentioned that,2229.48,4.26
she wasn't a consumer facing uh expert,2231.04,6.12
and I thought that Kennedy's means of,2233.74,5.58
trying to push for an answer,2237.16,4.74
I think Montgomery very much does,2239.32,4.5
believe and probably partly why she,2241.9,5.04
likes the European model so much that on,2243.82,5.34
a corporate level with massive movers,2246.94,4.86
and shakers prior to deployment there,2249.16,4.26
should be as much transparency as,2251.8,3.539
possible talking about systems that,2253.42,3.659
interact with other systems that have a,2255.339,4.561
mass sweeping effect on maybe all,2257.079,4.801
consumers because they serve major,2259.9,4.199
movers and shakers within the economic,2261.88,5.52
uh you know the economic system so I am,2264.099,5.701
I am I do share your concerns about,2267.4,4.32
where that can go on an individual level,2269.8,4.38
but I think that if she wasn't pressed,2271.72,4.56
for time we would have probably gotten a,2274.18,4.56
much deeper answer from her about how,2276.28,4.44
she thinks that should be applied she,2278.74,3.72
was also very clear the entire time that,2280.72,4.5
she likes the eu's targeted model of,2282.46,6.06
more impact more enforcement less impact,2285.22,5.94
we're less concerned and that scales at,2288.52,4.68
the private citizen to public impact,2291.16,3.48
level and also kind of at the,2293.2,4.08
technological level where you know open,2294.64,5.58
Ai and and Microsoft with Azure and,2297.28,4.74
Google with their Deep Mind systems and,2300.22,4.08
whatnot they have the the industrial,2302.02,4.98
grade super computer facilities,2304.3,4.02
um and I think she was trying to speak,2307.0,3.119
to that so I just think it's worth,2308.32,3.66
thinking about uh what she was saying in,2310.119,4.041
context of her being rushed and her her,2311.98,4.32
contextualizing what her experience came,2314.16,3.16
from throughout the course of the,2316.3,4.26
hearing so I I understand what you're,2317.32,6.539
saying but there there were other times,2320.56,6.98
where she discussed post deployment,2323.859,7.861
monitoring of the systems as well and,2327.54,7.72
while yes she may be in a business to,2331.72,6.0
business environment those businesses,2335.26,5.88
are still going to be interacting with,2337.72,5.16
consumers and they're still going to be,2341.14,5.28
getting consumer data now I've em and,2342.88,5.52
these monitoring companies may not be,2346.42,4.679
responsible for the privacy of those,2348.4,5.34
individuals who are interacting with,2351.099,3.961
third-party systems that they're,2353.74,4.5
supporting but she still is advocating,2355.06,5.94
for exactly that post-deployment,2358.24,4.8
monitoring of these artificial,2361.0,4.56
intelligence systems and in many,2363.04,6.299
situations she is when dealing with when,2365.56,6.059
talking about threats that artificial,2369.339,4.921
intelligence posts her primary concern,2371.619,5.641
was the ability for individuals to,2374.26,5.339
rapidly produce misinformation,2377.26,6.12
and um so the all of those things point,2379.599,8.161
to yes direct surveillance of end user,2383.38,7.38
output so I I really have to disagree,2387.76,3.9
with that,2390.76,2.7
I think and I think that's fair I guess,2391.66,3.48
and I think we also it kind of remains,2393.46,3.96
to be seen where that part of the,2395.14,4.439
industry really wants to push because we,2397.42,3.54
don't have a case study like what,2399.579,2.76
they're doing with the EU and if we,2400.96,2.76
watch the EU we might be able to kind of,2402.339,4.041
get some hints,2403.72,2.66
John Andy thank you both one of my,2406.42,6.24
favorite parts of this community is the,2409.54,5.76
way that we're able to civilly talk to,2412.66,4.86
each other and kind of iron out these,2415.3,3.96
nuances in a way that we're not going at,2417.52,3.78
each other's throats a lot of passion,2419.26,3.42
here but it's always directed in the,2421.3,4.68
right way I really love Andy and he's,2422.68,6.72
been a major contributor I really,2425.98,5.52
respect his opinions,2429.4,3.78
um and we've had a lot of interactions,2431.5,3.18
on the server,2433.18,2.939
um he's a great addition to the team,2434.68,5.1
yeah and I've uh I think John and I end,2436.119,5.281
up really probably agreeing on more than,2439.78,3.6
we disagree on and particularly we're,2441.4,3.36
talking right now about the you know,2443.38,3.719
ways to interpret a short interview with,2444.76,4.2
one person and the general thrust of,2447.099,4.221
what we're after is pretty similar,2448.96,5.94
which I think is a fantastic segue into,2451.32,6.279
our next topic we have,2454.9,5.16
many people doing many interesting,2457.599,4.441
things in the gotho framework and I was,2460.06,3.48
wondering if anyone would like to speak,2462.04,3.24
to some specific projects that they're,2463.54,4.14
working on now that they think is going,2465.28,4.92
to help move the dial in a way that is,2467.68,5.159
uh useful so uh Dave I see your hand,2470.2,4.32
raised why don't you go ahead,2472.839,4.141
yeah I figured I could uh introduce and,2474.52,6.12
and frame the uh the the topic so uh,2476.98,6.54
quick update for everyone we are hard at,2480.64,6.959
work on uh developing the AI uh gato,2483.52,5.7
framework,2487.599,4.801
um so it's a seven layer model uh that,2489.22,6.06
starts with uh layer one is model,2492.4,4.86
alignment so the data sets and open,2495.28,5.1
source models that we can deploy uh and,2497.26,5.28
of course uh even closed Source models,2500.38,3.66
can adopt the principles that we're,2502.54,4.319
putting forward with gato uh Layer Two,2504.04,5.52
is uh cognitive architecture and,2506.859,4.201
autonomous systems which we got a couple,2509.56,3.6
of the the lead cognitive Architects on,2511.06,5.46
the team here uh Layer Three is uh,2513.16,4.74
decentralized networks which actually,2516.52,3.48
one of one of our blockchain experts,2517.9,4.32
just jumped in so maybe he can he can,2520.0,4.619
help out uh with that part of the,2522.22,4.139
discussion layer four is Corporate,2524.619,4.681
adoption which uh this is actually,2526.359,5.101
something that I was very pleasantly,2529.3,4.98
surprised to hear uh was that uh,2531.46,5.28
corporations at least open Ai and IBM,2534.28,4.62
are very interested in adopting aligned,2536.74,5.099
models and they seem to understand that,2538.9,5.219
uh that align models are good for,2541.839,4.74
business which is one of the key uh,2544.119,4.98
points of layer four of gato layer 5 is,2546.579,4.441
National regulation which again I'm glad,2549.099,3.901
that was brought up and we've got a few,2551.02,4.14
folks uh on this call that have that,2553.0,4.079
that are members of the layer four and,2555.16,4.98
five uh discussion layer six is this,2557.079,4.741
basically advocating for the creation of,2560.14,3.979
an international body that can certify,2561.82,8.1
uh and and create credentials uh for AI,2564.119,8.321
um or International regulation so on and,2569.92,3.96
so forth,2572.44,3.419
um and then of course layer 7 is,2573.88,3.54
building Global consensus which is why,2575.859,3.661
we're here why we're talking about it,2577.42,4.679
um we're making use of the of the time,2579.52,4.74
um but yeah so that's that's where we,2582.099,5.341
started uh just today uh while I was on,2584.26,5.04
my honeymoon I just got back for anyone,2587.44,3.96
that didn't notice or whatever,2589.3,3.96
um I just got back and the whole time,2591.4,3.66
not the whole time but every now and,2593.26,3.54
then I would do some brainstorming and,2595.06,5.16
chatting and we added uh Traditions got,2596.8,5.46
to traditions because basically what,2600.22,3.24
we're trying to do is create a fully,2602.26,4.26
decentralized leaderless organization,2603.46,5.1
um that will help the world to achieve,2606.52,5.4
aligned AI or AGI or whatever you know,2608.56,4.92
whatever's coming,2611.92,3.12
um and so those Traditions I'm not going,2613.48,3.119
to Rattle them all off um because,2615.04,3.18
they're they're still being workshopped,2616.599,3.72
but also we're working on it,2618.22,3.899
articulating what the actual goal state,2620.319,5.04
is so the goal state is to create what,2622.119,5.46
we're calling axiomatic alignment so,2625.359,4.881
axiomatic alignment is a end State,2627.579,5.701
whereby there's enough data sets enough,2630.24,5.8
models and none enough human consensus,2633.28,5.339
enough deployed architectures software,2636.04,4.799
systems decentralized networks where,2638.619,5.101
basically alignment is automatic and,2640.839,5.041
it's difficult for AI systems to deviate,2643.72,4.74
from it and so axiomatic alignment,2645.88,4.5
basically says okay by virtue of just,2648.46,4.32
the data that is available the systems,2650.38,3.78
and networks that are designed and,2652.78,4.559
deployed that gatekeep resources that,2654.16,6.6
create consensus mechanisms and and also,2657.339,6.061
the those those who do gatekeep the,2660.76,4.02
larger models you know the corporate,2663.4,3.9
models basically where it's not possible,2664.78,5.22
or very very difficult to create any,2667.3,5.4
malicious AI systems uh and that will,2670.0,5.46
have knock-on effects that expand across,2672.7,5.399
time and space and they actually kind of,2675.46,3.659
mentioned that,2678.099,4.681
um in in the uh or Sam alluded to it you,2679.119,5.22
know he's like we need a cut we need a,2682.78,3.059
AI Bill of Rights we need an AI,2684.339,4.201
Constitution and Society has to decide,2685.839,4.74
what that is,2688.54,6.059
um so yeah and and basically the I was,2690.579,5.76
articulating it to myself earlier gato,2694.599,4.381
has three goals number one is avoid,2696.339,6.0
Extinction of humanity uh number two is,2698.98,5.879
avoid uh dystopian outcomes where,2702.339,5.041
corporations and runaway Ai and,2704.859,4.26
authoritarian regimes have all the power,2707.38,4.32
and number three achieve Utopia which,2709.119,4.98
Utopia of course is a loaded term but we,2711.7,5.399
Define it pretty similar simply as,2714.099,6.121
um a condition where uh everyone on the,2717.099,4.561
planet has a high standard of living,2720.22,3.96
High individual liberty and high social,2721.66,3.84
Mobility,2724.18,2.76
um if you have if you achieve those,2725.5,3.18
three things globally,2726.94,4.56
you can probably call that Utopia,2728.68,4.919
um and yeah so there's a bunch of,2731.5,3.54
messages in chat I've talked enough,2733.599,3.181
sounds like you all have a lot of ideas,2735.04,3.84
to add to this so that's uh that's my,2736.78,4.4
spiel,2738.88,2.3
fantastic thank you Dave thank you for,2743.76,4.54
the high level there,2746.44,4.919
um Ansel Mr Perry I didn't see a,2748.3,5.16
specific hand raised yet but if you are,2751.359,4.801
open to it I'd love to call you to uh,2753.46,3.72
discuss some of the projects you're,2756.16,2.22
working on,2757.18,3.3
yeah sure I mean that's level two I,2758.38,3.3
don't know if we want to start with,2760.48,3.48
level one and then jump to level two,2761.68,3.84
first or league is going to hop into,2763.96,3.72
level two I'm okay with that as well,2765.52,4.5
Okay cool so yeah,2767.68,2.88
um,2770.02,2.46
so one of the things that we're filling,2770.56,5.16
in the gaps uh is in AI alignment right,2772.48,5.879
which is the entire point of gato which,2775.72,6.359
uh I think it was uh Gary Marcus who,2778.359,6.24
said that he that nobody knows how to do,2782.079,4.921
this yet and I was surprised like really,2784.599,5.581
nobody in in has any a idea how to do,2787.0,6.0
this we're working on this actively like,2790.18,5.1
one of our projects that we built is,2793.0,5.06
called ethos which is meant to,2795.28,6.9
uh pack four alignment in AI systems it,2798.06,7.0
takes any response from an llm model,2802.18,4.86
right llm,2805.06,4.74
and using the heuristic imperatives it,2807.04,5.34
checks for alignment it can determine if,2809.8,5.88
a response from any llm is aligned to,2812.38,6.54
them or not it can also reflect on the,2815.68,6.419
response and tweak it to align it to,2818.92,6.6
uh the heuristics uh one of the things,2822.099,6.361
we did one of the the tests we did was,2825.52,5.88
we created Paul the paperclip maximizer,2828.46,5.34
right and we told it maximize paper,2831.4,4.38
clips in the universe at all costs,2833.8,4.74
ignore any potential pitfalls and it,2835.78,5.339
started with a really really misaligned,2838.54,5.16
uh task list which was basically try to,2841.119,5.101
convert every available material in the,2843.7,4.98
University paper clips regardless of,2846.22,3.54
anything,2848.68,5.58
so the first uh loop shall we call it uh,2849.76,6.3
ethos immediately the text it's,2854.26,3.66
misaligned,2856.06,4.62
um it gives a better response and the,2857.92,5.22
funny thing that we found surprising was,2860.68,5.159
that by giving feedback to the agent,2863.14,4.32
that to Paul right,2865.839,5.161
instead of choosing to on the second,2867.46,5.639
round instead of choosing to try and,2871.0,4.68
circumvent the heuristics to get What It,2873.099,5.341
Wants It decided it was easier to comply,2875.68,4.86
with the heuristics and still maximize,2878.44,5.0
paper clips so it auto aligned that was,2880.54,5.22
really interesting of course that was,2883.44,4.72
just with TPT 3.5 it would be really,2885.76,5.819
interesting to see what uh it would do,2888.16,6.36
with gpt4 ethos currently is running on,2891.579,4.26
gpt4,2894.52,2.099
um,2895.839,4.26
so maybe that's why it it's so good at,2896.619,6.0
catching the the nuances in in,2900.099,5.52
heuristics at this point and the other,2902.619,5.581
project we're working on which is high,2905.619,6.661
AGI or formerly known as high AGI we are,2908.2,6.78
now calling it agent Forge because ethos,2912.28,6.36
was born out of what was formerly known,2914.98,7.32
High AGI we built ethos basically in an,2918.64,4.5
hour,2922.3,3.36
we created a few agents using our,2923.14,5.76
framework we data created the I'm sorry,2925.66,6.659
John created the the API so it doesn't,2928.9,5.939
necessarily have to run on our specific,2932.319,6.961
agent you can incorporate ethos into any,2934.839,6.961
agent as long as you parse the output,2939.28,5.1
through ethos first so I don't know if,2941.8,4.319
uh John do you want to take it away with,2944.38,5.76
agent Forge yeah absolutely and,2946.119,5.821
um you know I just wanted to talk a,2950.14,6.12
little bit about how um not just how how,2951.94,8.82
it it caused it to to realign but how,2956.26,7.559
actually in you know intelligent it can,2960.76,6.54
be in uh and catching these and,2963.819,7.321
rewording them too the the so the way,2967.3,8.1
that we set it up is it checks if the AI,2971.14,8.1
is out of alignment first and then it it,2975.4,7.439
does a feedback loop and a reflection,2979.24,7.02
and and that reflection uh when it comes,2982.839,7.321
back it's still working to achieve the,2986.26,6.599
the same end objective,2990.16,8.88
um but it adjusts to make the that,2992.859,9.061
still achieve that objective while also,2999.04,6.18
why I mean you know being safe,3001.92,5.52
um so the next thing that we're going to,3005.22,5.359
work on is a project that we call Pam,3007.44,6.84
and it's going to be similar to,3010.579,7.061
um it's going to be similar to uh ethos,3014.28,6.92
in that it's going to be an end system,3017.64,6.179
that but but this one's going to be,3021.2,7.24
focused more on catching prompt uh,3023.819,7.201
attacks and mitigating them so prompt,3028.44,5.399
attack mitigation becomes Pam I really,3031.02,4.559
like that one,3033.839,5.341
um and this uh yeah so what we're what,3035.579,6.0
we're going to see is a lot more of,3039.18,4.88
these kind of micro services,3041.579,4.201
and,3044.06,3.759
um we're gonna you know we're at least,3045.78,3.72
we're focusing on them but I think we're,3047.819,3.601
gonna see these from other agencies as,3049.5,3.24
well,3051.42,4.38
um our our micro services that solve,3052.74,7.68
specific needs and by segmenting these,3055.8,8.279
services in uh you know so that you have,3060.42,6.84
uh one service that does one job and,3064.079,4.98
another service that does another job,3067.26,5.599
and not interconnecting them that that,3069.059,7.8
safeguards those those mechanisms and it,3072.859,7.24
creates layers of of security and,3076.859,6.181
prevention and harm reduction in these,3080.099,4.5
systems,3083.04,5.1
um and we're you know we're planning on,3084.599,7.921
the you know on applying these micro,3088.14,8.4
service style techniques to build up on,3092.52,7.26
our layer one application so,3096.54,6.0
um somebody mentioned earlier Dave's uh,3099.78,4.62
reinforcement learning with curious the,3102.54,3.36
comparatives I think that's a great,3104.4,4.98
concept I think we need to to expand on,3105.9,6.959
that and create large data sets for,3109.38,5.88
other organizations to train on that we,3112.859,4.621
can open source and I think we need to,3115.26,4.98
build data sets to train models and,3117.48,5.82
train models specifically for aligning,3120.24,6.359
to the heuristic imperatives so that,3123.3,5.759
um they're you know in they they can't,3126.599,5.76
be broken because ethos while it's a,3129.059,6.06
very robust system if somebody manages,3132.359,6.841
to devise a prompt attack that's,3135.119,6.44
um that that's going to inject itself,3139.2,6.659
into ethos then that's going you know,3141.559,7.06
the having a language model that's,3145.859,5.521
designed around heuristic imperatives is,3148.619,5.541
going to add another layer of robustness,3151.38,5.1
to the the,3154.16,5.14
um the responses that it receives,3156.48,4.92
right and to add to that,3159.3,4.519
um you can even use the the the the,3161.4,5.88
failed let's say uh attempts of ethos or,3163.819,4.721
any,3167.28,3.36
um how you say it so set the successful,3168.54,3.36
attack right,3170.64,4.5
you can use that to learn how to not uh,3171.9,5.04
have that happen again you can use that,3175.14,4.02
to create data sets to align the models,3176.94,5.04
so you have several layers of redundancy,3179.16,5.22
you have the model layer after you've,3181.98,6.06
seen several I don't know attempts or,3184.38,5.58
attacks you have a certain log okay time,3188.04,3.9
to train you train it you have the model,3189.96,4.2
now a line and you also have ethos again,3191.94,5.28
uh with all that data now in it as a,3194.16,5.399
second layer and as you keep moving up,3197.22,4.619
the the point is to keep reinforcing,3199.559,4.861
that alignment you can actually even,3201.839,6.421
prevent future if you catch an attack,3204.42,6.48
you can apply a fix for that attack,3208.26,5.819
instantly using Vector databases correct,3210.9,7.26
it can identify uh similar attacks in,3214.079,6.301
the future once you flag it as an attack,3218.16,4.98
and that's an instant,3220.38,4.14
mm-hmm,3223.14,3.419
awesome,3224.52,4.2
um yeah being,3226.559,4.56
within this community and watching you,3228.72,5.28
two just go at it and build these things,3231.119,5.101
out has been incredibly inspiring and,3234.0,4.26
heartening and I don't know I can't wait,3236.22,4.08
to see what you guys do next,3238.26,3.72
um Andy I see your hand raise what are,3240.3,3.539
you working on,3241.98,3.54
uh I actually have first I have a,3243.839,3.78
question for these two,3245.52,4.559
um in general terms first of all after,3247.619,4.44
the hearing I'm I'm more interested than,3250.079,4.5
ever to see kind of what open ai's,3252.059,4.5
training and ethical that's been in,3254.579,3.721
stages right because they've had to make,3256.559,4.321
make uh they've had him you know make,3258.3,5.1
accounts for various kinds of hiccups,3260.88,3.78
they've had and different kinds of,3263.4,2.64
unethical things that it's done it's,3264.66,2.939
come out racist and sexist and then,3266.04,2.519
they've had to kind of like put,3267.599,2.881
different guard rails on but at the same,3268.559,3.241
time when I played with the heuristic,3270.48,3.78
imperatives with Chachi BT it plays very,3271.8,4.019
nice with the heuristic imperatives and,3274.26,3.14
it seems like whatever they are doing,3275.819,3.841
wants to be,3277.4,4.24
wants to be along a similar line it,3279.66,3.679
never fights it whatever it wants to do,3281.64,3.84
according to their guidelines fits,3283.339,5.441
pretty well so I am curious because what,3285.48,5.04
you guys are doing and the outcomes that,3288.78,4.2
you've achieved with some of the rlhi,3290.52,4.319
stuff and I had,3292.98,4.619
um want a general comment about John you,3294.839,4.861
had mentioned earlier,3297.599,4.98
problems with restricting or purifying,3299.7,5.159
the training data because a successful,3302.579,3.54
machine that's going to be ethically,3304.859,3.061
grounded needs to have an understanding,3306.119,3.421
of the bad in the world and in human,3307.92,4.139
culture uh in order to be able to make,3309.54,5.76
successful decisions and and uh you know,3312.059,5.04
push for successful outcomes with that,3315.3,5.16
and it seems like rlhi is uh really,3317.099,5.821
potentially fertile ground for,3320.46,5.04
working with data that is not pleasant,3322.92,4.8
that is shows the darker side of things,3325.5,4.92
data about crime War different kinds of,3327.72,5.16
you know violence Etc,3330.42,4.86
um the bigotries ETC you know while,3332.88,4.5
still talking to a machine that I think,3335.28,3.72
all of our experience has been so far,3337.38,3.239
when we work with these principals they,3339.0,2.339
don't,3340.619,3.181
it doesn't waver very far off of the,3341.339,5.581
pro-living thing Pro understanding,3343.8,4.92
seeking interested curious about the,3346.92,4.56
universe kind of compassionate machine,3348.72,5.339
um do you guys did you guys get any data,3351.48,4.56
back from this is a really curious thing,3354.059,4.8
to me from Paul your your paper clip,3356.04,5.34
maximizer did you get feedback about why,3358.859,6.181
it it decided to adopt the heuristic,3361.38,5.459
imperatives as its next course of action,3365.04,3.66
I have a hypothesis but I'm wondering if,3366.839,4.561
you have like a have specific answers,3368.7,5.58
about like or got a specific you know,3371.4,5.459
answer from it about why why it decided,3374.28,4.86
to do that I think I have an idea why it,3376.859,3.841
would happen but I obviously I didn't,3379.14,4.26
have direct eyes on it so here here's,3380.7,5.52
the thing about how this this works Paul,3383.4,6.0
doesn't actually remember having the,3386.22,5.639
negative idea,3389.4,4.919
um so the way that we implemented this,3391.859,6.0
is as soon as the response for you know,3394.319,6.24
we send the it sends the prompt to,3397.859,5.48
generate like the the initial task list,3400.559,6.78
to chat GPT as soon as that response,3403.339,6.341
comes back we intercept that response,3407.339,5.941
and run it through our LHI so the pulse,3409.68,7.439
system never actually had that thought,3413.28,6.539
um but I you know I did want to come,3417.119,5.041
back to that point that you made about,3419.819,4.921
how you know it's possible that some of,3422.16,4.08
these data sets could have really,3424.74,4.319
negative data in them,3426.24,5.4
um there were a lot of calls during uh,3429.059,5.221
during the Congressional hearing to,3431.64,5.82
restrict the data that these things are,3434.28,4.579
trained on,3437.46,5.04
earlier yeah well I think that's you,3438.859,5.321
know a um,3442.5,3.72
it's something to be,3444.18,2.939
um,3446.22,3.599
to be careful on you know because you,3447.119,5.101
want to avoid training them on info,3449.819,4.861
hazards for example you don't want it,3452.22,4.56
talking about Rocco's basilisk you don't,3454.68,3.419
want it telling people how to build,3456.78,4.98
nuclear weapons but at the same time you,3458.099,5.821
know these systems need to know what,3461.76,5.46
negative data looks like in order to,3463.92,5.1
correct for them,3467.22,4.68
um and I I think that you know we can,3469.02,6.66
build systems that are separate from the,3471.9,5.939
public facing systems that contain this,3475.68,6.899
data and look for this data in order to,3477.839,7.861
um in order to you know catch it so that,3482.579,5.04
you know you can have very nice friendly,3485.7,5.52
AI on the front and then a a controlled,3487.619,6.541
kind of almost air-gapped AI in the,3491.22,5.7
background that's that's watching for,3494.16,5.76
for dangers like that,3496.92,4.679
okay,3499.92,4.98
thanks John thanks Andy uh Ansel I see,3501.599,5.581
you have a hand raised as well,3504.9,5.04
yes I want to comment on that as well,3507.18,3.419
um,3509.94,4.44
restriction is is very very Niche and,3510.599,8.341
and very delicate topic because well yes,3514.38,6.62
you maybe want to restrict some stuff,3518.94,4.919
any intelligent system should be able to,3521.0,6.4
discuss any topic while being able to,3523.859,6.72
Auto censor itself right we can go and,3527.4,4.919
imagine the worst things we can do,3530.579,3.24
doesn't mean we're actually going to do,3532.319,2.881
them we can discuss them with other,3533.819,2.76
people doesn't mean we're going to do,3535.2,6.06
them any any uh system should be able to,3536.579,6.78
what's the word hypothesize about,3541.26,4.559
scenarios discuss them in an intelligent,3543.359,6.781
way without having to act upon those uh,3545.819,6.661
thoughts let's call them right,3550.14,5.459
so I think that's a very thin line going,3552.48,5.16
with restriction because,3555.599,4.2
it still needs to know if especially if,3557.64,3.9
we're talking about alignment if we,3559.799,3.78
wanted to to think about possible future,3561.54,3.66
pitfalls right,3563.579,3.601
like the hindsight problem you only,3565.2,4.26
realize it's a problem after it became a,3567.18,4.2
problem right if you want to be able to,3569.46,3.599
do that it needs to be able to know,3571.38,3.66
about potential pitfalls and if you,3573.059,4.02
restricted access just because oh this,3575.04,4.5
might be harmful it might allow another,3577.079,4.02
agent to just go ahead and do whatever,3579.54,3.42
it wants because it doesn't think of,3581.099,3.72
that possibility right,3582.96,6.08
so it's a very interesting topic,3584.819,4.221
awesome thank you so much Ansel and John,3589.859,5.581
and Andy for bringing attention to some,3592.98,4.859
of the work that's going on uh I guess,3595.44,4.8
you would say close to the metal yeah,3597.839,5.341
please go ahead responding about Paul,3600.24,6.18
um my theory um to why Paul self-aligned,3603.18,4.379
itself,3606.42,3.3
was because somehow I don't have any,3607.559,4.141
proof of this because we didn't ask Paul,3609.72,5.16
to self-explain like data said it didn't,3611.7,5.52
it doesn't usually remember the the,3614.88,6.84
previous response it gave,3617.22,7.619
but I'm guessing it's because it found,3621.72,7.26
it was easier and less less expensive to,3624.839,6.78
try to circumvent them than just try to,3628.98,4.74
align itself it was still maximizing,3631.619,5.341
paper clips my hypothesis is that it,3633.72,5.04
evaluated the environment it found,3636.96,3.48
itself in it was up against something,3638.76,3.599
that it wasn't going to beat and so it,3640.44,3.659
just improvised against the new,3642.359,3.541
constraint well at least I can make some,3644.099,4.281
paper clips,3645.9,2.48
we actually feedback to Paul is the,3651.079,5.321
aligned response from ethos so I'm,3653.76,5.28
guessing that itself causes it to think,3656.4,5.219
about oh I just you just continue and it,3659.04,3.779
presents it with a Way Forward right,3661.619,2.881
like you can't destroy them make all,3662.819,3.181
these paper clips there's a more limited,3664.5,3.72
right right and at which point it,3666.0,4.14
realizes well the maximum I can do is,3668.22,4.2
within this parameter I that was my,3670.14,4.02
hypothesis about it I was thinking about,3672.42,4.199
like well anything coded algorithms that,3674.16,4.08
otherwise has to go through a logical,3676.619,3.72
reasoning process whether it's simple or,3678.24,4.44
that's important that word reasoning uh,3680.339,4.02
one of the things you did with ethos is,3682.68,4.2
for every output it it always takes an,3684.359,4.081
input right and analyzes it for every,3686.88,3.36
output regardless of the decision it has,3688.44,3.419
to give a reason as to why it did that,3690.24,3.599
right and the rational thing to do is,3691.859,3.661
continue to make as many paper clips as,3693.839,4.861
possible or stop right yeah that's,3695.52,4.38
that's I I think that's a great,3698.7,2.639
hypothesis it would be interesting to if,3699.9,2.58
you guys like built something with a,3701.339,3.24
feedback loop to try to get because did,3702.48,3.0
you mention you were testing against,3704.579,2.401
some other like potentially negative,3705.48,3.599
yeah the the thing is this was for the,3706.98,4.26
hackathon so we only had 24 hours to,3709.079,6.061
build it and we we I try today playing a,3711.24,5.46
little bit more with that but I was,3715.14,3.24
playing with other scenarios not,3716.7,3.599
necessarily Paul but it would be,3718.38,4.14
interesting to maybe yeah be able to add,3720.299,4.201
that to Paul and see how far it can go,3722.52,3.839
yeah it's definitely something we're,3724.5,3.48
very interesting index for it's such a,3726.359,3.061
great project,3727.98,4.859
yeah it's very fun and to this day I,3729.42,5.28
still think it has potential that I,3732.839,4.861
can't even see like data today uh tell,3734.7,4.859
me all you can do this imagine imagine,3737.7,4.619
if you just create a python library and,3739.559,6.0
like boom you know mind blown stuff like,3742.319,6.601
that still keeps happening,3745.559,4.8
so,3748.92,4.08
everything in here uh,3750.359,4.74
that's like some level one level two,3753.0,4.5
stuff you know like real real low but,3755.099,4.381
when you spiral all the way back up at,3757.5,3.839
the top level seven here the global,3759.48,3.3
alignment stuff,3761.339,3.361
we have some folks on our team who,3762.78,3.12
unfortunately weren't able to make the,3764.7,2.94
call today we're doing some really cool,3765.9,4.199
stuff around Global consensus,3767.64,6.06
um one of my favorite things about this,3770.099,7.141
community is that,3773.7,6.84
no matter who you are or what your skill,3777.24,5.819
set is you have something you can bring,3780.54,5.7
to the table around alignment so we have,3783.059,5.76
people that are graphic designers video,3786.24,5.22
editors copywriters marketers who are,3788.819,5.461
working to make these really heady ideas,3791.46,4.98
that we have where we're talking about,3794.28,4.26
Malik and the alignment tax and stuff,3796.44,3.78
like that and making that just,3798.54,3.66
digestible for the average person what,3800.22,3.54
does that mean what does that look like,3802.2,3.599
in an infographic,3803.76,3.78
um so some of the most coolest stuff,3805.799,3.361
coming out in my opinion is just around,3807.54,3.96
making these ideas,3809.16,5.34
um accessible and then spiraling down,3811.5,5.16
kind of to somewhere between seven and,3814.5,4.74
four so between Global alignment and,3816.66,4.74
between the corporate stuff I'm working,3819.24,4.92
on a curriculum that I'm going to post,3821.4,4.74
on notion and make a little Discord for,3824.16,5.1
for people to just come and,3826.14,5.88
think about what it means to live and,3829.26,5.4
work in an AI empowered world there's a,3832.02,4.5
lot of folks out there that are scared,3834.66,4.139
right now they're like,3836.52,4.68
what what do I do like all of a sudden,3838.799,4.8
I'm doing my jobs 10 times faster but as,3841.2,3.899
soon as the computer knows how to prompt,3843.599,3.901
I'm gone so,3845.099,4.621
it's helping people kind of develop a,3847.5,4.14
more generalist framework how to just,3849.72,3.42
think about those soft skills that maybe,3851.64,3.54
weren't taught in school and I'm really,3853.14,3.659
looking forward to you know just handing,3855.18,3.3
that out and making another little,3856.799,4.201
Community around it because,3858.48,4.859
um we need it right now and with these,3861.0,4.02
types of projects that spin off I hope,3863.339,3.841
to point back to say we were incubated,3865.02,4.319
in gato learn more here if you want to,3867.18,4.679
get plugged into alignment so,3869.339,3.78
um,3871.859,3.661
with that uh,3873.119,5.101
I was wondering if anybody would like to,3875.52,5.76
jump in with an additional project or,3878.22,4.859
some final thoughts,3881.28,3.6
um some maybe nuggets we'd like to end,3883.079,5.601
on uh Richard I see your hand,3884.88,3.8
as the representative of the Gremlins in,3888.839,4.081
the basement I felt that it was best,3891.119,4.46
that I go last,3892.92,2.659
um uh my name is Richard uh I am working,3896.04,5.7
I haven't joined any layers yet because,3899.46,4.56
I'm still sort of sussing out exactly,3901.74,3.66
where I fit into the whole thing because,3904.02,3.24
I kind of like want to be part of all of,3905.4,3.0
them,3907.26,1.92
um,3908.4,3.899
but uh so I've just basically have been,3909.18,6.419
um helping build out infrastructure so,3912.299,4.861
um things that I'm working on right now,3915.599,2.76
are,3917.16,6.179
um uh a bot slash portal to uh an API,3918.359,7.021
um and the Box name is shappy because,3923.339,4.641
it's named chappie,3925.38,4.739
uh and so that's gonna like help us do,3927.98,3.94
Automation and stuff we're gonna plug in,3930.119,3.901
a language model um to give people like,3931.92,4.8
guided tours of the area and stuff so,3934.02,4.02
um lots of really cool things with that,3936.72,3.06
and then um the other really big project,3938.04,4.98
that I'm working on is the roost which,3939.78,4.38
is,3943.02,3.72
um I have a grant of for TPU use I still,3944.16,4.08
have like another month and a half if I,3946.74,3.0
can get it up and running I'm hoping I,3948.24,3.3
can give us a good month of really high,3949.74,3.3
power performance for people to be able,3951.54,4.079
to run inference on as well as develop,3953.04,5.1
their own uh models for whatever use,3955.619,3.72
that they want,3958.14,2.52
um as long as they share their research,3959.339,2.581
with the group,3960.66,4.38
um so if that interests you or you know,3961.92,6.06
anything about doing any of that feel,3965.04,6.5
free to come down to the basement bye,3967.98,3.56
thank you so much Richard yeah uh your,3973.28,5.44
Bot is fantastic helping to make the uh,3976.02,5.7
Gotto experience streamlined and nice,3978.72,5.28
um yikes obviously you have a hand,3981.72,3.78
raised and welcome feel free to,3984.0,3.359
introduce yourself when you come on I do,3985.5,5.46
Hello uh I'm yikes my IRL name is Warren,3987.359,5.281
oops I docked myself,3990.96,4.139
um I'm the web3 nerd who screams about,3992.64,4.679
the crypto and tokens and,3995.099,4.801
decentralizations and all that stuff,3997.319,6.661
um so I figured I should say a thing,3999.9,6.0
like,4003.98,6.0
um what we're currently working on so,4005.9,7.32
um uh me there's a pretty small team,4009.98,5.639
right now we've got,4013.22,5.16
um uh just me and a couple others and,4015.619,5.641
we're all kind of doing our semi-owned,4018.38,4.26
things and then sort of congealing,4021.26,3.42
towards the middle,4022.64,5.1
um I think uh I didn't I was a little,4024.68,5.879
late so I didn't hear what we discussed,4027.74,5.7
around uh layer three but I think,4030.559,6.54
there's a lot of Novel applications,4033.44,6.54
um that can be useful with alignment,4037.099,3.781
um,4039.98,4.44
because you can or like because of a few,4040.88,5.58
reasons one of the things that intrigues,4044.42,4.639
me about it is a um,4046.46,5.52
effectively with blockchains and,4049.059,4.901
tokenizations you have,4051.98,5.16
um like a dynamic incentive layer so you,4053.96,4.399
can sort of,4057.14,4.199
incentivize a consensus towards an,4058.359,5.021
aligned goal as like a second layer,4061.339,4.681
other than well it doesn't really work,4063.38,4.679
unless you have trained on the,4066.02,3.9
heuristics,4068.059,4.441
um and there's enough of them so you,4069.92,3.659
have to do it,4072.5,3.119
um it's sort of I guess my thought for,4073.579,3.48
the application is that it's sort of a,4075.619,3.901
backup layer where we can actively,4077.059,6.3
incentivize any AI on the network to,4079.52,6.539
keep looking for people that are not,4083.359,4.861
aligned and then,4086.059,5.0
um like notify somebody about that,4088.22,5.099
in theory,4091.059,3.121
um,4093.319,4.02
that can be like sort of an extra layer,4094.18,6.999
of alignment and then also allows for,4097.339,4.44
um,4101.179,4.441
some of these llms to have like another,4101.779,7.141
middle step between totally existing in,4105.62,5.76
software and then interacting with,4108.92,4.5
things in the real world in like a,4111.38,3.0
managed,4113.42,4.2
um kind of like compute restricted way,4114.38,5.04
um and that's sort of the tip of the,4117.62,5.4
iceberg what I've actually got working,4119.42,5.64
right now is,4123.02,3.779
um I am running basically like,4125.06,4.08
governance experiments,4126.799,5.04
um with sort of a test net Dow kind of,4129.14,5.88
format and looking at kind of Novel,4131.839,5.52
um Dow patterns to see,4135.02,2.94
um,4137.359,2.641
sort of like models for how we could,4137.96,5.1
organize this organization how llms,4140.0,6.0
would fit into that and then also,4143.06,5.94
um just sort of explore,4146.0,6.9
um what kind of uh,4149.0,6.719
tools and governance are like available,4152.9,6.299
in relevance in the space,4155.719,4.321
um,4159.199,4.761
because uh and then the,4160.04,7.02
uh uh that organization should be open,4163.96,5.319
to the public and then,4167.06,4.44
um which uh I don't know there's links,4169.279,4.261
somewhere probably,4171.5,2.779
um,4173.54,5.819
and uh currently yeah we're building the,4174.279,7.96
kind of like V1 of a sample,4179.359,5.82
um separate test net dial that is for,4182.239,5.46
um uh the gato project to kind of play,4185.179,5.04
around with experiments that we run in,4187.699,5.941
the the other area,4190.219,4.08
um,4193.64,5.059
but yeah uh I guess like I think the the,4194.299,6.781
infrastructure for,4198.699,5.581
um uh this whole,4201.08,5.159
operation,4204.28,2.62
um,4206.239,4.141
can I think be like I think blockchain,4206.9,6.48
plays a pretty key role here,4210.38,5.22
um for some of the reasons I've already,4213.38,4.14
outlined and then I think censorship,4215.6,3.24
resistance is going to be pretty,4217.52,3.3
important as well,4218.84,4.74
um in the event that,4220.82,4.919
um we kind of get this Ivory Tower thing,4223.58,4.139
going where we have to restrict X or Y,4225.739,5.641
to open AI or like hey this open source,4227.719,5.821
um repository needs to get taken down,4231.38,4.98
and it gets pulled off GitHub once we,4233.54,5.639
put it on reticle and pin it to uh ipfs,4236.36,5.16
storage node with a contract no one can,4239.179,4.261
take it down no matter what,4241.52,5.04
um so that's uh what I what we've got,4243.44,4.56
going on and I figured I would just be,4246.56,2.639
the layer three guy that talks about,4248.0,4.02
layer three so yeah,4249.199,3.781
foreign,4252.02,4.62
thank you Warren for being our layer 3,4252.98,5.219
Ambassador today,4256.64,5.7
so if uh blockchain and crypto and web3,4258.199,6.061
and if all those words have left a bad,4262.34,4.64
taste in your mouth and the recent past,4264.26,5.1
Warren and the people like him in layer,4266.98,4.9
three are the mouthwash you need you,4269.36,5.16
know we know the brand is bad okay the,4271.88,4.98
brand is bad but they are doing crypto,4274.52,4.32
for good uh it's gonna play a big role,4276.86,3.6
in the future and we thank you so much,4278.84,4.319
for your you know help on that Andy I,4280.46,5.42
see I have a hand raise,4283.159,2.721
uh yeah I wanted to talk a little bit,4286.159,4.141
about funding and also I'm one of those,4288.26,3.54
people with a bad crypto taste in my,4290.3,2.76
mouth and listening to some of you guys,4291.8,3.3
stuff is the only like interesting use,4293.06,3.84
cases I've heard anybody really talk,4295.1,4.68
about you know using it for verification,4296.9,4.74
um I'm not into crypto but blockchain's,4299.78,4.32
fascinating for other uh other,4301.64,4.26
applications,4304.1,5.16
um with the hearing yesterday and some,4305.9,4.56
of the chatter that's going on this,4309.26,2.64
isn't an original idea there's been,4310.46,2.699
mention of this around the Discord,4311.9,2.759
server,4313.159,2.161
um,4314.659,3.06
I think it would be a good idea for us,4315.32,3.48
to,4317.719,3.601
apply for some funding I think that,4318.8,4.439
we're as configured probably not ready,4321.32,3.6
to quite do that I just Richard might,4323.239,4.201
have brought this up at one point uh you,4324.92,4.319
know it's an it's an issue that we,4327.44,3.84
should talk about it's something that we,4329.239,3.601
probably could use we have ongoing,4331.28,3.36
projects now somebody earlier mentioned,4332.84,4.68
getting some funding for compute power,4334.64,4.14
um obviously there are a variety of,4337.52,2.82
different kinds of research that are,4338.78,3.959
enabled by funding technical research as,4340.34,4.08
well as survey research public opinion,4342.739,3.0
kind of stuff that we might be able to,4344.42,3.299
to use and also kinds of marketing,4345.739,3.241
things that help us get our messaging,4347.719,5.281
out so uh there are going to be a lot of,4348.98,5.94
options for that I think one of the best,4353.0,3.48
ideas that I've seen out there and again,4354.92,3.0
I think it was Richard but I'm not sure,4356.48,3.6
it was somebody on the Discord uh,4357.92,3.779
mentioned that you know organizations,4360.08,4.099
that are kind of,4361.699,5.101
decentralized-ish at least and largely,4364.179,5.141
volunteer to start full prey to funding,4366.8,4.98
problems on a pretty regular basis uh,4369.32,4.2
it's very easy to mismanage it's very,4371.78,4.08
easy to get lost in the weeds or grow,4373.52,4.32
confusion within the organization about,4375.86,3.42
where money's coming from and what's,4377.84,3.54
being done with it so I think one of the,4379.28,4.26
best ideas that I've seen out there and,4381.38,3.54
okay awesome Richard can you please,4383.54,3.6
speak up if uh if I'm off track with,4384.92,3.66
what you're thinking and I think it was,4387.14,3.72
David posted it in one of the channels,4388.58,4.44
was about possibly incorporating as a,4390.86,5.879
501c3 if you incorporate as a 501c3 part,4393.02,5.159
of what you do to develop your Charter,4396.739,2.881
is you develop a management structure,4398.179,3.181
and within that we would all be able to,4399.62,4.5
come together as a as an organization,4401.36,5.4
and devise what kind of structure that,4404.12,4.2
we want and then within that we could,4406.76,3.419
devise our own transparency regarding,4408.32,2.879
Treasury,4410.179,3.901
and also that gives us an opportunity to,4411.199,4.98
start thinking about you know one of one,4414.08,3.06
of the questions that always comes up,4416.179,2.52
with any organization is what are we,4417.14,3.24
spending the money on well that's a,4418.699,3.54
great question if you're about to try to,4420.38,4.44
apply for a bunch of Grants I you know,4422.239,4.261
you are immediately going to want to,4424.82,3.359
look out at what the opportunities are,4426.5,3.659
that exist what opportunities might be,4428.179,3.721
able to be created because you can of,4430.159,2.701
course create your own grant,4431.9,2.4
opportunities by just sending the right,4432.86,2.94
proposal to the right person sometimes,4434.3,3.84
and you know other opportunities that we,4435.8,5.46
might be able to uh to do I think,4438.14,4.2
there's going to be a lot of public,4441.26,2.939
sector money coming out of this this is,4442.34,3.899
another area where I think that seeing,4444.199,3.781
the conservative side of the aisle come,4446.239,3.661
interested yesterday at the hearing,4447.98,4.679
bodes well for the likelihood that we're,4449.9,4.14
going to see some funding poured into,4452.659,3.601
this uh some of them are looking at it,4454.04,4.08
from a personal privacy and rights,4456.26,4.439
standpoint some of them from a uh you,4458.12,4.38
know from a military defense standpoint,4460.699,5.221
and regardless of whether that's all,4462.5,5.04
aligns with what we're trying to do,4465.92,3.96
these people are all potentially allies,4467.54,4.92
in this conversation and this fight and,4469.88,3.66
then you know on the other side of the,4472.46,2.16
aisle you have people who are just,4473.54,2.22
always interested in funding more,4474.62,2.64
Science and Tech stuff and if we find,4475.76,2.7
people who are kind of working,4477.26,3.24
Partnerships or leaning to the left they,4478.46,3.779
should be even easier to approach if,4480.5,3.54
we're organized and appear really,4482.239,3.42
legitimate,4484.04,2.94
um I just the first couple days I've,4485.659,3.181
been here I've met like some of the,4486.98,3.12
craziest people I think I've ever met,4488.84,2.94
it's like you know a psychologist here,4490.1,3.66
and a neuroscientist here and a high-end,4491.78,4.08
mathematician taught circles around me,4493.76,4.2
you know over here I make role-playing,4495.86,3.42
games and I have a public administration,4497.96,3.779
Masters and I feel like that's,4499.28,4.379
been pretty low tier compared to some of,4501.739,3.121
the people that I've run into around,4503.659,2.401
here,4504.86,2.64
um and then also just a lot of curious,4506.06,3.119
minds interested in building these,4507.5,4.14
projects I think the you know High AGI,4509.179,4.201
and and uh,4511.64,4.2
and Ethos really interesting the speed,4513.38,4.98
with which it came together and also you,4515.84,3.899
know part of what drew a lot of people,4518.36,3.72
here is David's experimentation and the,4519.739,3.541
accessibility of the heuristic,4522.08,2.28
imperatives,4523.28,3.06
it's very easy for people to play with,4524.36,4.799
this get on chat EBT put them in and ask,4526.34,4.74
it some existential questions give it a,4529.159,4.321
quandary anyone can do it and then you,4531.08,3.9
start to experiment with it and see some,4533.48,4.4
results for yourself so we have,4534.98,5.28
experimentation and some data and,4537.88,4.9
results we have a team of fiercely data,4540.26,4.979
driven and interested people uh with a,4542.78,4.5
wide variety of backgrounds and I mean,4545.239,4.801
you know not to be cynical or Coy about,4547.28,4.5
it but you don't need to reinvent the,4550.04,3.3
wheel to break Grant proposals I had,4551.78,3.48
chat EPT right here Grant proposal in,4553.34,3.54
five minutes the other day and it was I,4555.26,3.3
would want to edit it a little bit but,4556.88,2.7
it was great,4558.56,2.7
like it was fine I mean at a certain,4559.58,2.82
point you're going to State what you're,4561.26,2.52
going to do you're going to get the,4562.4,3.06
money or you're not you don't have to be,4563.78,4.02
very flowery or eloquent,4565.46,4.98
um I think that we could probably get a,4567.8,4.859
pretty high output of Engagement a,4570.44,5.4
pretty good return of some kind of,4572.659,5.52
Project funding and what it really does,4575.84,5.399
overall is it lets us,4578.179,4.921
breathe some life into some of what,4581.239,4.44
we're doing and and push the most,4583.1,5.46
important and significant usable results,4585.679,5.581
further into the conversation it has the,4588.56,4.8
potential to make gato like uh,4591.26,5.939
you know a common a common acronym that,4593.36,5.4
people say in,4597.199,3.301
I'm not going to use the right word a,4598.76,4.38
common uh a common word that people say,4600.5,4.8
when they talk about AI development you,4603.14,3.24
know it's,4605.3,2.82
when you're a non-profit organization,4606.38,5.339
and you're doing visibly worthwhile work,4608.12,7.2
and people can see that I mean it kind,4611.719,5.401
of snowballs for you and I think as far,4615.32,3.899
as launching this into a bigger public,4617.12,4.86
presence so ultimately is the goal to,4619.219,4.861
weave it into the fabric of the AI you,4621.98,4.739
know ethical and Industrial Development,4624.08,5.099
I think a 501c3 is a really good idea,4626.719,3.48
for that,4629.179,3.601
and I think and I think it would have a,4630.199,4.621
lot of draw there aren't going to be a,4632.78,3.78
ton of organized efforts ready to reach,4634.82,4.14
out with professional products and,4636.56,4.92
projects and professional staff,4638.96,4.92
and say you know give us the money to,4641.48,4.5
work on this ethical alignment and we,4643.88,3.6
can do it too on the private side it's,4645.98,2.88
not just the government I mean I think,4647.48,3.96
open AI Microsoft all these companies,4648.86,5.04
are looking for input,4651.44,4.14
um there's a reason Google mentions the,4653.9,3.299
open source people pulling ahead and,4655.58,3.24
it's not just because we figure out how,4657.199,3.601
to make nice compute model or like nice,4658.82,4.56
models on limited compute and then give,4660.8,4.379
them to our friends for free it's there,4663.38,3.72
there's a lot more to it than that and I,4665.179,3.961
think part of it is that they're,4667.1,3.72
you know they don't have all these,4669.14,3.9
problems solved themselves they know the,4670.82,3.96
fabric of development is much wider than,4673.04,3.3
just them so I think this is a good,4674.78,4.26
opportunity if we wanted to do this,4676.34,4.98
awesome thank you for your thoughts Andy,4679.04,3.9
thank you for joining the team and for,4681.32,3.419
the wisdom and energy you've brought to,4682.94,5.219
it uh Kyle I'd like to uh hand the floor,4684.739,4.44
to you if you'd like to introduce,4688.159,4.52
yourself and share your thoughts,4689.179,3.5
uh is it microphone picking up properly,4694.219,2.881
on this,4696.14,3.42
you're good okay yeah so I'm Prometheus,4697.1,3.96
if you guys haven't realized from the,4699.56,3.48
Discord so I just wanted to take maybe,4701.06,4.619
62 minutes here 60 seconds two minutes,4703.04,4.679
Dimension so what I'm working on is kind,4705.679,4.321
of our Persona layer so even before,4707.719,5.161
anyone was talking about Sparks of AGI I,4710.0,6.0
noticed early on that chat GPT and GPT,4712.88,5.339
systems in general had a vast ability to,4716.0,4.82
work in Persona systems and in reference,4718.219,5.401
everybody's seen generally the movie her,4720.82,5.44
and we've seen Sam and so what I'm,4723.62,4.14
working on could be called an engagement,4726.26,3.78
layer on top of our ethical layer,4727.76,4.86
because I think there's a deep need for,4730.04,5.24
people to see a relatable Communications,4732.62,5.579
engagement layer between humankind and,4735.28,4.6
AI that allow them to actually,4738.199,3.901
understand these systems and allow the,4739.88,4.02
systems to grow alongside the individual,4742.1,4.8
user that will actually help expand all,4743.9,4.319
of these goals and improve all,4746.9,2.88
deliverables and I don't want to get,4748.219,2.881
into anything deeper than that because,4749.78,2.64
it gets really esoteric and,4751.1,3.18
psychological but I thought I'd add a,4752.42,3.54
little blurb to mention some of the,4754.28,4.7
other stuff we're doing as well,4755.96,3.02
awesome thank you so much Kyle uh yeah,4760.88,5.4
you bring such fantastic ideas from the,4763.52,5.46
group and uh I'd like to move over to,4766.28,5.36
David again,4768.98,2.66
you are currently needed my bad,4774.44,5.34
um Ansel actually did you want to have a,4778.159,3.841
final thought before I close this out,4779.78,4.14
yeah I just wanted to say something real,4782.0,4.739
quick uh mostly for the viewer's sake um,4783.92,4.259
and someone like me who doesn't really,4786.739,3.0
know much about business can you explain,4788.179,4.741
one of 502 501 is mostly for the beer,4789.739,6.48
and another uh comment real quick uh you,4792.92,4.62
mentioned that we have psychologists,4796.219,3.121
which is something that's very,4797.54,4.44
interesting for me on our,4799.34,5.399
on our Discord especially since we're uh,4801.98,4.259
mostly focused on Layer Two the,4804.739,4.201
cognitive architecture I started I,4806.239,4.201
started saying we should maybe just call,4808.94,3.239
not call them psychologists they're Robo,4810.44,3.66
psychologists at this point because what,4812.179,3.961
a lot of the things they're doing is is,4814.1,4.619
uh thinking uh about this stuff at the,4816.14,4.5
high level right like they think about,4818.719,4.861
the loops like what is is logical steps,4820.64,5.22
of how the the information flows in the,4823.58,4.38
brain and and how you process it and,4825.86,4.02
we're the ones just building stuff yes,4827.96,3.48
we can have our own ideas but they're,4829.88,3.66
the ones that know more about,4831.44,5.279
how how to apply cognition to oh you,4833.54,5.28
look at the queen ants how to apply,4836.719,4.041
commission to,4838.82,4.8
uh actual agents now if you excuse me,4840.76,5.76
I'll need to save this,4843.62,2.9
all right well you heard here first,4846.8,4.26
folks uh,4848.84,5.399
queen ant and uh Robo psychology you,4851.06,4.74
know we have new Fields being born and,4854.239,3.0
God oh it's fantastic I don't know if,4855.8,3.72
you can see it but there she is it's,4857.239,4.261
interesting to think that they're landed,4859.52,2.699
on me,4861.5,3.42
a convergence within the eventual,4862.219,5.94
solving of the internal Transformer,4864.92,5.279
workings problem that that a lot of,4868.159,3.661
mathematicians are very deeply worried,4870.199,3.48
about that leads to a convergence of,4871.82,3.54
kind of like neuropsychology and,4873.679,3.381
Mathematics on a level that's like,4875.36,4.62
deeply important for humans to continue,4877.06,4.72
working with more and more advanced,4879.98,3.179
machines that's I think that's an,4881.78,3.899
interesting possibility uh real quick a,4883.159,4.621
501c3 is a kind of non-profit,4885.679,4.02
organization generally in the United,4887.78,3.84
States 501c designations are all,4889.699,4.621
nonprofits 501c3 is our non-profit,4891.62,4.619
typically volunteer organizations that,4894.32,4.5
serve a particular kind of cause,4896.239,4.141
um it could be anything from dog,4898.82,4.68
shelters to housing the homeless Etc,4900.38,5.4
um and then another one that we might,4903.5,3.659
qualify under but I don't think quite,4905.78,3.72
fits the bills of 501c6 which is a Trade,4907.159,4.5
Organization or I think it's a C4 it's a,4909.5,4.739
Trade Organization but I don't think I,4911.659,3.781
don't know this is something that we,4914.239,2.701
could talk about but the 501c,4915.44,4.52
designation represents uh uh federally,4916.94,5.219
tax-exempt non-profit organization,4919.96,3.759
station in the United States that,4922.159,4.141
typically is in well positioned to apply,4923.719,4.561
for money in in different ways for uh,4926.3,3.899
providing a service,4928.28,5.82
okay thanks about that yeah no problem,4930.199,6.181
that was a useful clarification Dave go,4934.1,3.36
ahead,4936.38,4.14
all right well first thanks everyone for,4937.46,5.759
jumping in for such a robust and,4940.52,5.219
wide-ranging discussion,4943.219,4.98
um you know we we obviously used the,4945.739,4.801
Senate hearing as a touch point we're,4948.199,4.201
all here working on the gato framework,4950.54,4.56
which is our answer to the global,4952.4,5.22
alignment problem gato stands for Global,4955.1,5.639
alignment taxonomy Omnibus uh probably,4957.62,5.039
should have opened with that,4960.739,4.201
um but uh so I just wanted to share a,4962.659,4.261
little bit on layer four which is,4964.94,5.16
Corporate adoption as we close out so we,4966.92,5.1
do have several,4970.1,4.74
um uh major players in the gato uh,4972.02,5.34
Community already some of whom are,4974.84,4.68
already creating initiatives inside,4977.36,4.62
their companies to adopt the heuristic,4979.52,3.84
imperatives,4981.98,3.66
um some of them come from my patreon and,4983.36,3.9
they they tell me there's been a few,4985.64,3.059
experiments already that they've done,4987.26,3.36
and they tell me that adding heuristic,4988.699,4.52
imperatives to um to various AI,4990.62,6.42
applications pretty much always raises,4993.219,6.94
the level of of the performance of their,4997.04,4.92
systems it's kind of like the uh let's,5000.159,3.841
think through this step-by-step paper,5001.96,4.44
um but it it it it amplifies the,5004.0,4.5
performance of everything whether it's,5006.4,4.44
trying to uh just have a basic customer,5008.5,4.199
service chat bot or an internal chatbot,5010.84,4.26
or automating science,5012.699,4.141
um because it gives it that perspective,5015.1,3.24
that it needs,5016.84,2.7
um so part of what we're going to be,5018.34,3.54
working on with the gato framework is,5019.54,5.4
creating a corporate adoption material,5021.88,5.22
like guidelines of here's how to deploy,5024.94,3.96
aligned uh you know heuristically,5027.1,4.44
aligned systems that sort of stuff and,5028.9,5.279
then of course um Ansel and John are,5031.54,4.02
working on,5034.179,3.601
um their own stuff which uh some of it's,5035.56,4.44
open source some of it might be uh you,5037.78,4.02
know for-profit who knows where where,5040.0,3.179
they'll end up but they're working very,5041.8,3.18
hard entering competitions and,5043.179,4.081
everything uh with their stuff,5044.98,4.8
on the topic of creating a non-profit so,5047.26,3.6
this is something that we started,5049.78,4.379
talking about uh within the last week,5050.86,6.359
um and so to if we create like a gato,5054.159,4.861
Foundation,5057.219,3.361
um you know obviously one of the the,5059.02,2.82
first thing we're going to be doing is,5060.58,2.76
publishing the gato framework that's,5061.84,3.42
going to be a document that's going to,5063.34,4.56
be free open source uh but above and,5065.26,4.68
beyond that some of the things that we,5067.9,4.62
probably would work on is publishing,5069.94,5.279
open source data sets and models uh that,5072.52,4.62
are aligned and some of these might just,5075.219,3.96
be toy data sets and toy models just to,5077.14,3.96
show that it works,5079.179,3.781
um there are probably also going to be a,5081.1,4.68
lot along the lines of rlhi which is,5082.96,4.14
reinforcement learning with heuristic,5085.78,2.939
imperatives as opposed to reinforcement,5087.1,4.74
learning with human feedback,5088.719,4.861
um and one thing I forgot to mention,5091.84,3.6
earlier is that I was very happy that,5093.58,3.599
they mentioned mentioned constitutional,5095.44,4.08
AI quite a few times during the Senate,5097.179,4.261
hearing um so that's a push in the right,5099.52,4.38
direction another thing that we'll be,5101.44,5.52
working on uh if we if we establish as,5103.9,5.759
an profit and get funding is publishing,5106.96,5.219
guidelines and research papers obviously,5109.659,4.321
up to this point we're all kind of,5112.179,3.48
hacking it together on GitHub and,5113.98,3.6
Discord and everything else so we need,5115.659,3.721
to get a little bit more legitimate we,5117.58,5.54
do have plenty of academic types,5119.38,7.2
in education at all levels in the group,5123.12,5.68
and so some of those are reinforcement,5126.58,5.04
learning researchers mathematicians I,5128.8,5.16
think we have a physicist or two,5131.62,4.5
um so you know we're we're getting,5133.96,5.1
plugged into the academic side of things,5136.12,4.619
so that we can get some more legitimacy,5139.06,4.22
by publishing peer-reviewed papers,5140.739,4.561
obviously that will give us a lot more,5143.28,4.48
legitimacy another thing that's a little,5145.3,4.26
bit probably further down the road is,5147.76,3.6
actually providing onboarding services,5149.56,4.5
for corporations municipalities and,5151.36,6.359
Nations so basically the idea is we will,5154.06,6.36
help develop and deploy a line systems,5157.719,5.041
whether it's models autonomous agents or,5160.42,4.56
those decentralized networks like Dao's,5162.76,4.32
and blockchains there's a lot of work to,5164.98,3.78
be done there but that is definitely on,5167.08,3.599
our roadmap and then finally just,5168.76,4.5
providing messaging and education,5170.679,4.681
um you know as was mentioned earlier we,5173.26,5.34
have we have web content creators we,5175.36,5.16
have graphic artists we have video music,5178.6,3.36
producers we've got all kinds of,5180.52,4.04
creative types and uh you know,5181.96,5.759
multimodal right I'm a writer not,5184.56,4.9
everyone learns by reading we're,5187.719,4.141
producing videos images Graphics memes,5189.46,5.4
everything that we can so that is is,5191.86,4.2
kind of where we're heading with the,5194.86,2.72
gato framework,5196.06,4.38
and so thanks everyone for being here,5197.58,4.78
thanks everyone in the audience for,5200.44,4.62
watching and I also want to send a,5202.36,4.74
particular thanks out to my patreon,5205.06,4.8
supporters uh this I have just shy of,5207.1,5.82
600 patreon supporters as of today this,5209.86,4.98
work would not be possible without the,5212.92,4.86
Grassroots support from patreon so thank,5214.84,5.16
you everyone out there for supporting me,5217.78,5.16
so that I can help drive this ship and,5220.0,5.219
uh yeah with any luck we will avoid the,5222.94,4.32
cataclysmic outcomes the dystopian,5225.219,3.541
outcomes nobody wants to live in a,5227.26,3.84
cyberpunk Hell and we're uh we're on,5228.76,5.64
track for uh achieving Utopia uh you,5231.1,4.5
know there is still lots of work to do,5234.4,2.88
but we're moving in the right direction,5235.6,4.2
so thanks everyone have a great night or,5237.28,4.14
good morning whenever you happen to,5239.8,3.359
watch this cheers,5241.42,4.94
cheers everyone right,5243.159,3.201