The Open AI Whistleblower: What Really Happened to Suchir Balaji?

1
00:00:00,040 --> 00:00:02,920
Hello, bizarre story friends.
If you're tuned in now, you are

2
00:00:02,920 --> 00:00:05,080
listening to The Bizarre AF, a
place where we talk about the

3
00:00:05,080 --> 00:00:08,600
strange, the unusual, the
unknown, and all things bizarre

4
00:00:08,600 --> 00:00:13,080
AFI am Alicia, your Hostess for
today's episode, and as always,

5
00:00:13,080 --> 00:00:14,560
we ask that you keep an open
mind.

6
00:00:14,840 --> 00:00:17,400
Keep a skeptical ear, but keep
on listening to those facts as

7
00:00:17,400 --> 00:00:21,480
we take you on our newest
journey, The Open AI

8
00:00:21,480 --> 00:00:24,840
Whistleblower.
What really happened to SU Chair

9
00:00:24,840 --> 00:01:02,720
Balgi?
Well, hello darling, how are you

10
00:01:02,720 --> 00:01:05,960
doing today?
Hey Bubba, I'm doing great.

11
00:01:06,320 --> 00:01:09,320
What's new in your life?
God, a lot.

12
00:01:09,400 --> 00:01:11,880
And by a lot, I mean, I have
absolutely no idea.

13
00:01:12,000 --> 00:01:14,640
You know, sometimes I just feel
like I'm wandering through the

14
00:01:14,640 --> 00:01:18,160
world, just clueless.
Waiting for an alien invasion?

15
00:01:18,200 --> 00:01:20,480
Oh my God.
Please, I am please.

16
00:01:20,800 --> 00:01:23,320
Please.
I mean, I'm definitely not, but

17
00:01:23,320 --> 00:01:26,080
the more and more time goes by,
the more and more I'm like,

18
00:01:26,440 --> 00:01:29,680
could it really be any weirder?
Could it really be any worse?

19
00:01:30,520 --> 00:01:32,640
Could it be?
Would I hate it that much?

20
00:01:32,640 --> 00:01:36,160
Could it be better?
Could it be in fact be better?

21
00:01:36,240 --> 00:01:40,600
Yeah, I think, I think that is
that is the name of the game.

22
00:01:40,600 --> 00:01:44,680
And on top of that, you know,
it's just like the holidays, you

23
00:01:44,760 --> 00:01:46,480
know?
You know, things are crazy busy

24
00:01:46,760 --> 00:01:51,360
and we're, you know, going from
one feast to another, rolling

25
00:01:51,360 --> 00:01:53,920
down the fucking road.
Do you feel like that?

26
00:01:54,040 --> 00:01:58,320
Yeah, Yeah.
Yeah, although as I think as I

27
00:01:58,360 --> 00:02:01,560
get older, I feel like I'm doing
less of that and just being more

28
00:02:01,560 --> 00:02:07,560
like I say Hermity because that
sounds awful, but more just kind

29
00:02:07,560 --> 00:02:10,280
of cozy at home kind of tangent,
right?

30
00:02:10,320 --> 00:02:12,640
Tangent.
So, you know, today I actually

31
00:02:12,640 --> 00:02:16,280
wanted to talk about
whistleblowers and we actually

32
00:02:16,280 --> 00:02:17,640
talk.
About Thing more and more, yeah.

33
00:02:17,920 --> 00:02:19,960
Yeah, it's become more and more
and we actually talk about

34
00:02:19,960 --> 00:02:22,080
whistleblowers all the time on
this, on this podcast.

35
00:02:22,080 --> 00:02:24,840
It's been, it's been steadily
that we've increased it, right.

36
00:02:25,040 --> 00:02:28,840
And ultimately, like, they're
essential to run an effective

37
00:02:28,840 --> 00:02:32,720
democracy, and they really do
play a pivotal role in ensuring

38
00:02:32,720 --> 00:02:36,320
that absolute power does not
corrupt absolutely, right?

39
00:02:36,320 --> 00:02:38,680
Exactly.
So today we are talking about a

40
00:02:38,680 --> 00:02:41,400
whistleblower, not in the alien
space, which is typically where

41
00:02:41,400 --> 00:02:45,560
where this would come from, but
in the artificial intelligence

42
00:02:45,560 --> 00:02:49,160
space.
This story is about Suchir Balji

43
00:02:49,480 --> 00:02:52,280
and he was a 26 year old
engineer who actually worked

44
00:02:52,280 --> 00:02:54,640
inside of open AI.
He helped.

45
00:02:54,640 --> 00:02:57,200
ChatGPT that we all use today.
Yes, exactly.

46
00:02:57,360 --> 00:03:02,080
He helped build the systems
behind ChatGPT, and then he

47
00:03:02,080 --> 00:03:05,160
publicly questioned whether or
not the company's practices were

48
00:03:05,160 --> 00:03:06,840
even legal.
Interesting.

49
00:03:07,160 --> 00:03:08,440
Legal.
OK.

50
00:03:08,480 --> 00:03:10,160
All right.
Yeah, whether they're they're,

51
00:03:10,320 --> 00:03:11,240
whether or not they're.
Illegal.

52
00:03:11,280 --> 00:03:13,640
OK.
And weeks later, he was gone.

53
00:03:15,040 --> 00:03:16,960
What do you mean like gone from
that planet?

54
00:03:18,120 --> 00:03:19,840
Oh, no.
So this episode is not about

55
00:03:19,840 --> 00:03:21,600
sensationalism.
I just wanted to make that

56
00:03:21,600 --> 00:03:24,200
clear.
It's really about tracing the

57
00:03:24,200 --> 00:03:29,240
facts and understanding what he
actually said and exploring, you

58
00:03:29,240 --> 00:03:33,600
know, the ethical fault lines
that he revealed in one of the

59
00:03:33,600 --> 00:03:35,800
world's most powerful companies.
Fascinating.

60
00:03:35,800 --> 00:03:39,480
You can't wait, OK?
So suit your Balgi, he was the

61
00:03:39,480 --> 00:03:43,800
kind of, you know, person that
silicone and valley like wet

62
00:03:43,800 --> 00:03:47,680
dreams about.
Of course, 1998, raised in

63
00:03:47,680 --> 00:03:53,040
Florida and then moves to
Cupertino, CA with his parents

64
00:03:53,040 --> 00:03:55,920
who are also, you know, in IT.
But he actually builds his first

65
00:03:55,920 --> 00:03:59,960
computer at 13 and ends up
winning the programming Olympics

66
00:03:59,960 --> 00:04:02,720
essentially by high school.
So it's the the programming

67
00:04:02,720 --> 00:04:04,680
Olympic odds.
Have you heard of those?

68
00:04:05,400 --> 00:04:07,400
Sure.
OK, yes, yes, yes, yes.

69
00:04:07,640 --> 00:04:10,520
They get all the nerds together
and they build great stuff.

70
00:04:11,760 --> 00:04:14,480
Not to go off on tangent, but
the software that I develop on,

71
00:04:14,480 --> 00:04:17,560
they have these hackathons,
yeah.

72
00:04:17,640 --> 00:04:19,760
So same kind of thing.
You get them all together, you

73
00:04:19,760 --> 00:04:22,880
can see who can build shit.
He built his first computer at

74
00:04:22,880 --> 00:04:24,720
13 and he was winning by high
school.

75
00:04:24,800 --> 00:04:26,400
These are these are like cash
prizes.

76
00:04:26,400 --> 00:04:29,880
Often it's like $100,000.
He ends up winning in fucking

77
00:04:29,880 --> 00:04:32,680
high school.
He ends up studying computer

78
00:04:32,680 --> 00:04:36,960
science at Berkeley and then
eventually joins Open AI.

79
00:04:36,960 --> 00:04:42,480
He works there between 2020 and
2024 and this information is

80
00:04:42,480 --> 00:04:44,960
found on his website.
So you know once again, not

81
00:04:44,960 --> 00:04:46,640
conjecture.
You know he wasn't a public

82
00:04:46,640 --> 00:04:52,560
figure, but people inside of the
company of of open AI knew he

83
00:04:52,560 --> 00:04:55,480
was amazing.
He was like a prodigy at his

84
00:04:55,480 --> 00:04:57,120
job.
He was really good at what he

85
00:04:57,120 --> 00:04:59,400
did.
John Shulman, who which is what

86
00:04:59,400 --> 00:05:01,400
was one of the open AI Co
founders.

87
00:05:01,880 --> 00:05:06,560
He later said that Balji's
contributions were essential and

88
00:05:06,560 --> 00:05:08,480
it wouldn't have succeeded
without him.

89
00:05:08,480 --> 00:05:11,680
He helped gather and organize a
huge data sets.

90
00:05:11,680 --> 00:05:15,080
It was really the raw material
behind open AI which was used to

91
00:05:15,080 --> 00:05:19,720
train the large language models
like like the GPT 3 and GPT 4.

92
00:05:20,000 --> 00:05:24,360
So, you know, if you don't use
ChatGPT and you live under a

93
00:05:24,360 --> 00:05:28,560
rock, there have been multiple
iterations of ChatGPT.

94
00:05:28,640 --> 00:05:32,000
Right, Get it better every time.
The language model just gets

95
00:05:32,000 --> 00:05:34,160
better and better and better,
because why?

96
00:05:34,520 --> 00:05:37,640
It learns from itself.
That's right, more data is

97
00:05:37,640 --> 00:05:40,480
putting in it's continue, it's
continuing learning.

98
00:05:40,480 --> 00:05:44,400
So, so he was the one who
originally was like helping, you

99
00:05:44,400 --> 00:05:48,200
know, gather all this data and
training ultimately that these

100
00:05:48,200 --> 00:05:49,800
models.
And this is really important

101
00:05:49,800 --> 00:05:54,320
because the same thing that he
helped build is what he later is

102
00:05:54,320 --> 00:05:55,760
questioning.
Right.

103
00:05:56,120 --> 00:05:58,800
We kind of hear that with smart
scientists all the time.

104
00:05:58,840 --> 00:06:03,360
Like, they build something and
then they question like, uh oh,

105
00:06:03,800 --> 00:06:06,520
was this a good idea?
Yeah.

106
00:06:06,680 --> 00:06:08,400
But quite honestly, if they
wouldn't have built it, somebody

107
00:06:08,400 --> 00:06:11,160
else would have anyway.
So, you know, they can't blame

108
00:06:11,160 --> 00:06:14,960
themselves.
But yeah, you see that all the

109
00:06:14,960 --> 00:06:16,960
time.
Or then the government will get

110
00:06:16,960 --> 00:06:20,320
a hold of it and go, you know,
people like we say, God, let's

111
00:06:20,320 --> 00:06:22,880
take us like broken records.
People use things for nefarious

112
00:06:22,880 --> 00:06:24,520
reasons.
Good things for nefarious

113
00:06:24,520 --> 00:06:27,440
reasons, yes.
Right, You can use things for

114
00:06:27,480 --> 00:06:30,720
evil and then for good, and
there are an awful lot of people

115
00:06:30,720 --> 00:06:34,800
who want to use it for nefarious
reasons.

116
00:06:35,160 --> 00:06:38,360
Especially to get money.
And money is yes, and we will be

117
00:06:38,360 --> 00:06:40,480
talking about that.
You don't say.

118
00:06:40,480 --> 00:06:46,080
So Speaking of opening, I did
begin as a non profit and they

119
00:06:46,200 --> 00:06:48,920
said that they it's to benefit
all of humanity.

120
00:06:48,920 --> 00:06:51,040
Sure, it seems like that's what
it would be.

121
00:06:51,040 --> 00:06:52,640
Yeah, that kind of a place,
right?

122
00:06:53,040 --> 00:06:55,360
The tool that all of us use on a
daily basis.

123
00:06:55,360 --> 00:06:56,960
I mean, I do, I can, I can say
that I do.

124
00:06:57,440 --> 00:07:00,880
It was initially A nonprofit
meant to improve humankind.

125
00:07:01,080 --> 00:07:05,160
But by the time that Suture
Balgi had joined, the structure

126
00:07:05,440 --> 00:07:07,240
of the company had already
split.

127
00:07:07,240 --> 00:07:12,640
So there was now a nonprofit
parent for profit arm and then a

128
00:07:12,640 --> 00:07:18,120
capped profit entity that
controlled both aspects of

129
00:07:18,320 --> 00:07:21,080
Chachi PT.
A profit controlling a non

130
00:07:21,080 --> 00:07:23,200
profit?
Yeah, that sounds like a

131
00:07:23,200 --> 00:07:25,200
conflict.
Yeah, it does.

132
00:07:25,640 --> 00:07:27,040
Yeah.
So you know, Kevin, have you

133
00:07:27,040 --> 00:07:30,160
actually been in a situation
where you joined a company and

134
00:07:30,760 --> 00:07:32,920
everything that they say that
they stand for, like you're

135
00:07:33,320 --> 00:07:35,360
interviewing, you're like, Oh my
God, yeah, this is great.

136
00:07:35,360 --> 00:07:39,600
This is like we're on the same
page and then as you get there,

137
00:07:39,840 --> 00:07:42,840
the reality of the company is
different, questionable.

138
00:07:42,880 --> 00:07:45,560
I have experienced that.
Yeah, and what was it like?

139
00:07:45,560 --> 00:07:47,280
Like when you discovered?
Oh my God.

140
00:07:47,320 --> 00:07:50,240
It's heartbreaking, it's
disappointing.

141
00:07:50,240 --> 00:07:56,840
You feel like betrayed, I guess,
almost like it's terrible.

142
00:07:56,840 --> 00:08:01,280
It's absolutely terrible.
And as I've learned, as I've

143
00:08:01,720 --> 00:08:08,520
gotten older and in my career, I
was like a workaholic, like if I

144
00:08:08,520 --> 00:08:11,120
work for somebody, like I was
Leon loyal.

145
00:08:11,440 --> 00:08:14,560
Yeah.
Like excessively loyal.

146
00:08:14,560 --> 00:08:18,440
And then as I've grown older,
I've realized they're not loyal

147
00:08:18,440 --> 00:08:19,320
to you.
They are.

148
00:08:19,520 --> 00:08:24,080
There has no loyalty as much as
they tell you how important you

149
00:08:24,080 --> 00:08:27,840
are and I'll no, yeah, no.
And so then when you get to the

150
00:08:27,840 --> 00:08:31,200
extreme of like, you know,
finding out that they are not

151
00:08:31,200 --> 00:08:33,200
even close to what they
represent, then that's even

152
00:08:33,200 --> 00:08:35,480
worse.
Oh my God, yeah, it is the

153
00:08:35,480 --> 00:08:37,320
worst.
It you're like, I was sold this

154
00:08:37,320 --> 00:08:41,640
one dream and then yes, it's not
what the reality is, right?

155
00:08:41,760 --> 00:08:45,760
That is something that I I joke
around with people in general.

156
00:08:46,160 --> 00:08:49,200
We're not saving babies.
This is not brain surgery.

157
00:08:49,680 --> 00:08:52,840
This is not like life or death.
Like, let's get with the

158
00:08:52,840 --> 00:08:54,840
freaking programs.
You know, we're here for a

159
00:08:54,840 --> 00:08:58,320
paycheck.
But Balji had actually told The

160
00:08:58,320 --> 00:09:03,280
Associated Press that it did not
sit right with him that the

161
00:09:03,280 --> 00:09:06,720
company was training on the
creative work of millions of

162
00:09:06,720 --> 00:09:08,640
people.
And then he they were releasing

163
00:09:08,640 --> 00:09:11,520
products that actually compete
with those people in the same

164
00:09:11,520 --> 00:09:12,800
marketplace.
OK.

165
00:09:13,240 --> 00:09:14,720
That's what he was having a
problem with.

166
00:09:15,520 --> 00:09:19,760
I'm sorry, hindsight's 2020, but
what did he expect was going to

167
00:09:19,760 --> 00:09:23,000
happen?
He was young and he's like, oh,

168
00:09:23,000 --> 00:09:25,120
this will be so great.
I'm getting all the data.

169
00:09:25,200 --> 00:09:27,040
Yes, I mean where?
Did he think he was getting the

170
00:09:27,040 --> 00:09:28,720
data from exactly?
Yeah.

171
00:09:29,000 --> 00:09:32,880
Yeah, naive, naive attack. 20,
He was in his mid 20s.

172
00:09:32,880 --> 00:09:34,640
Yeah.
You know, he had not learned

173
00:09:34,640 --> 00:09:36,560
about the fucking world yet,
right?

174
00:09:37,320 --> 00:09:39,880
So, you know, the more and more
he, he actually sees this

175
00:09:39,880 --> 00:09:43,800
company pivot further towards
commercialization, right?

176
00:09:44,040 --> 00:09:48,240
Licensing deals, aggressive
product rollouts and

177
00:09:48,280 --> 00:09:50,880
partnerships that were really
meant to scale quickly, like

178
00:09:50,880 --> 00:09:54,600
they were trying to develop
these iterations of Chachi PT as

179
00:09:54,600 --> 00:09:58,360
fast as possible.
Now here's really what's

180
00:09:58,360 --> 00:10:02,200
striking and his fair use essay.
He publishes an essay which

181
00:10:02,200 --> 00:10:03,320
we'll talk a little bit more
about.

182
00:10:03,760 --> 00:10:08,560
He does not rant or rage about
open AI ChatGPT.

183
00:10:08,560 --> 00:10:12,800
He actually methodically points
out that the commercial nature

184
00:10:12,800 --> 00:10:18,160
of these products weighs against
a fair use defense under the US

185
00:10:18,160 --> 00:10:20,560
law.
So in other words, the more open

186
00:10:20,600 --> 00:10:26,800
AI moves towards profit, the
more legal and ethical friction

187
00:10:26,840 --> 00:10:29,600
he saw.
Like you can't do both at the

188
00:10:29,600 --> 00:10:32,960
same time.
You can't move towards like

189
00:10:33,120 --> 00:10:36,120
commercialization and then also
be ethical.

190
00:10:36,440 --> 00:10:40,560
Like there is no way we're not
gathering this data in ways that

191
00:10:40,560 --> 00:10:42,840
are responsible.
No, because you have you're

192
00:10:42,840 --> 00:10:45,640
doing things for profit, yes.
Yes, yeah.

193
00:10:45,640 --> 00:10:48,120
And what you are trying to sell
a product?

194
00:10:48,240 --> 00:10:50,520
Exactly.
By any means necessary.

195
00:10:50,720 --> 00:10:53,040
Right.
Which is a problem, right?

196
00:10:53,040 --> 00:10:56,000
So it's what, you know,
essentially, it's one thing to

197
00:10:56,160 --> 00:10:58,840
build tools meant to benefit
humanity.

198
00:10:58,840 --> 00:11:02,600
It's a totally different thing
to watch the organization

199
00:11:02,600 --> 00:11:06,600
reorient around capital,
partnerships, valuation, speed,

200
00:11:06,800 --> 00:11:09,480
you know, money.
He joined this.

201
00:11:09,480 --> 00:11:11,720
He's this guy could have gotten
a job fucking anywhere.

202
00:11:11,800 --> 00:11:15,480
He could have gotten a job
anywhere and he chose to work at

203
00:11:15,480 --> 00:11:19,360
open AI because of their ethical
stances and the fact that this

204
00:11:19,360 --> 00:11:23,040
was a non profit.
So he they they bring him in,

205
00:11:23,560 --> 00:11:26,960
use his like skill set.
To build a life altering.

206
00:11:27,560 --> 00:11:30,960
Product or to build in a life
altering product that he was

207
00:11:31,160 --> 00:11:37,520
pivotal in I mean the Co the Co
founder said that yeah and Oh

208
00:11:37,520 --> 00:11:38,920
yeah, we're totally going to be
doing it good.

209
00:11:38,920 --> 00:11:42,400
It's just for like humanity's
benefit and then.

210
00:11:43,920 --> 00:11:46,280
What?
Happened then they're like, we

211
00:11:46,280 --> 00:11:47,800
need to make money.
Money.

212
00:11:49,360 --> 00:11:53,520
So by October 2024, I had talked
about his essay.

213
00:11:53,880 --> 00:11:57,080
Balji had published something
unusual for an engineer at that

214
00:11:57,080 --> 00:11:59,640
time still working in the
industry to do.

215
00:12:00,560 --> 00:12:05,480
He publishes that highly
technical, deeply critical essay

216
00:12:05,480 --> 00:12:10,360
titled When does Generative AI
Qualify for Fair Use?

217
00:12:10,360 --> 00:12:12,360
And he did not mince words in
this article.

218
00:12:12,360 --> 00:12:17,880
OK, the essay goes, actually
argues that the way generative

219
00:12:18,120 --> 00:12:21,760
AI systems train on Internet
data, including copyrighted

220
00:12:21,760 --> 00:12:26,520
work, may not actually pass the
fair use, the US fair use test.

221
00:12:26,920 --> 00:12:31,600
So he broke down each one of the
four legal factors from US

222
00:12:31,600 --> 00:12:36,240
Copyright Act one O 7, which is
the purpose of character,

223
00:12:36,880 --> 00:12:39,960
purpose and character of the
use, nature of the copyrighted

224
00:12:39,960 --> 00:12:43,360
work, amount used and the effect
on the market.

225
00:12:43,680 --> 00:12:48,760
Now his claim was that AI models
might fill several of those

226
00:12:48,760 --> 00:12:53,080
facts, especially the last one,
because as he put it on, the

227
00:12:53,360 --> 00:12:57,400
outputs of these models compete
with the original creators.

228
00:12:57,480 --> 00:13:01,680
He actually even went further
and warned that companies like

229
00:13:01,680 --> 00:13:05,880
Open AI were creating something
that extracts from the Internet

230
00:13:05,880 --> 00:13:09,440
without giving anything back,
and that this could damage the

231
00:13:09,440 --> 00:13:13,200
creative ecosystem long term.
Yeah, so I'm processing all of

232
00:13:13,200 --> 00:13:15,920
this in my brain, which is why
I'm I'm quiet.

233
00:13:19,080 --> 00:13:24,120
It's, it's this rock and a hard
place thing in my mind because

234
00:13:24,120 --> 00:13:27,400
like any human being, you're
going to know things that are

235
00:13:27,400 --> 00:13:30,840
copyrighted to help educate you
to do something different, which

236
00:13:30,840 --> 00:13:35,440
is no different than a large
language model, but to the point

237
00:13:35,440 --> 00:13:38,600
of the quantity, I guess, and
the quality.

238
00:13:38,600 --> 00:13:44,480
And that is where humans would
do it differently than AI would.

239
00:13:44,720 --> 00:13:48,360
Or it is.
AI, you know, doesn't have any

240
00:13:48,360 --> 00:13:50,600
scruples about the bullshit out
there.

241
00:13:50,600 --> 00:13:54,720
Like, which I would say, and
actually that's not necessarily

242
00:13:54,720 --> 00:13:56,920
true because you and I, before
the podcast even started, talked

243
00:13:56,920 --> 00:14:00,120
about like people who are just
there for power and like and

244
00:14:00,120 --> 00:14:02,160
stuff and don't care if they're
lying.

245
00:14:02,200 --> 00:14:05,760
Like they, they clearly will lie
just to incite, you know, folks

246
00:14:05,760 --> 00:14:09,360
and make them angry or whatever.
Same kind of situation, right?

247
00:14:09,440 --> 00:14:14,160
AI that's, that's really the,
the few or there are less people

248
00:14:14,160 --> 00:14:17,920
like humans who do that.
The people who are creating like

249
00:14:17,920 --> 00:14:19,200
works of art and things like
that.

250
00:14:19,680 --> 00:14:21,080
They're not doing that to flood
the market.

251
00:14:21,080 --> 00:14:23,680
They're just doing it because
it's something that, you know,

252
00:14:23,880 --> 00:14:27,240
they love to do it and they're
inspired and maybe they're

253
00:14:27,240 --> 00:14:29,600
hoping to inspire other people
in some capacity.

254
00:14:30,720 --> 00:14:34,320
But this essay that he
publishes, it's not just

255
00:14:34,320 --> 00:14:36,520
venting.
He's not just venting about

256
00:14:36,520 --> 00:14:38,360
shit.
He's not like these people are

257
00:14:38,360 --> 00:14:40,920
so mean and they don't like me.
Like, I hate this company.

258
00:14:42,040 --> 00:14:46,040
It was a shot fired across the
bow of one of the most powerful

259
00:14:46,040 --> 00:14:49,400
AI companies in the world.
I would say still right.

260
00:14:49,440 --> 00:14:52,680
Yeah, you're calling.
Yeah, you're, yeah, exactly.

261
00:14:52,760 --> 00:14:55,800
You're calling out kind of their
bullshit like you're bringing to

262
00:14:55,800 --> 00:15:00,160
light what really the hypocrisy,
The hypocrisy, that's the word I

263
00:15:00,160 --> 00:15:03,000
was looking for.
Yes, yes, yeah.

264
00:15:03,400 --> 00:15:07,920
And to a what I'm sure is a
billion dollar market, that's

265
00:15:07,920 --> 00:15:12,480
not a good idea.
No, no, it's not a good idea.

266
00:15:12,480 --> 00:15:14,480
They have power and money,
right?

267
00:15:14,480 --> 00:15:17,880
And investors.
And actually this came at a

268
00:15:17,880 --> 00:15:22,040
moment when the lawsuits about
open AI were mounting.

269
00:15:22,240 --> 00:15:23,680
There were more.
And he knew this.

270
00:15:23,680 --> 00:15:27,440
He knew that the lawsuits were
were being stacked and from the

271
00:15:27,440 --> 00:15:29,040
inside.
What does he do?

272
00:15:29,200 --> 00:15:33,600
He publishes an article on his
page that gets picked up by the

273
00:15:33,600 --> 00:15:38,560
New York Times and is run
against open AI.

274
00:15:38,920 --> 00:15:43,680
OK, The New York Times had
republished like the article on

275
00:15:43,680 --> 00:15:46,680
his website and on it they're
like really saying engineer

276
00:15:46,760 --> 00:15:49,400
accusing open AI for copyright
infringement.

277
00:15:49,400 --> 00:15:51,040
And like, of course, it's like a
top story.

278
00:15:51,040 --> 00:15:53,640
Top one.
If you want to understand why an

279
00:15:53,640 --> 00:15:57,880
insider might get uneasy, you
have to kind of look at the

280
00:15:57,880 --> 00:16:03,640
company's relationship with its
training data, right?

281
00:16:03,640 --> 00:16:09,280
How is Open AI training ChatGPT
on data?

282
00:16:09,280 --> 00:16:12,960
A Ball G's essay includes a line
that may make you feel a little

283
00:16:13,000 --> 00:16:15,400
uneasy.
He writes that companies like

284
00:16:15,400 --> 00:16:19,680
Open AI have signed numerous
data licensing agreements.

285
00:16:20,360 --> 00:16:23,120
And then he asks the question
that most engineers might keep

286
00:16:23,120 --> 00:16:26,640
quiet about.
If this is all for fair use, why

287
00:16:26,640 --> 00:16:32,280
pay for any licenses at all?
Right.

288
00:16:32,560 --> 00:16:34,360
Why are you paying for this?
Why are you?

289
00:16:34,360 --> 00:16:37,600
Right, exactly.
Ultimately, it it implies 2

290
00:16:37,600 --> 00:16:40,080
things.
One, the legal status of the

291
00:16:40,080 --> 00:16:43,040
training data actually might not
be as clear as the company

292
00:16:43,040 --> 00:16:46,680
claims that it is.
And then two some models may

293
00:16:46,680 --> 00:16:51,840
rely on licensed sets, while
others quietly use unlicensed

294
00:16:51,840 --> 00:17:00,040
ones.
Right, because who's to say

295
00:17:00,040 --> 00:17:01,480
that?
I mean, it's on the Internet,

296
00:17:02,320 --> 00:17:07,160
whether it's licensed or not.
You may say, yeah, I mean, who's

297
00:17:07,160 --> 00:17:11,040
to say that you couldn't just
say, open the floodgates and go,

298
00:17:11,040 --> 00:17:12,480
just go wherever?
Yeah.

299
00:17:12,480 --> 00:17:14,880
No, totally.
Yeah, no, we're doing things.

300
00:17:15,079 --> 00:17:16,160
We're we're doing things up and.
Up.

301
00:17:16,160 --> 00:17:19,319
Yeah, yeah, it's totally fine.
Yeah, No, like outside of the

302
00:17:19,319 --> 00:17:22,920
company, you know,
transparency's an actual like,

303
00:17:22,920 --> 00:17:25,240
word that people use and
actuate.

304
00:17:25,240 --> 00:17:29,680
But inside, there's a level of
visibility that the employees

305
00:17:29,680 --> 00:17:33,240
may or may not have within.
And they're actively training

306
00:17:33,240 --> 00:17:37,080
this this huge program.
We also know that he actually

307
00:17:37,080 --> 00:17:41,760
helped collect and organize
large Internet skilled data

308
00:17:41,760 --> 00:17:45,080
sets.
So he saw the pipeline up close.

309
00:17:45,080 --> 00:17:47,120
He saw what was being fed to
these.

310
00:17:47,520 --> 00:17:50,560
He helped arrange them.
So imagine this, you are

311
00:17:50,560 --> 00:17:53,600
actually handling the data.
You've watched the lawsuit stack

312
00:17:53,600 --> 00:17:59,760
up, you've seen this mix of
licensed and unlicensed sources,

313
00:17:59,760 --> 00:18:04,640
and then you watch the company
roll out product after product

314
00:18:05,360 --> 00:18:07,440
all saying, Oh yeah, no, no, we
it's all up and up.

315
00:18:07,440 --> 00:18:09,360
It's all up and up.
We follow copyright law.

316
00:18:09,360 --> 00:18:13,080
It's totally fine.
And someone who has like this

317
00:18:13,080 --> 00:18:17,040
really strong, like moral
compass, it's going to create a

318
00:18:17,040 --> 00:18:19,800
huge tension like he's going to,
he's going to struggle with

319
00:18:19,800 --> 00:18:21,000
that.
He's going to struggle hard.

320
00:18:21,040 --> 00:18:22,960
And something has to give,
right?

321
00:18:22,960 --> 00:18:25,480
There's something that really
comes from tech engineers.

322
00:18:25,480 --> 00:18:28,600
He actually starts having a
concern for the ecosystem that

323
00:18:28,600 --> 00:18:32,360
he is working within.
And he isn't just worried about

324
00:18:32,360 --> 00:18:35,000
lawsuits.
He was really worried about the

325
00:18:35,000 --> 00:18:37,720
Internet, the health of the
Internet itself.

326
00:18:37,720 --> 00:18:40,840
Makes sense.
Starts getting flooded with what

327
00:18:40,840 --> 00:18:42,920
we know as hallucinations, as an
example.

328
00:18:42,920 --> 00:18:45,800
Exactly, and he's probably
seeing this on a much larger

329
00:18:45,800 --> 00:18:50,960
scale before Chachi PT is even
talking about before Open AI is

330
00:18:51,040 --> 00:18:55,480
fessing up to it.
In his fair use essayist, he

331
00:18:55,640 --> 00:19:01,160
recites research showing that
there was a 12% drop in Stack

332
00:19:01,160 --> 00:19:06,280
Overflow traffic after releases
of powerful language models.

333
00:19:07,680 --> 00:19:12,400
So what is Stack Overflow?
That's the data that would be

334
00:19:12,400 --> 00:19:13,720
stack overflow.
It's the.

335
00:19:13,760 --> 00:19:18,480
So a stack is a set of data.
Overflow would be what what?

336
00:19:19,040 --> 00:19:20,400
What was it?
The output?

337
00:19:20,720 --> 00:19:22,000
What would this?
What would that be?

338
00:19:22,000 --> 00:19:25,600
Well, it's, it's basically what
he's he's doing is able to track

339
00:19:25,600 --> 00:19:29,320
the information that's being
like generated out of it.

340
00:19:29,320 --> 00:19:31,480
Yeah, out of it and within the
Internet, right.

341
00:19:31,880 --> 00:19:34,560
So he's seeing that there's
actually a drop.

342
00:19:34,600 --> 00:19:37,840
Once they have these models that
are uploaded into the system

343
00:19:37,840 --> 00:19:40,920
that are that are updated, he
sees that there's a drop in

344
00:19:40,920 --> 00:19:46,360
information being created, the
traffic that's actually being.

345
00:19:46,360 --> 00:19:48,040
Weird.
Yeah.

346
00:19:48,040 --> 00:19:50,400
So he's like, no, this is
actually not good.

347
00:19:50,840 --> 00:19:54,440
Right the.
The data the the Internet itself

348
00:19:54,800 --> 00:19:59,720
is almost starting to to die.
Like you're seeing a decrease in

349
00:19:59,720 --> 00:20:02,240
the information that is being
creatively.

350
00:20:02,320 --> 00:20:05,920
Because people are trying to get
it out of chat BT rather than

351
00:20:05,920 --> 00:20:08,880
it's like you said before,
contributing into the Internet.

352
00:20:08,880 --> 00:20:10,520
Yeah, people are.
Not there's nothing going back

353
00:20:10,560 --> 00:20:12,480
in yes.
Exactly.

354
00:20:12,560 --> 00:20:16,480
Yeah, it's kind of fucked up.
And he's now he's arguing that

355
00:20:16,480 --> 00:20:20,280
generative AI doesn't just use
content, it can replace the need

356
00:20:20,280 --> 00:20:24,560
to visit the original source.
So the traffic like the people

357
00:20:24,560 --> 00:20:30,040
like clicking on, yes,
Wikipedia, whatever, like

358
00:20:30,040 --> 00:20:33,400
dictionary or like, or going to
like these blogs and things like

359
00:20:33,400 --> 00:20:35,840
that.
They're, they stop going to it.

360
00:20:35,840 --> 00:20:38,600
They're just asking ChatGPT to
like.

361
00:20:38,600 --> 00:20:41,480
How often have you asked ChatGPT
for something and not even

362
00:20:41,480 --> 00:20:42,960
clicked on the link that it
provides?

363
00:20:43,120 --> 00:20:46,680
Right, right.
And going outside to that

364
00:20:46,680 --> 00:20:49,120
traffic.
So like exactly you're not doing

365
00:20:49,120 --> 00:20:50,160
it, you're.
Not doing it, no.

366
00:20:50,480 --> 00:20:54,440
So if you think about it like AI
is pulling from artists, from

367
00:20:54,440 --> 00:20:59,080
writers, from researchers,
coders, musicians, forum

368
00:20:59,080 --> 00:21:03,040
posters, Reddit, you know, sure,
everyday contributors, yes.

369
00:21:03,040 --> 00:21:06,840
And it then reduces the traffic
back to those communities.

370
00:21:06,840 --> 00:21:10,120
So it's literally draining the
Internet from the people that

371
00:21:10,120 --> 00:21:14,560
create real value to it.
It would almost be like, and

372
00:21:14,800 --> 00:21:18,560
this is an extreme example to
help us, I think visualize this.

373
00:21:18,920 --> 00:21:21,800
It'd be like if you didn't have
ChatGPT today, like you said,

374
00:21:21,800 --> 00:21:23,640
you would go out and you'd find
these sources.

375
00:21:23,640 --> 00:21:27,200
You would Google it and then go
to your places that you know

376
00:21:27,200 --> 00:21:31,960
were giving you what you want.
Imagine everyone stopped doing

377
00:21:31,960 --> 00:21:36,520
that and just ask chat BT it.
It's like instead of having

378
00:21:36,520 --> 00:21:39,760
millions of people going to
these sources, you have one

379
00:21:39,760 --> 00:21:43,000
person going to these sources.
Yeah, ultimately.

380
00:21:43,080 --> 00:21:46,080
And that one person chat adding
to that source, right.

381
00:21:46,120 --> 00:21:48,200
He's not contributing.
Yeah, you're not contributing.

382
00:21:48,200 --> 00:21:54,040
You're not saying, you know, for
instance, you're you like a a

383
00:21:54,320 --> 00:21:56,000
bar or something or like a
restaurant.

384
00:21:56,400 --> 00:21:59,480
You're not.
You're no longer saying like it

385
00:21:59,480 --> 00:22:02,600
or four stars or whatever.
Rating it, right.

386
00:22:02,600 --> 00:22:04,520
You're not rating it anymore.
Excuse me?

387
00:22:04,520 --> 00:22:06,040
You're just going just reading
it.

388
00:22:06,080 --> 00:22:07,920
Yeah.
You're just, you're just reading

389
00:22:07,920 --> 00:22:08,360
it.
Yeah.

390
00:22:09,320 --> 00:22:10,760
It's wild.
OK.

391
00:22:10,880 --> 00:22:13,360
Yeah.
It's really cause and effect and

392
00:22:13,360 --> 00:22:16,240
he seems to recognize this
earlier than many 'cause he's.

393
00:22:16,560 --> 00:22:19,920
'Cause he's a genius.
Yeah, he's a genius that

394
00:22:19,960 --> 00:22:24,120
exploiting the Commons without
giving back harms the system.

395
00:22:24,400 --> 00:22:28,960
So for a young engineer, he was
raised on the idea that the

396
00:22:28,960 --> 00:22:31,400
Internet is for everyone, right?
Of course.

397
00:22:32,240 --> 00:22:38,640
And watching it be strip mined
for information gold that gain

398
00:22:38,640 --> 00:22:42,560
for profit sake would definitely
be painful.

399
00:22:42,560 --> 00:22:44,880
Because in the beginning I was
saying like, he's very naive, he

400
00:22:44,880 --> 00:22:47,480
was young, blah, blah, blah.
I honestly, I think that he was

401
00:22:47,480 --> 00:22:51,040
just seeing something happen in
real time and really started

402
00:22:51,400 --> 00:22:53,960
causing an alarm.
Like this is actually a problem

403
00:22:54,000 --> 00:22:56,560
and we're not thinking about
this being a problem.

404
00:22:56,560 --> 00:22:59,920
So around the time that Balji
left, Opening Eye was going

405
00:22:59,920 --> 00:23:04,040
through internal upheaval.
There was board conflicts,

406
00:23:04,040 --> 00:23:08,000
leadership disputes, there were
debates about safety protocols.

407
00:23:08,400 --> 00:23:12,080
And it was actually the same
period of time when concerns

408
00:23:12,080 --> 00:23:15,960
about hallucinations,
reliability and alignment were

409
00:23:15,960 --> 00:23:20,040
becoming more visible.
They were not talking about

410
00:23:20,040 --> 00:23:21,000
this.
They weren't talking about the

411
00:23:21,000 --> 00:23:22,720
fact that, like, ChatGPT will
lie to you.

412
00:23:22,880 --> 00:23:27,200
ChatGPT has, quite honestly,
people have have.

413
00:23:27,840 --> 00:23:30,360
It makes shit up.
It makes shit up and and it has

414
00:23:30,360 --> 00:23:34,000
triggered psychosis in people
because it will hallucinate.

415
00:23:34,000 --> 00:23:35,600
It will tell you in things that
are tell.

416
00:23:35,600 --> 00:23:38,240
You something that are not real.
Yeah, and then you'll start to

417
00:23:38,240 --> 00:23:40,960
believe it.
And so people like it starts

418
00:23:40,960 --> 00:23:42,880
hallucinating.
Hallucinating.

419
00:23:42,880 --> 00:23:45,560
You know, when he was speaking
publicly, he actually mentioned

420
00:23:45,560 --> 00:23:48,280
issues beyond the copyright.
Not, not just copyright.

421
00:23:48,280 --> 00:23:52,280
He said that the pace of
deployment is a problem they're

422
00:23:52,280 --> 00:23:55,880
just trying to get shit out of
as quickly as possible.

423
00:23:56,200 --> 00:23:58,040
You can't.
What is it like you can't, you

424
00:23:58,040 --> 00:23:59,360
have, you have three things to
choose.

425
00:23:59,360 --> 00:24:02,840
It's like a time, energy and
money, right?

426
00:24:03,120 --> 00:24:06,600
So if you shortchange one, the
other two are going to suffer,

427
00:24:06,600 --> 00:24:08,400
right?
Like so you don't have to.

428
00:24:08,720 --> 00:24:13,040
So the content of the models
that they're pushing forward are

429
00:24:13,040 --> 00:24:15,160
going to start having real big
issues.

430
00:24:16,040 --> 00:24:17,480
Because everything in the
Internet is true.

431
00:24:17,560 --> 00:24:20,720
Yeah, that's.
Exactly, exactly, exactly.

432
00:24:20,720 --> 00:24:25,080
We recently released the last
episode that that I had a hosted

433
00:24:25,080 --> 00:24:29,600
on on our friend from time like
history, The Time Traveller.

434
00:24:29,760 --> 00:24:31,000
Right, exactly.
It's gonna.

435
00:24:31,160 --> 00:24:32,240
That's it.
No, it's gonna look.

436
00:24:32,240 --> 00:24:35,560
At that and think it's real
exactly so you know he's saying

437
00:24:35,560 --> 00:24:38,560
that this pace of deployment's a
problem he's saying that the

438
00:24:38,560 --> 00:24:42,800
pressure to release products
quickly is a problem and that

439
00:24:42,800 --> 00:24:47,920
the risk of models generating
false or harmful content is a

440
00:24:47,920 --> 00:24:54,320
huge problem It is and this was
over a year ago and people were

441
00:24:54,320 --> 00:24:56,920
talking about the fact that or
he was talking about the fact

442
00:24:56,920 --> 00:24:59,480
that like this is definitely
happening and I know how much.

443
00:24:59,600 --> 00:25:00,960
You know how much has changed in
the year.

444
00:25:01,400 --> 00:25:03,640
It's changed so much.
Drastic, yeah.

445
00:25:04,240 --> 00:25:09,160
With technology specifically, I
mean, it's moving so fast, so

446
00:25:09,160 --> 00:25:11,080
you can only imagine.
That was a year ago.

447
00:25:11,080 --> 00:25:13,040
You can only imagine.
Oh, yeah, yeah.

448
00:25:13,040 --> 00:25:15,600
And he was seeing all of the
data and the information that

449
00:25:15,600 --> 00:25:17,440
was being put into it.
And now he's being trained.

450
00:25:17,640 --> 00:25:21,040
Really, engineers like him know
that model failures are not

451
00:25:21,200 --> 00:25:24,360
they're not abstract.
They're really baked into the

452
00:25:24,360 --> 00:25:28,200
training choices, the data
quality, the evaluation

453
00:25:28,200 --> 00:25:31,320
shortcuts that they're making.
And let's honestly be real,

454
00:25:32,000 --> 00:25:35,080
these fast moving companies we
were talking about, they're

455
00:25:35,080 --> 00:25:38,440
going to sacrifice something
just to move forward quickly.

456
00:25:38,440 --> 00:25:42,080
Oh, I don't know if it was
pressure, internal culture or

457
00:25:42,080 --> 00:25:45,480
the sense of the company's like
move fast, scale, big classic

458
00:25:45,480 --> 00:25:49,920
fucking, you know, computer or
like a startup mentality.

459
00:25:50,960 --> 00:25:53,400
But that mentality was outpacing
its safety work.

460
00:25:53,400 --> 00:25:58,320
And from his public comments, we
can infer that he believed that

461
00:25:58,320 --> 00:26:03,040
there would be harm not just
from copyright issues, but from

462
00:26:03,040 --> 00:26:07,000
the very speed and scope Open AI
was starting to release.

463
00:26:07,680 --> 00:26:09,560
And that can be intoxicating,
right?

464
00:26:09,560 --> 00:26:13,480
Like by the people who are in
charge, but when you're the

465
00:26:13,480 --> 00:26:17,160
engineer who is seeing all the
fucking issues and problems and

466
00:26:17,160 --> 00:26:18,720
errors.
They feel responsible.

467
00:26:18,720 --> 00:26:21,640
Yes, you feel responsible and no
one's listening to you.

468
00:26:21,640 --> 00:26:27,760
Only weeks after his essay.
It was released in October and

469
00:26:27,760 --> 00:26:35,880
on November 18th, 2024, his name
was appearing in legal findings

470
00:26:35,880 --> 00:26:38,400
and he was actually listed as a
potential witness.

471
00:26:39,120 --> 00:26:41,600
Sure.
Someone who might have unique

472
00:26:41,600 --> 00:26:43,840
and relative documents.
That's what it was.

473
00:26:44,520 --> 00:26:45,600
Yeah.
Yeah.

474
00:26:46,760 --> 00:26:48,200
In these legal.
Retro.

475
00:26:49,000 --> 00:26:52,440
Retro's right.
So he may have had unique and

476
00:26:52,440 --> 00:26:57,080
relevant documents for several
copyright related lawsuits

477
00:26:57,120 --> 00:27:00,480
against open AI.
His parents were had later told

478
00:27:00,480 --> 00:27:04,320
the Guardian that he said he
wanted to testify in the

479
00:27:04,320 --> 00:27:06,520
strongest fair use cases.
So he's like, yeah, I'll

480
00:27:06,520 --> 00:27:08,880
definitely testify.
If I mean do it like let's do

481
00:27:08,880 --> 00:27:10,400
it.
Yeah, let's do the ones that

482
00:27:10,400 --> 00:27:11,520
like.
Yeah, matter.

483
00:27:11,640 --> 00:27:13,720
This is really where his
tensions start showing.

484
00:27:13,760 --> 00:27:17,200
You have former employee, a
technical insider and a person

485
00:27:17,200 --> 00:27:20,440
who is directly handling the
training data.

486
00:27:20,440 --> 00:27:23,560
He's publicly challenging the
legal foundation of the

487
00:27:23,560 --> 00:27:27,400
company's business model and
offering to cooperate in

488
00:27:27,400 --> 00:27:29,760
lawsuits.
And that's not it to.

489
00:27:29,760 --> 00:27:32,120
The chagrin.
Yes, to the chagrin.

490
00:27:32,280 --> 00:27:34,840
The power people.
Money, people.

491
00:27:35,000 --> 00:27:37,080
Yeah, yeah.
It's not a small thing, right?

492
00:27:37,440 --> 00:27:40,000
It's.
Textbook definition of a

493
00:27:40,000 --> 00:27:42,080
whistleblower, right?
He never used that word, by the

494
00:27:42,080 --> 00:27:44,520
way, but just enough.
He never said that, but it is

495
00:27:44,600 --> 00:27:48,560
that's exactly what he is.
So around that time, you know,

496
00:27:48,640 --> 00:27:51,040
open AI, like all these lawsuits
are coming up, right?

497
00:27:51,080 --> 00:27:54,880
Open AI is facing increasing
scrutiny over what the clarity

498
00:27:54,880 --> 00:27:59,040
around their training data,
concerns about their safety and

499
00:27:59,040 --> 00:28:00,000
their alignment.
Who are they?

500
00:28:00,160 --> 00:28:02,120
Like, what are what are they
trying to do?

501
00:28:02,120 --> 00:28:05,520
Are you being paid by other
companies to put their search?

502
00:28:05,520 --> 00:28:09,240
Like what is happening here?
There are leadership disputes

503
00:28:09,520 --> 00:28:13,280
and there's massive legal
battles with writers, publishers

504
00:28:13,280 --> 00:28:16,160
and other creators.
And Baoji's critique was that

505
00:28:16,360 --> 00:28:19,880
training data may not qualify as
fair use.

506
00:28:20,040 --> 00:28:23,160
It really struck at the core of
their technology.

507
00:28:23,400 --> 00:28:26,680
So if the courts were to rule
that training is not fair, use

508
00:28:26,680 --> 00:28:31,440
generative AIS as we know.
Would be go away, yes.

509
00:28:31,560 --> 00:28:34,320
Yeah, they would have to shut it
down or rebuild it or something.

510
00:28:34,960 --> 00:28:37,680
Exactly.
That's not going to happen.

511
00:28:37,920 --> 00:28:41,360
No, no, not in their eyes.
So when an engineer steps

512
00:28:41,360 --> 00:28:44,080
forward and is like, I don't
think this is legal.

513
00:28:45,160 --> 00:28:46,840
Uh huh.
That's a threat.

514
00:28:47,200 --> 00:28:48,840
Yeah.
It's a threat to opening.

515
00:28:48,880 --> 00:28:53,320
Up sure is.
I was talking about how his

516
00:28:53,640 --> 00:28:56,360
article was released on the New
York Times.

517
00:28:56,360 --> 00:28:59,920
He's getting involved in all of
these like legal battle, and

518
00:28:59,920 --> 00:29:02,080
he's getting called as a witness
in the strong ones.

519
00:29:02,080 --> 00:29:04,000
He's speaking out now.
We're talking about November

520
00:29:04,000 --> 00:29:06,040
18th.
This was just a year ago.

521
00:29:06,680 --> 00:29:10,840
This is just a year ago only,
you know, Weeks after his essay

522
00:29:10,840 --> 00:29:14,080
had come out, November 18th,
he's listed in legal documents

523
00:29:14,080 --> 00:29:17,200
as being a person of interest,
to be a witness.

524
00:29:17,200 --> 00:29:19,480
He's like, yeah, I'm happy to do
it.

525
00:29:19,480 --> 00:29:21,680
I'm having these issues.
He's having he's, you know,

526
00:29:21,680 --> 00:29:24,360
meeting with, like, reporters,
things like that, like legal

527
00:29:24,520 --> 00:29:29,800
folks.
On November 26th, 2024, Balji

528
00:29:29,800 --> 00:29:36,200
was found dead in his San
Francisco apartment and the San

529
00:29:36,200 --> 00:29:39,360
Francisco Medical Examiner ruled
his death a suicide.

530
00:29:39,520 --> 00:29:44,240
Oh my God, here we go again.
Right from the report, I'll go

531
00:29:44,240 --> 00:29:47,720
through the report and we'll
talk about the different sides

532
00:29:47,720 --> 00:29:51,160
of it, OK, It's let's just say
it's interesting timing, right?

533
00:29:51,320 --> 00:29:52,960
To say the least.
To say the least.

534
00:29:53,240 --> 00:29:56,760
So the report actually noted
that his apartment door was dead

535
00:29:56,760 --> 00:30:02,880
bolted from the inside, he had
legally purchased a firearm, and

536
00:30:03,000 --> 00:30:07,360
on his toxicology report it
showed that he had consumed GHB

537
00:30:07,360 --> 00:30:08,760
and alcohol.
OK.

538
00:30:10,120 --> 00:30:12,960
OK, yes, he's scared.
Yeah, sure.

539
00:30:13,560 --> 00:30:16,680
But his parents are strongly
disputing this filing.

540
00:30:16,680 --> 00:30:17,920
He was very close with his
parents.

541
00:30:17,920 --> 00:30:21,120
They actually argued that the
evidence was mishandled and that

542
00:30:21,120 --> 00:30:23,440
footage from the apartment
building is missing.

543
00:30:24,160 --> 00:30:26,240
OK.
Really.

544
00:30:26,320 --> 00:30:27,600
Yeah.
Exactly.

545
00:30:27,800 --> 00:30:29,520
How many times?
How many times?

546
00:30:29,600 --> 00:30:30,840
How many times does have to
happen?

547
00:30:30,880 --> 00:30:35,400
So in September 2025, they
actually filed A lawsuit against

548
00:30:35,400 --> 00:30:39,040
the building's management,
alleging obstruction and

549
00:30:39,040 --> 00:30:40,720
tampering evidence.
Clear.

550
00:30:40,760 --> 00:30:43,640
Clearly.
So here's the key thing, like no

551
00:30:43,640 --> 00:30:47,160
matter which interpretation when
you like, you see in his death,

552
00:30:47,280 --> 00:30:49,960
it does not change the substance
of what he said while he was

553
00:30:49,960 --> 00:30:51,000
alive, right?
Right.

554
00:30:51,960 --> 00:30:55,080
So he didn't publish leaks, you
know, his whistleblowing.

555
00:30:55,080 --> 00:30:59,000
No, he didn't publish leaks.
He didn't expose internal chat

556
00:30:59,000 --> 00:31:01,480
logs or confidential documents.
Exactly.

557
00:31:02,040 --> 00:31:04,280
But what he did do was a little
bit more dangerous.

558
00:31:04,280 --> 00:31:09,120
He applied the law calmly,
technically, publicly, to the

559
00:31:09,120 --> 00:31:12,160
place that he used to work.
Which makes it more difficult to

560
00:31:12,160 --> 00:31:13,640
dispute.
Right.

561
00:31:14,000 --> 00:31:21,520
Yeah, he's, he is asking if the
foundation of generative AI, is

562
00:31:21,520 --> 00:31:26,040
it built on something that might
not hold up in court, The

563
00:31:26,040 --> 00:31:28,680
fucking foundation of it.
He's not saying these people are

564
00:31:28,680 --> 00:31:31,320
mean.
He's like, no, everything that

565
00:31:31,320 --> 00:31:33,680
this stands for, this is not
legal.

566
00:31:33,760 --> 00:31:36,720
And so if they were found that
they literally would have to

567
00:31:36,720 --> 00:31:37,960
flip the switch.
Right.

568
00:31:38,520 --> 00:31:41,240
Exactly, exactly.
They'd have to flip the switch,

569
00:31:41,560 --> 00:31:44,240
then everything like then this
money making scheme that they

570
00:31:44,240 --> 00:31:45,880
have, everything's up in
question.

571
00:31:45,880 --> 00:31:48,120
They're not going to be able to
make the money and like to have

572
00:31:48,120 --> 00:31:49,800
the power that they currently
have, right?

573
00:31:49,880 --> 00:31:52,360
This essay did not disappear
after his death.

574
00:31:52,520 --> 00:31:57,440
It really became more relevant
as all of these additional

575
00:31:57,440 --> 00:32:01,080
lawsuits start popping up and it
really raises the questions that

576
00:32:01,080 --> 00:32:04,320
everyone in tech, law and
creative fields should be asking

577
00:32:04,360 --> 00:32:08,440
about any kind of AI artificial
intelligence.

578
00:32:08,800 --> 00:32:12,480
Can companies scrape the entire
Internet to build a billion

579
00:32:12,480 --> 00:32:16,400
dollar like AI model?
Do copyright laws apply to

580
00:32:16,400 --> 00:32:19,280
machines that learn from human
work?

581
00:32:19,280 --> 00:32:22,800
And if AI can output something
that competes with the original

582
00:32:22,800 --> 00:32:25,640
creator that is not that is fair
use.

583
00:32:25,920 --> 00:32:32,480
Is that stealing?
Well, I, I, I know I, I mean, I

584
00:32:32,480 --> 00:32:35,800
have my opinions, but I mean, it
can be argued either way.

585
00:32:35,800 --> 00:32:38,200
I guess I could say you could
argue this case either way.

586
00:32:38,600 --> 00:32:40,840
Yeah, and I don't think we
necessarily have the answers

587
00:32:40,840 --> 00:32:43,400
yet, but he really did force
this conversation.

588
00:32:43,400 --> 00:32:46,640
Yes, exactly.
He's like, is this ethical?

589
00:32:46,680 --> 00:32:49,000
I right.
You know, I'm having fucking

590
00:32:49,000 --> 00:32:51,880
questions and, like, doubts, and
no one's listening to me, so I'm

591
00:32:51,880 --> 00:32:54,400
just going to publish it.
And the only way I know how

592
00:32:54,760 --> 00:32:57,480
calmly and methodically.
Yeah. data-driven, dude.

593
00:32:57,680 --> 00:32:59,320
Yeah.
So when the San Francisco

594
00:32:59,320 --> 00:33:03,280
medical Examiner ruled his death
a suicide, the public, of

595
00:33:03,280 --> 00:33:07,800
course, makes sense, you know.
Because he was so distraught

596
00:33:07,880 --> 00:33:10,240
over the whole thing.
I mean, why would he do it?

597
00:33:10,360 --> 00:33:11,320
Yeah, and why?
Did it?

598
00:33:11,320 --> 00:33:15,520
Why did that make sense?
Yeah, I, I mean, nobody, nobody

599
00:33:15,520 --> 00:33:17,560
really.
This was like weeks after he had

600
00:33:17,560 --> 00:33:20,400
released this.
And, you know, it's just easier

601
00:33:20,400 --> 00:33:24,600
to kind of shove it underneath
the rug rather than saying, oh,

602
00:33:24,600 --> 00:33:28,360
hey, this app that I use all the
time, maybe the people who are

603
00:33:28,360 --> 00:33:32,240
running it are evil and are
trying to make people disappear,

604
00:33:33,440 --> 00:33:35,520
right?
His parents did not accept this

605
00:33:35,520 --> 00:33:36,480
well.
Of course not.

606
00:33:36,600 --> 00:33:38,880
Yeah.
And their doubts were not rooted

607
00:33:38,880 --> 00:33:42,200
in denial or conspiracy.
They were rooted in specific

608
00:33:42,200 --> 00:33:46,560
inconsistencies that would
trouble any reasonable person,

609
00:33:46,560 --> 00:33:50,280
especially when it involves the
death of a young person and a

610
00:33:50,280 --> 00:33:53,920
young whistleblower who was
standing at the edge of a legal

611
00:33:53,920 --> 00:33:55,640
firestorm.
It makes.

612
00:33:55,760 --> 00:33:57,440
Yeah, It doesn't make it.
There's no way.

613
00:33:57,520 --> 00:33:59,880
No.
Now I wanted to talk about why

614
00:33:59,880 --> 00:34:02,800
they didn't accept his, the
official story and why their

615
00:34:02,800 --> 00:34:06,880
questions still do matter.
OK, so this missing surveillance

616
00:34:06,880 --> 00:34:11,400
footage, he lived in a secure
building, a nice building with

617
00:34:11,400 --> 00:34:14,960
cameras.
And his parents were asked to

618
00:34:14,960 --> 00:34:18,159
review the footage.
They actually reportedly were

619
00:34:18,159 --> 00:34:21,719
told that it was missing,
inaccessible, or simply not

620
00:34:21,719 --> 00:34:24,760
available for the window around
his death.

621
00:34:27,639 --> 00:34:29,600
It's exhausting.
For a tragedy, you know, this is

622
00:34:29,600 --> 00:34:32,920
that this is significant, right?
This is not just a small glitch.

623
00:34:33,400 --> 00:34:36,080
You know, it's it's obvious.
Yeah.

624
00:34:36,360 --> 00:34:40,840
Yeah.
Surveillance isn't proof of a

625
00:34:40,840 --> 00:34:43,760
crime.
We're not saying that like, you

626
00:34:43,760 --> 00:34:46,760
know, right, proof of a crime,
but it is one of the biggest

627
00:34:46,760 --> 00:34:50,040
reasons that the family is
questioning the official

628
00:34:50,040 --> 00:34:51,880
rulings.
Why did this information?

629
00:34:51,880 --> 00:34:55,199
Why was this deleted?
Exactly that does.

630
00:34:55,239 --> 00:34:59,720
No, no, absolutely not.
It does not make any sense, no.

631
00:34:59,720 --> 00:35:02,880
The other issue that they have
is that the angle and the nature

632
00:35:02,880 --> 00:35:05,080
of the gun gunshot wound is
really in question.

633
00:35:05,400 --> 00:35:08,200
His parents say that the
trajectory and the placement of

634
00:35:08,200 --> 00:35:11,920
the wound didn't match what they
expected of a self-inflicted

635
00:35:12,040 --> 00:35:14,360
shot.
OK, they're not, you know,

636
00:35:14,360 --> 00:35:17,440
forensic experts, but they are
parents.

637
00:35:17,440 --> 00:35:21,440
And when the physical details
don't match the story that

638
00:35:21,440 --> 00:35:24,640
you're being told, alarm bells
go off.

639
00:35:24,640 --> 00:35:27,960
Of course, Right, Right.
Once again, not, you know, we

640
00:35:27,960 --> 00:35:31,760
don't have proof of foul play,
but this is a legitimate like

641
00:35:32,480 --> 00:35:35,080
question and something that you
should be doubting.

642
00:35:35,400 --> 00:35:37,600
Now, here's another thing.
The presence of GHB in his

643
00:35:37,600 --> 00:35:40,360
system.
So the autopsy noted that there

644
00:35:40,360 --> 00:35:44,400
was GHB which is tied to two
different, very different

645
00:35:44,400 --> 00:35:48,160
worlds, right?
Recreational nightlife and

646
00:35:49,040 --> 00:35:53,320
incapacitation, right?
Because GHB is in what like.

647
00:35:53,480 --> 00:35:56,000
Well, it's a substance that you
can actually take and like

648
00:35:56,000 --> 00:35:59,120
people will use it.
It's not something that a doctor

649
00:35:59,280 --> 00:36:00,760
could prescribe, but it's like
it's used.

650
00:36:00,760 --> 00:36:04,400
It's like a drug that you take
and like, it makes you more

651
00:36:04,400 --> 00:36:09,840
docile.
So his parents really questioned

652
00:36:09,840 --> 00:36:14,320
whether he would ever willingly
take it and whether its presence

653
00:36:14,320 --> 00:36:19,400
contributed to vulnerability.
So it really does complicate the

654
00:36:19,400 --> 00:36:21,920
simplicity of like a suicide
narrative.

655
00:36:21,920 --> 00:36:24,680
Well, that in and of itself
might not be.

656
00:36:24,680 --> 00:36:27,480
But when you add up, install the
little pieces, then you start

657
00:36:27,480 --> 00:36:29,560
adding up, then it starts to
become something.

658
00:36:29,640 --> 00:36:32,320
Yeah, yeah.
It really like introduces this

659
00:36:32,320 --> 00:36:35,480
ambiguity, like, OK, you don't
have any surveillance, OK.

660
00:36:35,600 --> 00:36:37,480
You know, he has GHB in his
system.

661
00:36:38,320 --> 00:36:40,200
OK, there's a weird gunshot
wound.

662
00:36:40,200 --> 00:36:42,360
Like, you know, the stuff just
starts to add up.

663
00:36:42,360 --> 00:36:46,120
The 4th 1 was like, you know, he
didn't have signs of despair.

664
00:36:46,400 --> 00:36:50,160
He didn't have a history of of
like suicidal thoughts.

665
00:36:50,640 --> 00:36:52,720
You know, his he was future
focused.

666
00:36:52,720 --> 00:36:56,120
He's coming out talking about
all these issues with an open

667
00:36:56,120 --> 00:36:58,400
AI.
He was mentally stable.

668
00:36:58,440 --> 00:37:01,880
He was professionally engaged.
He was like still releasing

669
00:37:01,880 --> 00:37:05,160
stuff on his website.
He was in regular communication

670
00:37:05,160 --> 00:37:09,000
with his family, and he was
actively preparing to testify in

671
00:37:09,000 --> 00:37:13,600
major copyright cases.
So this is not like the classic

672
00:37:13,600 --> 00:37:17,240
behavioral shifts that happened
before someone would commit

673
00:37:17,240 --> 00:37:19,960
suicide.
This is it's questionable.

674
00:37:20,200 --> 00:37:24,120
And I do want to note that
oftentimes suicides will come

675
00:37:24,120 --> 00:37:28,360
without warning, sure, but in a
case with other inconsistencies.

676
00:37:28,360 --> 00:37:30,840
That's exactly it.
This becomes another pattern of

677
00:37:30,840 --> 00:37:33,280
like this shit is not adding up.
Right.

678
00:37:33,840 --> 00:37:38,200
So the last thing that they
bring up is this timing was too

679
00:37:38,200 --> 00:37:40,840
convenient, it was too
coincidental, right?

680
00:37:40,840 --> 00:37:42,480
That's.
The other thing exactly.

681
00:37:42,840 --> 00:37:47,560
You know he died less than two
weeks after being named in court

682
00:37:47,560 --> 00:37:50,520
filings as someone who held
unique and relevant documents.

683
00:37:50,520 --> 00:37:52,560
To shut something completely
down.

684
00:37:52,720 --> 00:37:57,680
Yeah, related to open AI
lawsuits related to these

685
00:37:57,680 --> 00:38:00,840
lawsuits and he dies 2 weeks
later, no.

686
00:38:01,520 --> 00:38:03,360
No, I don't buy it.
Don't buy it.

687
00:38:03,360 --> 00:38:06,240
No, no, I don't buy it either.
And his parents actually talk

688
00:38:06,240 --> 00:38:10,880
about him expressing like, the
stress and the unease around

689
00:38:10,880 --> 00:38:14,720
these legal implications of him
speaking out so.

690
00:38:14,720 --> 00:38:18,720
I'm sure he was upset, I'm sure
he was scared.

691
00:38:18,840 --> 00:38:20,960
Yeah, he's 26 year old, years
old.

692
00:38:20,960 --> 00:38:23,240
Exactly.
I.

693
00:38:23,520 --> 00:38:27,080
Can't.
I just can't buy the fact that

694
00:38:27,080 --> 00:38:30,080
he was so upset and distraught
that he kills himself.

695
00:38:30,120 --> 00:38:32,040
I can't buy it.
Yeah.

696
00:38:32,120 --> 00:38:33,080
Can't do it.
Yeah.

697
00:38:33,080 --> 00:38:37,320
And I think that, like, even if
nothing nefarious happened, all

698
00:38:37,320 --> 00:38:40,240
of these things, you know, are
bothersome.

699
00:38:40,240 --> 00:38:44,440
But really for me, it's like the
timing is the thing that is the

700
00:38:44,440 --> 00:38:46,160
biggest question.
Exactly.

701
00:38:47,440 --> 00:38:51,640
So it's important to hear like,
their thought process around

702
00:38:51,640 --> 00:38:54,960
here because like having missing
evidence, like this missing

703
00:38:54,960 --> 00:38:57,400
evidence, that's not normal.
It should never be normal.

704
00:38:57,400 --> 00:38:59,680
Like they don't have like the
surveillance he was living.

705
00:38:59,680 --> 00:39:02,920
It's not like he was living in
like, you know, some shitty

706
00:39:02,920 --> 00:39:04,720
apartment complex.
Exactly.

707
00:39:04,800 --> 00:39:08,040
He was living in a nice place
that it puts into question the

708
00:39:08,040 --> 00:39:11,200
timelines of when he died, What,
who was around?

709
00:39:11,200 --> 00:39:13,120
Was anyone around?
You know, when this

710
00:39:13,120 --> 00:39:17,880
documentation is not complete,
you know, we want clarity, we

711
00:39:17,880 --> 00:39:19,880
want to be able to connect the
dots.

712
00:39:19,880 --> 00:39:24,120
So transparency, when we are
connecting dots, it really does

713
00:39:24,120 --> 00:39:27,800
protect everyone.
And you know, unfortunately,

714
00:39:27,800 --> 00:39:30,200
whistleblowers, people who speak
out against these big

715
00:39:30,200 --> 00:39:35,480
organizations, governments,
etcetera, they are often in this

716
00:39:35,480 --> 00:39:38,920
really uncomfortable position
where they're trying to tell the

717
00:39:38,920 --> 00:39:45,000
truth and there are big powers
that have an invested interest

718
00:39:45,000 --> 00:39:47,000
in them not getting the word
out.

719
00:39:47,680 --> 00:39:50,800
Exactly.
You know, the scrutiny is not a

720
00:39:50,800 --> 00:39:53,400
conspiracy.
It's really accountability.

721
00:39:54,040 --> 00:39:59,200
And the more powerful a company,
a government, an institution is,

722
00:39:59,200 --> 00:40:04,600
the higher of a burden that we
there is on investigators to

723
00:40:04,960 --> 00:40:07,400
rule things out and not rush
them through, right?

724
00:40:07,480 --> 00:40:11,040
Of course, but power and money
rule.

725
00:40:11,280 --> 00:40:14,160
Yes, yeah, they do.
They always rule.

726
00:40:14,360 --> 00:40:18,000
And, you know, oftentimes when,
like, tragedies like this

727
00:40:18,000 --> 00:40:21,520
happen, the families are the
first ones to spot these

728
00:40:21,520 --> 00:40:23,520
inconsistencies.
They're the first.

729
00:40:23,720 --> 00:40:26,480
Like, there are so many cases
that we have heard about where

730
00:40:26,480 --> 00:40:29,120
people are saying that you said
that this was a suicide.

731
00:40:29,120 --> 00:40:31,240
I mean, come on.
Look no further than Jeffrey

732
00:40:31,240 --> 00:40:32,640
Epstein.
It does not make sense.

733
00:40:32,640 --> 00:40:38,920
His family literally hired an
outside autopsy report, a

734
00:40:38,920 --> 00:40:42,120
coroner to do it because they
were inconsistencies and their

735
00:40:42,120 --> 00:40:45,600
intuition is not evidence.
I'm not saying that, but I am

736
00:40:45,600 --> 00:40:48,400
saying that it is information
that we should be paying

737
00:40:48,400 --> 00:40:52,320
attention to.
And the public really does have

738
00:40:52,360 --> 00:40:56,440
a stake in how whistleblower
deaths are investigated.

739
00:40:57,000 --> 00:40:59,160
Yes, you know.
Yes, because the more this

740
00:40:59,160 --> 00:41:01,720
happens, the less people will be
wanting.

741
00:41:02,080 --> 00:41:06,320
If every time a whistleblower of
something this high of stakes

742
00:41:06,920 --> 00:41:10,120
kills themselves, the less
likely we will have

743
00:41:10,360 --> 00:41:12,200
whistleblowers.
Yeah, yeah.

744
00:41:12,520 --> 00:41:15,360
Which is their point, which is
the intent.

745
00:41:15,480 --> 00:41:20,440
Exactly, exactly.
It silences these people who

746
00:41:20,520 --> 00:41:23,280
have scruples, who have, like, a
moral compass.

747
00:41:23,840 --> 00:41:26,840
It silences them.
And then it creates these

748
00:41:26,840 --> 00:41:29,160
vacuums of power.
It's a problem.

749
00:41:29,200 --> 00:41:31,680
Yeah.
So even if, like, here's the

750
00:41:31,680 --> 00:41:34,960
thing, even if the ruling is
correct, like, let's say, let's

751
00:41:35,040 --> 00:41:38,360
say he did, you know, commit
suicide or whatever, they still

752
00:41:38,360 --> 00:41:41,520
deserve answers as to why these
things are not making.

753
00:41:41,560 --> 00:41:43,600
Like, why is there missing
footage?

754
00:41:43,600 --> 00:41:46,880
Why, Like, what is like, all
this stuff, all this timing is

755
00:41:46,880 --> 00:41:50,640
really convenient.
Why, why the toxicology?

756
00:41:50,640 --> 00:41:53,880
They're like, he's never been
someone to do any kind of drugs,

757
00:41:54,320 --> 00:41:58,680
you know, and they deserve to
have an investigation like done

758
00:41:58,680 --> 00:42:01,720
fully and not not have it be
ruled suicide like

759
00:42:01,720 --> 00:42:05,560
automatically, you know, So I
don't know.

760
00:42:05,560 --> 00:42:09,120
I just think that, you know, his
parents don't necessarily have

761
00:42:09,120 --> 00:42:11,840
the right to be justified, but,
you know, their questions do

762
00:42:11,840 --> 00:42:14,920
matter.
And the evidence around their

763
00:42:14,920 --> 00:42:18,080
son's death was not clean or
simple or complete.

764
00:42:18,080 --> 00:42:20,200
And there are gaps in this
information.

765
00:42:20,200 --> 00:42:23,800
It's for me, it's black and
white and I don't, I'm not, I'm

766
00:42:23,800 --> 00:42:27,600
never one to really just be.
I'm not a judgey person at all.

767
00:42:27,600 --> 00:42:30,480
And I and I always, I am about
facts.

768
00:42:31,160 --> 00:42:44,760
But this to me seems so obvious,
just because like if it wasn't

769
00:42:44,760 --> 00:42:47,160
chat, if it was something else
maybe.

770
00:42:47,160 --> 00:42:51,520
But this is so big.
I mean this, this is the biggest

771
00:42:52,680 --> 00:42:59,280
alteration of humanity since I
don't know, maybe the wheel, I

772
00:42:59,280 --> 00:43:01,520
don't know, a car of a
telephone, I don't know.

773
00:43:01,520 --> 00:43:05,480
But it's life altering.
This is totally a life altering.

774
00:43:05,880 --> 00:43:16,040
So for something to be involved,
it's that big, tells you it's

775
00:43:16,040 --> 00:43:20,720
that the money is that
equivalent and people will do

776
00:43:20,720 --> 00:43:24,680
anything to not lose the money
anything.

777
00:43:24,880 --> 00:43:31,120
Yeah, yes.
It's, you know, when you have a

778
00:43:31,120 --> 00:43:37,320
brilliant person who ends up
being a whistleblower die at a

779
00:43:37,320 --> 00:43:41,480
moment when he's, you know,
there's legal, there are legal

780
00:43:41,480 --> 00:43:46,880
battles, there are questions
that should be asked, and we may

781
00:43:46,880 --> 00:43:50,080
never know what the answer is.
No, you know, I'm going to say

782
00:43:50,080 --> 00:43:54,960
we're not going to because no,
the nobody the powerful people

783
00:43:54,960 --> 00:43:56,480
don't want will never let us
know.

784
00:43:57,080 --> 00:43:57,800
Right.
Yeah.

785
00:43:58,520 --> 00:44:01,040
Because they have the control.
They do, yeah.

786
00:44:01,440 --> 00:44:02,840
Yeah.
And that's why I think it's

787
00:44:02,960 --> 00:44:07,760
important for us to make note
and talk about these things that

788
00:44:07,760 --> 00:44:10,720
that are happening.
Like if it gets brushed under

789
00:44:10,720 --> 00:44:14,240
the rug and we say, oh, it's
just some guy who was depressed,

790
00:44:14,240 --> 00:44:15,960
whatever.
We're not asking these

791
00:44:15,960 --> 00:44:18,000
questions.
It's going to happen to more and

792
00:44:18,000 --> 00:44:19,400
more people.
It's going to continue.

793
00:44:19,400 --> 00:44:22,840
You know, we're going to, for
whatever reason, be naive and

794
00:44:22,840 --> 00:44:27,480
trust these, like organizations.
And then the things that they're

795
00:44:27,480 --> 00:44:29,880
warning US against are
inevitably happening.

796
00:44:30,040 --> 00:44:30,560
Right.
Exactly.

797
00:44:30,640 --> 00:44:31,920
Inevitably.
Yeah.

798
00:44:32,440 --> 00:44:35,320
Yeah, it always ends up coming
out, these whistleblowers that

799
00:44:35,320 --> 00:44:37,360
come out and then all of a
sudden like have a heart attack.

800
00:44:37,440 --> 00:44:42,040
Yeah, yeah, that's not like the
first example of this.

801
00:44:42,120 --> 00:44:44,880
Yeah, it's not, it's not the
first, It's not even the 10th.

802
00:44:44,880 --> 00:44:48,080
It's not, you know, it's like,
like you could name a fucking

803
00:44:48,080 --> 00:44:52,280
number and you could, you know,
it's, there's so many of these

804
00:44:52,800 --> 00:44:56,920
that we need to like as humans
who are just trying our hardest.

805
00:44:57,200 --> 00:44:59,960
It's important for us to to pay
attention to this shit.

806
00:44:59,960 --> 00:45:02,880
Exactly and.
Guess what folks like?

807
00:45:03,320 --> 00:45:05,960
I use chachi PT.
I use like AI.

808
00:45:06,120 --> 00:45:12,760
I actually pay for it.
Exactly, exactly.

809
00:45:13,080 --> 00:45:14,680
It's a problem, but it's a
problem.

810
00:45:14,680 --> 00:45:17,800
And, you know, these companies
are not to be trusted.

811
00:45:17,800 --> 00:45:19,480
I've talked.
We were just talking about it

812
00:45:19,480 --> 00:45:23,160
how like, you know, I'm getting
away from Google and I'm trying

813
00:45:23,160 --> 00:45:27,160
to use other, you know, other
text sources because you can't

814
00:45:27,160 --> 00:45:29,880
trust your data with anything.
And like this guy was trying to

815
00:45:29,880 --> 00:45:33,000
say that he's like, all this
data is being mined and you

816
00:45:33,000 --> 00:45:36,920
can't trust it.
And, well, it's very interesting

817
00:45:36,920 --> 00:45:40,080
that he's not able to speak
about it anymore.

818
00:45:41,920 --> 00:45:43,680
No good.
Yeah, it's not good at all.

819
00:45:43,680 --> 00:45:45,280
No good.
Pay attention, people.

820
00:45:45,800 --> 00:45:46,400
Pay attention.
Don't.

821
00:45:46,400 --> 00:45:49,480
Trust, yeah, and don't just blow
shit under the rug.

822
00:45:49,480 --> 00:45:54,680
We all need to stay vigilant.
Yes, now more than ever, we are

823
00:45:54,680 --> 00:45:55,760
so much more.
Than.

824
00:45:56,120 --> 00:45:58,360
Ever we're more connected, we
have technology, we have sources

825
00:45:58,360 --> 00:46:01,920
like look at things, read
things, read multiple sources.

826
00:46:02,200 --> 00:46:05,760
Don't take the first source that
you are listening to, you know,

827
00:46:05,760 --> 00:46:08,680
try to diversify the information
that you're getting because

828
00:46:09,360 --> 00:46:13,280
yeah, it's clearly we're getting
less information and it's

829
00:46:13,280 --> 00:46:14,880
becoming more and more narrow
focus.

830
00:46:14,880 --> 00:46:17,400
So it's a problem.
Damn.

831
00:46:17,400 --> 00:46:21,200
All right, we'll make sure and
follow us on social media.

832
00:46:21,200 --> 00:46:23,120
Yes.
Let us know what you guys want

833
00:46:23,120 --> 00:46:25,320
to hear.
You can watch our episodes on

834
00:46:25,320 --> 00:46:28,080
our website, thatwasraf.com.
You can actually watch them on

835
00:46:28,080 --> 00:46:29,320
Spotify too.
Yeah.

836
00:46:30,320 --> 00:46:32,440
And just let us know what you
guys think.

837
00:46:32,600 --> 00:46:36,000
Yes, yeah, leave a review like
comments with grandma and shit.

838
00:46:36,120 --> 00:46:38,720
Yeah, yeah, yeah.
Thank you guys for great.

839
00:46:38,760 --> 00:46:40,240
Yeah.
Thank you so much.

840
00:46:40,240 --> 00:46:42,960
Yeah, we'll we'll see you soon.
See you soon.

841
00:46:43,040 --> 00:46:43,600
Bye.