1
00:00:00,000 --> 00:00:05,200
All right, so get this, imagine like you're trying to build software, but the tools you're

2
00:00:05,200 --> 00:00:08,800
using are like constantly changing, like while you're using them.

3
00:00:08,800 --> 00:00:09,800
Yeah.

4
00:00:09,800 --> 00:00:11,800
That's kind of the world that OpenAI is working in right now.

5
00:00:11,800 --> 00:00:13,040
Yeah, definitely.

6
00:00:13,040 --> 00:00:17,640
And we dove into, you know, the stack of articles and research you gave us, even that transcript

7
00:00:17,640 --> 00:00:19,560
from the tech conference.

8
00:00:19,560 --> 00:00:24,000
And it feels like the future of AI is a lot like that quote we found.

9
00:00:24,000 --> 00:00:25,000
Okay.

10
00:00:25,000 --> 00:00:27,520
The technology floor is not fixed.

11
00:00:27,520 --> 00:00:31,760
It is a fascinating time to be a part of the AI world.

12
00:00:31,760 --> 00:00:34,760
It really is like building on shifting sand.

13
00:00:34,760 --> 00:00:35,760
Oh, wow.

14
00:00:35,760 --> 00:00:39,960
Because what's possible is changing so rapidly, it's kind of hard to wrap your head around.

15
00:00:39,960 --> 00:00:40,960
Really?

16
00:00:40,960 --> 00:00:41,960
Yeah.

17
00:00:41,960 --> 00:00:43,360
Okay, so we have this transcript from Peter Dang, right?

18
00:00:43,360 --> 00:00:44,360
Yeah.

19
00:00:44,360 --> 00:00:46,120
And he's worked at some pretty big companies, right?

20
00:00:46,120 --> 00:00:47,120
Twitter, Instagram.

21
00:00:47,120 --> 00:00:48,640
He's got quite a resume.

22
00:00:48,640 --> 00:00:51,800
And now OpenAI, talking about a career trajectory.

23
00:00:51,800 --> 00:00:53,720
I know, right?

24
00:00:53,720 --> 00:00:59,240
But he actually says that OpenAI doesn't always know what their AI can do until like…

25
00:00:59,240 --> 00:01:00,240
Until it's ready, yeah.

26
00:01:00,240 --> 00:01:01,600
…until the model's fully baked.

27
00:01:01,600 --> 00:01:02,600
Is that…

28
00:01:02,600 --> 00:01:03,600
Baked, yeah.

29
00:01:03,600 --> 00:01:04,600
…is that normal?

30
00:01:04,600 --> 00:01:07,520
It feels like baking a cake and you don't know if it's chocolate or vanilla until you

31
00:01:07,520 --> 00:01:08,520
take a bite.

32
00:01:08,520 --> 00:01:09,520
Yeah.

33
00:01:09,520 --> 00:01:12,960
It's more common than you might think in the AI world these days.

34
00:01:12,960 --> 00:01:13,960
Really?

35
00:01:13,960 --> 00:01:15,480
Especially with these newer models that are coming out.

36
00:01:15,480 --> 00:01:16,480
Okay.

37
00:01:16,480 --> 00:01:17,480
Yeah.

38
00:01:17,480 --> 00:01:21,040
And Dang really highlights that traditional tech usually has, you know, a relatively stable

39
00:01:21,040 --> 00:01:22,400
base to build on.

40
00:01:22,400 --> 00:01:23,400
Right.

41
00:01:23,400 --> 00:01:27,360
And with OpenAI, that base is constantly evolving, always changing.

42
00:01:27,360 --> 00:01:28,360
Okay.

43
00:01:28,360 --> 00:01:29,360
Yeah.

44
00:01:29,360 --> 00:01:32,000
And so he even says it can feel like a guessing game sometimes.

45
00:01:32,000 --> 00:01:36,560
So it's less about having like a detailed roadmap and more about…

46
00:01:36,560 --> 00:01:37,560
Exploration.

47
00:01:37,560 --> 00:01:38,560
…exploration and experimentation.

48
00:01:38,560 --> 00:01:39,560
Hey.

49
00:01:39,560 --> 00:01:40,560
So it's…

50
00:01:40,560 --> 00:01:41,560
That's gotta see what sticks.

51
00:01:41,560 --> 00:01:42,560
Yeah.

52
00:01:42,560 --> 00:01:44,440
And that kind of feeds into another point that Dang makes about the speed at which AI

53
00:01:44,440 --> 00:01:48,040
is becoming, you know, both cheaper and more powerful at the same time.

54
00:01:48,040 --> 00:01:49,040
Okay.

55
00:01:49,040 --> 00:01:52,720
And he uses the example of GPT-3 versus GPT-4.

56
00:01:52,720 --> 00:01:56,960
And he was saying that in under two years, GPT-4 became significantly more capable.

57
00:01:56,960 --> 00:01:57,960
Wow.

58
00:01:57,960 --> 00:02:00,760
And the cost to run it dropped by like 99%.

59
00:02:00,760 --> 00:02:01,760
Seriously?

60
00:02:01,760 --> 00:02:02,760
Yeah.

61
00:02:02,760 --> 00:02:05,560
That's the kind of exponential progress that we're talking about here.

62
00:02:05,560 --> 00:02:10,320
That's like saying that your monthly software subscription just got practically free.

63
00:02:10,320 --> 00:02:11,320
Right.

64
00:02:11,320 --> 00:02:13,440
But it also does 10 times what it used to.

65
00:02:13,440 --> 00:02:14,440
Exactly.

66
00:02:14,440 --> 00:02:17,480
What does that kind of leap even mean for developers who are using this?

67
00:02:17,480 --> 00:02:19,320
I mean, it's a game changer.

68
00:02:19,320 --> 00:02:21,960
It unlocks so many possibilities.

69
00:02:21,960 --> 00:02:27,360
Dang was talking about how AI is becoming more accessible to, you know, a wider range

70
00:02:27,360 --> 00:02:28,960
of developers and applications.

71
00:02:28,960 --> 00:02:33,160
And it's impacting fields like education and healthcare.

72
00:02:33,160 --> 00:02:34,200
Wow.

73
00:02:34,200 --> 00:02:38,360
Even travel in ways that we could only dream of a few years ago.

74
00:02:38,360 --> 00:02:39,360
Yeah.

75
00:02:39,360 --> 00:02:40,840
And it's not just theoretical anymore either.

76
00:02:40,840 --> 00:02:41,840
Right.

77
00:02:41,840 --> 00:02:47,520
Because Dang mentions using ChatGPT as a real-time translator while he was traveling in Seoul

78
00:02:47,520 --> 00:02:48,520
and Tokyo.

79
00:02:48,520 --> 00:02:49,520
Oh, wow.

80
00:02:49,520 --> 00:02:51,520
That's real-world impact right there.

81
00:02:51,520 --> 00:02:52,520
That's incredible.

82
00:02:52,520 --> 00:02:53,520
Yeah.

83
00:02:53,520 --> 00:02:54,520
And he says that's just the start.

84
00:02:54,520 --> 00:02:55,520
Really?

85
00:02:55,520 --> 00:02:56,520
Yeah.

86
00:02:56,520 --> 00:02:58,920
He says we're moving towards AI that goes beyond just answering questions to actually

87
00:02:58,920 --> 00:03:01,040
completing complex tasks.

88
00:03:01,040 --> 00:03:02,040
Like what?

89
00:03:02,040 --> 00:03:05,920
Think like, you know, AI personal assistants that can handle multi-step projects.

90
00:03:05,920 --> 00:03:07,720
So like booking a whole trip for you?

91
00:03:07,720 --> 00:03:08,720
Yeah.

92
00:03:08,720 --> 00:03:10,160
Or like writing a podcast script?

93
00:03:10,160 --> 00:03:16,280
Maybe Dang emphasizes this shift from AI as a tool to AI as a collaborator.

94
00:03:16,280 --> 00:03:17,280
Collaborator.

95
00:03:17,280 --> 00:03:18,280
Yeah.

96
00:03:18,280 --> 00:03:20,880
Imagine interacting with AI the same way you brainstorm with a colleague, right?

97
00:03:20,880 --> 00:03:22,040
That natural back and forth.

98
00:03:22,040 --> 00:03:24,800
That's what he says we need to be aiming for.

99
00:03:24,800 --> 00:03:28,040
So are we talking about like designing the personality of AI?

100
00:03:28,040 --> 00:03:29,040
Kind of.

101
00:03:29,040 --> 00:03:30,320
That's a little spooky, isn't it?

102
00:03:30,320 --> 00:03:31,320
It's interesting.

103
00:03:31,320 --> 00:03:35,680
He talks about user experience and how it's not just about making AI smarter.

104
00:03:35,680 --> 00:03:36,680
Yeah.

105
00:03:36,680 --> 00:03:40,480
But it's also about making it more human-like in its interactions.

106
00:03:40,480 --> 00:03:41,480
So meaning what?

107
00:03:41,480 --> 00:03:42,800
Like response times?

108
00:03:42,800 --> 00:03:43,800
Yeah.

109
00:03:43,800 --> 00:03:46,600
How it thinks through problems, even its tone?

110
00:03:46,600 --> 00:03:47,600
All of that.

111
00:03:47,600 --> 00:03:48,600
Okay.

112
00:03:48,600 --> 00:03:53,480
So he does make AI more like us in how it interacts and how it problem solves.

113
00:03:53,480 --> 00:03:58,480
But Dang also worked at companies like Twitter and Instagram, which are much more like established

114
00:03:58,480 --> 00:03:59,480
tech companies.

115
00:03:59,480 --> 00:04:03,000
And he talks about the difference in pace of development between those companies and

116
00:04:03,000 --> 00:04:04,000
OpenAI.

117
00:04:04,000 --> 00:04:05,000
He does.

118
00:04:05,000 --> 00:04:06,000
Yeah.

119
00:04:06,000 --> 00:04:09,760
He says that the biggest contrast he's seen is this constant state of flux at OpenAI.

120
00:04:09,760 --> 00:04:10,760
Yeah.

121
00:04:10,760 --> 00:04:12,920
Like changing all the time, you know?

122
00:04:12,920 --> 00:04:17,480
So imagine being an engineer and the ground rules of your work are changing every few

123
00:04:17,480 --> 00:04:18,480
months.

124
00:04:18,480 --> 00:04:19,480
Oh, wow.

125
00:04:19,480 --> 00:04:20,480
That's wild.

126
00:04:20,480 --> 00:04:21,480
Yeah.

127
00:04:21,480 --> 00:04:24,120
That's like trying to build a house while the blueprints are still being drawn.

128
00:04:24,120 --> 00:04:25,120
It really is.

129
00:04:25,120 --> 00:04:30,720
How do you even begin to create a product roadmap in that kind of environment?

130
00:04:30,720 --> 00:04:31,720
That's a great question.

131
00:04:31,720 --> 00:04:32,720
Yeah.

132
00:04:32,720 --> 00:04:34,440
And it's something that Dang really grapples with.

133
00:04:34,440 --> 00:04:40,760
He says that often OpenAI doesn't have like this crystal clear roadmap because they're

134
00:04:40,760 --> 00:04:44,440
discovering new capabilities of their models all the time.

135
00:04:44,440 --> 00:04:45,440
Oh, wow.

136
00:04:45,440 --> 00:04:46,640
Even during the development process.

137
00:04:46,640 --> 00:04:50,360
So it's less about let's build feature X and it's more about like...

138
00:04:50,360 --> 00:04:53,400
It's like let's see what this thing can do and then figure out what to build.

139
00:04:53,400 --> 00:04:54,400
And see what it can do.

140
00:04:54,400 --> 00:04:55,400
Okay.

141
00:04:55,400 --> 00:05:00,620
He uses the term iterative to describe how they approach building products.

142
00:05:00,620 --> 00:05:02,120
They release a model.

143
00:05:02,120 --> 00:05:05,680
They see how people use it, see what works, what doesn't work.

144
00:05:05,680 --> 00:05:06,680
Okay.

145
00:05:06,680 --> 00:05:07,680
And then they refine it.

146
00:05:07,680 --> 00:05:11,920
So they're constantly like adapting and iterating based on real world feedback.

147
00:05:11,920 --> 00:05:12,920
Yes.

148
00:05:12,920 --> 00:05:15,000
Which makes sense given what we talked about earlier, right?

149
00:05:15,000 --> 00:05:17,520
About how the technology floor is always shifting.

150
00:05:17,520 --> 00:05:18,520
Always.

151
00:05:18,520 --> 00:05:22,320
But that also must be a challenge when it comes to like monetizing these products.

152
00:05:22,320 --> 00:05:23,320
Yeah.

153
00:05:23,320 --> 00:05:24,320
Right.

154
00:05:24,320 --> 00:05:25,320
Huge question.

155
00:05:25,320 --> 00:05:30,960
If you're constantly iterating, the landscape is changing so rapidly, how do you build a

156
00:05:30,960 --> 00:05:33,640
sustainable business model around that?

157
00:05:33,640 --> 00:05:35,640
That's what everybody's trying to figure out.

158
00:05:35,640 --> 00:05:39,920
And Dang even admits that OpenAI doesn't have all the answers yet.

159
00:05:39,920 --> 00:05:47,160
He says that traditional approaches like advertising premium subscriptions might not translate very

160
00:05:47,160 --> 00:05:48,960
well to the world of AI.

161
00:05:48,960 --> 00:05:51,760
Especially if the goal is to make AI more accessible.

162
00:05:51,760 --> 00:05:52,760
Exactly.

163
00:05:52,760 --> 00:05:58,040
It's hard to imagine charging a hefty subscription fee for something that's supposed to be as

164
00:05:58,040 --> 00:06:01,120
ubiquitous as like email.

165
00:06:01,120 --> 00:06:02,120
The internet.

166
00:06:02,120 --> 00:06:03,120
Yeah.

167
00:06:03,120 --> 00:06:04,120
Exactly.

168
00:06:04,120 --> 00:06:07,880
So it raises some really interesting questions about the future of AI, how it's going to

169
00:06:07,880 --> 00:06:09,240
be funded and governed.

170
00:06:09,240 --> 00:06:10,240
Yeah.

171
00:06:10,240 --> 00:06:15,080
If AI is going to be this transformative force, we can't just leave it up to market forces

172
00:06:15,080 --> 00:06:18,520
to decide how it's developed and who benefits from it.

173
00:06:18,520 --> 00:06:20,120
Dang seems to agree.

174
00:06:20,120 --> 00:06:25,400
He stresses that we need to start thinking now about the economic and social structures

175
00:06:25,400 --> 00:06:28,000
that will shape the future of AI.

176
00:06:28,000 --> 00:06:31,440
He says it's not just a technological challenge, it's a societal one as well.

177
00:06:31,440 --> 00:06:32,440
Okay.

178
00:06:32,440 --> 00:06:35,840
So big picture stuff aside, let's get back to like the tech for a second.

179
00:06:35,840 --> 00:06:39,680
Dang talks about a new open AI model called O1.

180
00:06:39,680 --> 00:06:40,800
Yes.

181
00:06:40,800 --> 00:06:42,600
What's so special about it?

182
00:06:42,600 --> 00:06:43,600
It's really fascinating.

183
00:06:43,600 --> 00:06:44,600
Okay.

184
00:06:44,600 --> 00:06:47,360
It gives us a glimpse into the future of AI.

185
00:06:47,360 --> 00:06:51,200
Dang describes O1 as a completely different kind of model.

186
00:06:51,200 --> 00:06:52,200
Really?

187
00:06:52,200 --> 00:06:53,200
Yeah.

188
00:06:53,200 --> 00:06:56,320
He says it goes beyond just retrieving information.

189
00:06:56,320 --> 00:06:59,800
It actually reasons through problems almost like a human would.

190
00:06:59,800 --> 00:07:00,800
Wait, hold on.

191
00:07:00,800 --> 00:07:03,120
Is O1 actually like thinking on its own?

192
00:07:03,120 --> 00:07:04,120
Thinking.

193
00:07:04,120 --> 00:07:07,040
Is there just enough sci-fi movies to know how that usually ends?

194
00:07:07,040 --> 00:07:08,040
Yeah.

195
00:07:08,040 --> 00:07:09,040
Not well.

196
00:07:09,040 --> 00:07:10,920
Yeah, not quite like in the movies, at least not yet.

197
00:07:10,920 --> 00:07:11,920
Okay.

198
00:07:11,920 --> 00:07:16,720
But Dang says O1 operates more like a detective solving a case, you know?

199
00:07:16,720 --> 00:07:17,720
Okay.

200
00:07:17,720 --> 00:07:21,480
It forms hypotheses, tests them against its knowledge base.

201
00:07:21,480 --> 00:07:22,480
Wow.

202
00:07:22,480 --> 00:07:24,880
And then kind of refines its thinking based on the results.

203
00:07:24,880 --> 00:07:25,880
So it's like...

204
00:07:25,880 --> 00:07:26,880
Imagine doing that.

205
00:07:26,880 --> 00:07:28,160
And constantly going back to the drawing board?

206
00:07:28,160 --> 00:07:29,160
Exactly.

207
00:07:29,160 --> 00:07:30,160
But at lightning speed.

208
00:07:30,160 --> 00:07:31,160
Wow.

209
00:07:31,160 --> 00:07:32,160
On a massive scale.

210
00:07:32,160 --> 00:07:33,160
Okay.

211
00:07:33,160 --> 00:07:35,600
And a little intimidating at the same time.

212
00:07:35,600 --> 00:07:36,600
Right.

213
00:07:36,600 --> 00:07:38,680
How do you even build something that can reason like that?

214
00:07:38,680 --> 00:07:40,680
I mean, it's incredibly complex.

215
00:07:40,680 --> 00:07:41,680
Yeah.

216
00:07:41,680 --> 00:07:47,080
But Dang explains that one of the key breakthroughs with O1 is its ability to handle those like

217
00:07:47,080 --> 00:07:49,000
multi-step reasoning tasks.

218
00:07:49,000 --> 00:07:50,000
Multi-step, okay.

219
00:07:50,000 --> 00:07:51,000
Yeah.

220
00:07:51,000 --> 00:07:55,640
He gives this example of a senior lawyer who used O1 to draft a legal brief.

221
00:07:55,640 --> 00:07:56,640
Right.

222
00:07:56,640 --> 00:07:57,640
Something that would usually take hours.

223
00:07:57,640 --> 00:07:58,640
Right.

224
00:07:58,640 --> 00:08:00,160
O1 was able to do it in minutes.

225
00:08:00,160 --> 00:08:01,160
Whoa.

226
00:08:01,160 --> 00:08:02,160
Yeah.

227
00:08:02,160 --> 00:08:03,160
So it's huge.

228
00:08:03,160 --> 00:08:04,160
It's a big deal.

229
00:08:04,160 --> 00:08:05,160
And not just for lawyers.

230
00:08:05,160 --> 00:08:06,160
Right.

231
00:08:06,160 --> 00:08:07,760
But for like anyone who works with complex information.

232
00:08:07,760 --> 00:08:08,760
Yeah.

233
00:08:08,760 --> 00:08:13,000
Anybody who has to, you know, analyze data, look for patterns, make decisions.

234
00:08:13,000 --> 00:08:14,160
It has huge implications.

235
00:08:14,160 --> 00:08:15,160
Okay.

236
00:08:15,160 --> 00:08:18,760
So Dang sees O1 as like a glimpse into this future.

237
00:08:18,760 --> 00:08:19,760
Yeah.

238
00:08:19,760 --> 00:08:22,800
Where AI isn't just answering our question.

239
00:08:22,800 --> 00:08:25,600
It's like actively collaborating with us.

240
00:08:25,600 --> 00:08:26,600
Working alongside us.

241
00:08:26,600 --> 00:08:28,320
On these really complex projects.

242
00:08:28,320 --> 00:08:29,320
Yeah.

243
00:08:29,320 --> 00:08:30,320
So we're talking like...

244
00:08:30,320 --> 00:08:31,320
AI partners.

245
00:08:31,320 --> 00:08:32,320
Okay.

246
00:08:32,320 --> 00:08:35,160
So does that mean that robots are taking our jobs?

247
00:08:35,160 --> 00:08:36,160
Well...

248
00:08:36,160 --> 00:08:40,200
Like seriously, what does this kind of AI mean for the average person?

249
00:08:40,200 --> 00:08:42,160
I mean that's the big question, right?

250
00:08:42,160 --> 00:08:47,120
Dang acknowledges that there is the potential for disruption in the workforce.

251
00:08:47,120 --> 00:08:48,120
Okay.

252
00:08:48,120 --> 00:08:49,400
But he's optimistic overall.

253
00:08:49,400 --> 00:08:50,400
Okay.

254
00:08:50,400 --> 00:08:53,680
He believes that AI will ultimately create new opportunities.

255
00:08:53,680 --> 00:08:54,680
Okay.

256
00:08:54,680 --> 00:08:57,080
You know, free us up from these tedious tasks.

257
00:08:57,080 --> 00:08:58,080
Right.

258
00:08:58,080 --> 00:09:02,560
And focus on the things that require, you know, creativity and critical thinking.

259
00:09:02,560 --> 00:09:04,120
Things that AI isn't quite ready for.

260
00:09:04,120 --> 00:09:05,120
Okay.

261
00:09:05,120 --> 00:09:08,020
I can get behind that, but it also sounds like we've all got some learning and adapting

262
00:09:08,020 --> 00:09:09,020
to do.

263
00:09:09,020 --> 00:09:10,020
For sure.

264
00:09:10,020 --> 00:09:12,320
Like to stay ahead of the curve, so to speak.

265
00:09:12,320 --> 00:09:14,720
Lifelong learning is going to be more important than ever.

266
00:09:14,720 --> 00:09:16,600
But Dang's advice is really interesting.

267
00:09:16,600 --> 00:09:21,520
He says instead of, you know, being afraid of what AI can do, focus on what it can't

268
00:09:21,520 --> 00:09:22,520
do yet.

269
00:09:22,520 --> 00:09:23,520
Okay.

270
00:09:23,520 --> 00:09:27,160
Because those are the areas where humans will continue to add unique value.

271
00:09:27,160 --> 00:09:30,640
So like find the gaps and then build your expertise there.

272
00:09:30,640 --> 00:09:31,640
Yeah.

273
00:09:31,640 --> 00:09:32,640
So I like that.

274
00:09:32,640 --> 00:09:34,880
And remember what we talked about earlier, that technology floor.

275
00:09:34,880 --> 00:09:35,880
Yeah.

276
00:09:35,880 --> 00:09:36,880
Constantly shifting.

277
00:09:36,880 --> 00:09:37,880
Right.

278
00:09:37,880 --> 00:09:41,000
So what AI can't do today, it might be able to do tomorrow.

279
00:09:41,000 --> 00:09:42,000
Wow.

280
00:09:42,000 --> 00:09:43,000
Yeah.

281
00:09:43,000 --> 00:09:48,720
So just staying curious, adapting to new tools and ways of thinking, that's going to be key.

282
00:09:48,720 --> 00:09:51,320
This entire deep dive has been a wild ride, let me tell you.

283
00:09:51,320 --> 00:09:52,320
I know, right?

284
00:09:52,320 --> 00:09:59,240
We've gone from AI that translates languages in real time to AI that can draft legal documents

285
00:09:59,240 --> 00:10:00,240
in minutes.

286
00:10:00,240 --> 00:10:01,240
It's incredible.

287
00:10:01,240 --> 00:10:02,240
My head is spinning.

288
00:10:02,240 --> 00:10:03,240
Mine too.

289
00:10:03,240 --> 00:10:05,000
And we've just scratched the surface.

290
00:10:05,000 --> 00:10:06,040
Just the beginning.

291
00:10:06,040 --> 00:10:08,120
The future of AI is full of possibility.

292
00:10:08,120 --> 00:10:09,120
Endless possibility.

293
00:10:09,120 --> 00:10:11,640
It's going to be fascinating to see how it all unfolds.

294
00:10:11,640 --> 00:10:12,640
It really is.

295
00:10:12,640 --> 00:10:17,240
And if this conversation is any indication, it's definitely not going to be boring.

296
00:10:17,240 --> 00:10:18,240
I agree.

297
00:10:18,240 --> 00:10:23,320
So for our listeners out there, keep exploring those what if questions because the future

298
00:10:23,320 --> 00:10:26,360
of AI is being written right now.

299
00:10:26,360 --> 00:10:27,440
And who knows?

300
00:10:27,440 --> 00:10:29,160
Maybe you hold the key to the next breakthrough.

301
00:10:29,160 --> 00:10:30,160
You never know.

302
00:10:30,160 --> 00:10:49,160
Until next time, happy exploring.