1
00:00:00,000 --> 00:00:10,080
Welcome to Artificially Intelligent Marketing, a weekly podcast where we stay on top of the

2
00:00:10,080 --> 00:00:15,800
latest trends, tips and tools in the world of marketing AI, helping you get the best

3
00:00:15,800 --> 00:00:18,640
results from your marketing efforts.

4
00:00:18,640 --> 00:00:24,200
Now let's join our hosts, Paul Avery and Martin Broadhurst.

5
00:00:24,200 --> 00:00:30,080
Welcome to Artificially Intelligent Marketing, a podcast for marketers that goes into all

6
00:00:30,080 --> 00:00:34,760
the aspects of how AI can help them improve their marketing efforts.

7
00:00:34,760 --> 00:00:38,920
I am joined today by the man, the legend, Martin Broadhurst.

8
00:00:38,920 --> 00:00:40,200
How are you, Martin?

9
00:00:40,200 --> 00:00:44,520
I am splendid on this fine Friday.

10
00:00:44,520 --> 00:00:48,160
It's absolutely just cleared up outside, blue skies, but if we'd have started this podcast

11
00:00:48,160 --> 00:00:53,080
recording about 15 minutes ago, I don't think you'd have heard me due to the incredibly

12
00:00:53,080 --> 00:00:55,840
loud hailstorm taking place.

13
00:00:55,840 --> 00:00:56,840
You'd have been washed away.

14
00:00:56,840 --> 00:01:01,240
Are you sure that hailstorm wasn't a hailing down of AI related news stories?

15
00:01:01,240 --> 00:01:02,840
It does feel like that, doesn't it?

16
00:01:02,840 --> 00:01:04,320
It's been a bit crazy this week, isn't it?

17
00:01:04,320 --> 00:01:10,880
I read on a forum today that 2023 has been an interesting decade, which I thought was

18
00:01:10,880 --> 00:01:16,160
quite apt, but then I sat back and thought, actually, I think March 2023 has been an interesting

19
00:01:16,160 --> 00:01:18,280
decade and it hasn't even ended yet.

20
00:01:18,280 --> 00:01:20,880
Yeah, just like the last four nights.

21
00:01:20,880 --> 00:01:22,760
Madness, madness.

22
00:01:22,760 --> 00:01:25,920
So right, a deep breath.

23
00:01:25,920 --> 00:01:27,200
There is so much to cover today.

24
00:01:27,200 --> 00:01:31,800
We're going to try and rattle through it, waffle less than usual and give you hopefully

25
00:01:31,800 --> 00:01:32,800
just the salient facts.

26
00:01:32,800 --> 00:01:35,320
Don't point that face at me, Martin.

27
00:01:35,320 --> 00:01:36,800
I don't want to over promise.

28
00:01:36,800 --> 00:01:37,800
That is true.

29
00:01:37,800 --> 00:01:38,800
I'm waffling now.

30
00:01:38,800 --> 00:01:39,800
It's been already broken.

31
00:01:39,800 --> 00:01:40,800
I thought it was straight.

32
00:01:40,800 --> 00:01:41,800
Right.

33
00:01:41,800 --> 00:01:42,920
This is what we're going to talk about.

34
00:01:42,920 --> 00:01:45,840
We're going to talk about Bard, ChapGPT plugins.

35
00:01:45,840 --> 00:01:48,640
We're going to talk about camera AI, Adobe Firefly.

36
00:01:48,640 --> 00:01:51,040
That sounds like a bit of a wrap.

37
00:01:51,040 --> 00:01:52,840
We're going to talk about Nvidia foundations.

38
00:01:52,840 --> 00:01:56,520
We're going to talk about how Bing got an image generator, Runway ML teased the launch

39
00:01:56,520 --> 00:02:02,520
of their text video and Microsoft published 150 page paper on why GPT-4 is a big leap

40
00:02:02,520 --> 00:02:06,720
forward and a strong step towards artificial general intelligence.

41
00:02:06,720 --> 00:02:11,560
We're also going to look at what this means in relation to OpenAI's recent paper on what

42
00:02:11,560 --> 00:02:13,680
this is going to mean for people's jobs.

43
00:02:13,680 --> 00:02:16,440
We're going to use it to our usual look at tool of the week.

44
00:02:16,440 --> 00:02:19,960
We've got a couple of those in there and we've got a new section this week, which is our

45
00:02:19,960 --> 00:02:22,840
Ninja marketing and sales application of AI.

46
00:02:22,840 --> 00:02:27,960
Tell you a little story about what someone's doing and it's way around to the end for that

47
00:02:27,960 --> 00:02:29,760
because you're going to go, sorry, what?

48
00:02:29,760 --> 00:02:30,760
How can I do that?

49
00:02:30,760 --> 00:02:32,080
Because it is rather cool.

50
00:02:32,080 --> 00:02:33,080
Right.

51
00:02:33,080 --> 00:02:34,080
Let's get cracking.

52
00:02:34,080 --> 00:02:36,840
Let's talk ChapGPT plugins first of all.

53
00:02:36,840 --> 00:02:42,760
So yesterday my LinkedIn and my WhatsApp, thanks to Martin, tried to drive my phone

54
00:02:42,760 --> 00:02:50,280
into meltdown when ChapGPT launched its new plugins platform.

55
00:02:50,280 --> 00:02:51,280
Is it a platform?

56
00:02:51,280 --> 00:02:52,280
Yes.

57
00:02:52,280 --> 00:02:55,720
Is it additions to how ChapGPT works already?

58
00:02:55,720 --> 00:02:56,720
Yes.

59
00:02:56,720 --> 00:02:57,720
It's kind of mad.

60
00:02:57,720 --> 00:02:58,720
So what is it?

61
00:02:58,720 --> 00:03:05,320
So in essence, people can now build third party plugins on top of ChapGPT.

62
00:03:05,320 --> 00:03:09,040
And as a proof of principle, OpenAI is already built with few already, which includes things

63
00:03:09,040 --> 00:03:10,480
like a web browser.

64
00:03:10,480 --> 00:03:15,320
So for example, you can now query ChapGPT in natural language like you've been doing

65
00:03:15,320 --> 00:03:20,480
already, but it can now go visit a website and collect information that it thinks is

66
00:03:20,480 --> 00:03:23,420
relevant to your query and then bring it back to you.

67
00:03:23,420 --> 00:03:29,360
So the criticism that many people, including the folks on this podcast have had of ChapGPT

68
00:03:29,360 --> 00:03:33,800
is that it's only trained on data up to the middle of 2021, but now it can go visit the

69
00:03:33,800 --> 00:03:34,800
web.

70
00:03:34,800 --> 00:03:37,160
Assuming it goes to the right places.

71
00:03:37,160 --> 00:03:39,240
That's no longer a criticism anymore.

72
00:03:39,240 --> 00:03:47,200
They've also partnered up OpenAI with around 20 sort of initial providers, including Kayak.

73
00:03:47,200 --> 00:03:51,400
So you can now ask ChapGPT to go and find you hotels, flights, et cetera, based on a

74
00:03:51,400 --> 00:03:52,700
brief you give it.

75
00:03:52,700 --> 00:03:55,160
So I'd like to go to Paris at some point in April.

76
00:03:55,160 --> 00:03:59,080
I'm not willing to spend more than 400 quid, but I want to stay in a three star hotel,

77
00:03:59,080 --> 00:04:02,040
go and find some options and then book it for me.

78
00:04:02,040 --> 00:04:03,280
Pretty awesome.

79
00:04:03,280 --> 00:04:08,680
And also Zapier is included in that Zapier, Zapier, which in essence allows you to connect

80
00:04:08,680 --> 00:04:11,600
ChapGPT to 5,000 other apps.

81
00:04:11,600 --> 00:04:15,880
So you can ask for things in natural language and it will do it like add this person to

82
00:04:15,880 --> 00:04:21,440
my CRM, look for this piece of information in the spreadsheet, yada, yada, yada.

83
00:04:21,440 --> 00:04:23,840
So that's, that's pretty crazy.

84
00:04:23,840 --> 00:04:25,880
First thoughts on that, Martin.

85
00:04:25,880 --> 00:04:31,760
When I saw that Zapier was one of the launch partners, I knew this was interesting.

86
00:04:31,760 --> 00:04:33,960
Immediately you just go, Oh wow.

87
00:04:33,960 --> 00:04:40,680
They have really unlocked like they say 5,000 other apps that has access to.

88
00:04:40,680 --> 00:04:48,680
We were impressed the other week when we looked at ChapGPT being integrated into HubSpot with

89
00:04:48,680 --> 00:04:53,120
ChatSpot and EinsteinGPT.

90
00:04:53,120 --> 00:04:56,440
This is a whole leap forward yet again.

91
00:04:56,440 --> 00:04:59,480
So think about all of the integrations that you're going to be able to do.

92
00:04:59,480 --> 00:05:05,560
First I'm very excited to get my hands on that particular integration, but yeah, first

93
00:05:05,560 --> 00:05:09,280
thoughts were, wow, they are serious, aren't they?

94
00:05:09,280 --> 00:05:13,600
This is a platform play that is a big move and very exciting.

95
00:05:13,600 --> 00:05:14,600
Yeah.

96
00:05:14,600 --> 00:05:17,440
I think the platform play is really interesting one because what does this do?

97
00:05:17,440 --> 00:05:22,200
I mean, this basically opens up the opportunity for others to build plugins on top of it.

98
00:05:22,200 --> 00:05:26,720
So in essence, this is the first formings one assumes of like an app marketplace for

99
00:05:26,720 --> 00:05:27,720
ChatGPT.

100
00:05:27,720 --> 00:05:29,920
Yeah, very much so.

101
00:05:29,920 --> 00:05:34,480
And if you look at what that's done for other businesses, take the example of Salesforce

102
00:05:34,480 --> 00:05:35,920
App Exchange.

103
00:05:35,920 --> 00:05:41,480
That's something that has its whole ecosystem all on its own.

104
00:05:41,480 --> 00:05:44,760
I was to quick Google search before I jumped into this.

105
00:05:44,760 --> 00:05:48,840
1.6 billion USD is the value of that marketplace.

106
00:05:48,840 --> 00:05:51,360
That's the Salesforce one.

107
00:05:51,360 --> 00:05:55,440
How big is the ChatGPT marketplace going to be?

108
00:05:55,440 --> 00:05:56,440
Absolutely.

109
00:05:56,440 --> 00:06:00,600
I saw a comment on Twitter that the way this is going, we're going to be spending 90% of

110
00:06:00,600 --> 00:06:07,720
our time at work in front of a ChatGPT style interface, whether you agree with that or

111
00:06:07,720 --> 00:06:08,720
not.

112
00:06:08,720 --> 00:06:13,560
But that's kind of insane when you think about how fundamentally it will change how we work

113
00:06:13,560 --> 00:06:17,960
and therefore the ecosystem around that and how much value it's going to create for the

114
00:06:17,960 --> 00:06:22,200
people who can come up with the killer apps, which of course was the race in the smartphone

115
00:06:22,200 --> 00:06:25,200
era, right?

116
00:06:25,200 --> 00:06:33,720
So, yeah, angry birds coming to a ChatGPT window near you soon, except for you have to ask

117
00:06:33,720 --> 00:06:35,920
them what to do instead of clicking.

118
00:06:35,920 --> 00:06:37,640
Yes, very, very cool.

119
00:06:37,640 --> 00:06:41,800
I think the other thing, and I'm not a developer, so I can't really comment too much on this,

120
00:06:41,800 --> 00:06:44,480
but I did see something on Twitter which I thought was amazing, which is that it sounds

121
00:06:44,480 --> 00:06:47,400
like it's super easy to create your own plugin.

122
00:06:47,400 --> 00:06:53,840
So to quote Mitchell Hashemott, who was on Twitter, he wrote, you write an open API manifest

123
00:06:53,840 --> 00:06:58,920
for your API, you use human language descriptions for everything, and that's it.

124
00:06:58,920 --> 00:07:03,400
You let the model figure out how to auth chain calls, process data in between, format it

125
00:07:03,400 --> 00:07:05,120
for viewing, et cetera.

126
00:07:05,120 --> 00:07:08,880
So this is my complete layman's interpretation of that.

127
00:07:08,880 --> 00:07:14,680
All of the complexity that would usually come with building an API that can ensure data

128
00:07:14,680 --> 00:07:22,360
is moving between systems and being effectively read and interpreted by open AI system and

129
00:07:22,360 --> 00:07:29,760
ChatGPT is massively simplified because ChatGPT in essence does a lot of the heavy lifting

130
00:07:29,760 --> 00:07:33,000
of figuring out how your plugin should even work.

131
00:07:33,000 --> 00:07:35,720
Is that how you interpreted that, Mime?

132
00:07:35,720 --> 00:07:36,720
Yeah, exactly.

133
00:07:36,720 --> 00:07:42,840
And Greg Brodman, co-founder of open AI, basically confirmed that you write your instructions

134
00:07:42,840 --> 00:07:47,440
in human language for the machine to read and go, oh yeah, I got this.

135
00:07:47,440 --> 00:07:50,040
I can turn that into an easy plugin.

136
00:07:50,040 --> 00:07:51,040
No problem.

137
00:07:51,040 --> 00:07:52,040
Yeah, you leave it to me.

138
00:07:52,040 --> 00:07:53,040
It's mad, isn't it?

139
00:07:53,040 --> 00:07:54,040
It's mad.

140
00:07:54,040 --> 00:07:56,680
So we've been thinking like, what does this mean for marketers?

141
00:07:56,680 --> 00:08:00,880
And there's so much stuff to try and unpick here, so we'll crack through a few bits.

142
00:08:00,880 --> 00:08:06,560
I think the first thing is expect a lot more tools to emerge that let you do a whole host

143
00:08:06,560 --> 00:08:08,320
of things, right?

144
00:08:08,320 --> 00:08:11,740
Creating content based on info that you provide it access to.

145
00:08:11,740 --> 00:08:18,640
So an example would be you had a G-Drive folder with 10, I don't know, PDFs or HTML articles

146
00:08:18,640 --> 00:08:23,720
or something that in essence was the briefing material that then ChatGPT could create the

147
00:08:23,720 --> 00:08:27,640
content based off the back of that, not just its prediction engine based on what it's been

148
00:08:27,640 --> 00:08:28,640
trained on.

149
00:08:28,640 --> 00:08:33,100
You can go query your CRM database and ask it questions about your leads and your customers

150
00:08:33,100 --> 00:08:38,120
and who bought the most X from us last month and that type of stuff.

151
00:08:38,120 --> 00:08:42,080
And you can carry out a series of actions that automatically that would normally take

152
00:08:42,080 --> 00:08:47,160
a series of clicks, but driven by a natural language request.

153
00:08:47,160 --> 00:08:53,320
So it's almost like asking an assistant to go look in a spreadsheet or database or folder

154
00:08:53,320 --> 00:08:58,580
or email or something to go retrieve a bit of information for you and potentially even

155
00:08:58,580 --> 00:09:02,460
synthesize and summarize and tell you what's important from it, right?

156
00:09:02,460 --> 00:09:06,240
So that's pretty, pretty cool.

157
00:09:06,240 --> 00:09:09,840
I think the other thing that I've been thinking about, Martin, is if it's really that easy

158
00:09:09,840 --> 00:09:15,080
to build on, we might find that marketplaces pop up that make it easy for you to basically

159
00:09:15,080 --> 00:09:19,840
request, if you're not a developer, your own specific plugins that you can buy for your

160
00:09:19,840 --> 00:09:24,080
own use cases and you can get them built at a very low price point.

161
00:09:24,080 --> 00:09:28,800
So one of the ideas we were throwing about is I would love to be able to just ask questions

162
00:09:28,800 --> 00:09:33,080
of our project management system here at Biostrata in natural language and have the data fed

163
00:09:33,080 --> 00:09:39,120
to me, like how long on average does it take for us to write a blog post?

164
00:09:39,120 --> 00:09:40,120
I can do that.

165
00:09:40,120 --> 00:09:42,560
We have dashboards that we built for that, but it's pretty painstaking.

166
00:09:42,560 --> 00:09:48,360
It'd be pretty awesome if I could just ask TrackGPT to go surface that data and understood

167
00:09:48,360 --> 00:09:51,520
what I was asking for.

168
00:09:51,520 --> 00:09:55,160
And if it's that easy and the price comes down low enough, you won't even have to buy

169
00:09:55,160 --> 00:09:56,160
commercial tools.

170
00:09:56,160 --> 00:09:59,400
You might be able to get ones made specifically for you.

171
00:09:59,400 --> 00:10:02,840
So those are some of the things I was thinking on.

172
00:10:02,840 --> 00:10:06,560
I know you had some questions about what does this mean for SEO, Martin?

173
00:10:06,560 --> 00:10:07,560
Yeah.

174
00:10:07,560 --> 00:10:11,120
I mean, well, I mean, I don't really know the answer.

175
00:10:11,120 --> 00:10:15,280
We were talking about it on WhatsApp when we were saying, how does this impact, if this

176
00:10:15,280 --> 00:10:21,800
fundamentally changes the way that we interface with computers and technology wholesale, right?

177
00:10:21,800 --> 00:10:25,840
Now that this can go out and browse the web, now that this can go and literally do the

178
00:10:25,840 --> 00:10:29,960
search on Kayak, now that this can do all of the things that you just explained, when

179
00:10:29,960 --> 00:10:34,000
am I going to Google?

180
00:10:34,000 --> 00:10:35,440
Just how?

181
00:10:35,440 --> 00:10:38,800
And if you think about what that means from, even if it's just a small percentage drop,

182
00:10:38,800 --> 00:10:44,160
let's say, let's say it's a 15, 20% drop in me going to Google, like to begin with, that's

183
00:10:44,160 --> 00:10:47,840
a 15, 20% drop in revenue for Google.

184
00:10:47,840 --> 00:10:55,640
That's not insignificant for a company that is striving for continual growth year on year.

185
00:10:55,640 --> 00:10:58,840
For SEOs, how do you optimize for this?

186
00:10:58,840 --> 00:11:01,600
How do you make sure that your content is seen first?

187
00:11:01,600 --> 00:11:05,040
I just think it fundamentally is going to shift the way that we think about channels

188
00:11:05,040 --> 00:11:06,200
for marketers.

189
00:11:06,200 --> 00:11:14,040
If you are a software, if you've got anything like a digital product at all, so software

190
00:11:14,040 --> 00:11:15,040
companies, right?

191
00:11:15,040 --> 00:11:18,980
When it's a startup, one of the first things that they look to integrate with will be something

192
00:11:18,980 --> 00:11:20,120
like Zapier, right?

193
00:11:20,120 --> 00:11:25,040
Because it immediately makes your product accessible to a wide range of other people.

194
00:11:25,040 --> 00:11:31,240
Some people might integrate with, is it Microsoft Power Automate or something like that, and

195
00:11:31,240 --> 00:11:33,600
to access that ecosystem.

196
00:11:33,600 --> 00:11:37,600
The first integration that software companies are going to be doing now is a plugin with

197
00:11:37,600 --> 00:11:40,000
ChatGPT.

198
00:11:40,000 --> 00:11:46,600
I think it's absolutely spot on and it makes me think of the quote from Mark Andresen Horowitz,

199
00:11:46,600 --> 00:11:50,680
who said, I don't know, it was about a decade ago, wasn't it?

200
00:11:50,680 --> 00:11:52,240
The software is eating the world.

201
00:11:52,240 --> 00:11:55,040
I think that was absolutely true then.

202
00:11:55,040 --> 00:12:00,280
But software in the form of ChatGPT plugins could really eat the world as we know it,

203
00:12:00,280 --> 00:12:03,320
including eating all the software that came before it.

204
00:12:03,320 --> 00:12:07,360
So software is eating the software that's eating the world, or I guess more accurately,

205
00:12:07,360 --> 00:12:12,600
ChatGPT is eating the software that ate the world.

206
00:12:12,600 --> 00:12:14,000
Because I think you're absolutely right.

207
00:12:14,000 --> 00:12:20,520
One of the things I saw on the Twittersphere yesterday was, will there be a rush of websites?

208
00:12:20,520 --> 00:12:25,240
Because you can block your website from being accessed by ChatGPT so that it can't go query

209
00:12:25,240 --> 00:12:26,480
it.

210
00:12:26,480 --> 00:12:32,040
Will we see a rush of, especially the bigger publishing houses blocking ChatGPT from being

211
00:12:32,040 --> 00:12:33,440
able to go to their website?

212
00:12:33,440 --> 00:12:36,320
Because in essence, it means a user doesn't have to.

213
00:12:36,320 --> 00:12:46,840
And will that drive even more the emergence of pay to access content, gated content?

214
00:12:46,840 --> 00:12:53,360
Because in essence, ChatGPT will be surfacing information, but not really necessarily passing

215
00:12:53,360 --> 00:12:57,320
on any credit per se to those websites, maybe a provider link.

216
00:12:57,320 --> 00:13:00,720
But what if I click a link if I thought my query had already been answered?

217
00:13:00,720 --> 00:13:01,720
I don't know.

218
00:13:01,720 --> 00:13:04,200
And OpenAI have kind of acknowledged this themselves, haven't they?

219
00:13:04,200 --> 00:13:08,440
They said, look, this is going to change the way that people interact with tech.

220
00:13:08,440 --> 00:13:14,120
And we're aware that we will be taking some traffic away from places.

221
00:13:14,120 --> 00:13:21,680
And yeah, they pretty much thrown a hand grenade into the content mix and said, we appreciate

222
00:13:21,680 --> 00:13:23,880
your feedback on how we can figure this out.

223
00:13:23,880 --> 00:13:26,600
Yeah, what was that quote I sent you?

224
00:13:26,600 --> 00:13:28,840
I can't remember it.

225
00:13:28,840 --> 00:13:30,120
But in essence, it was that.

226
00:13:30,120 --> 00:13:34,920
It was like OpenAI went, yeah, this is going to cause some real major changes in how people

227
00:13:34,920 --> 00:13:36,520
create and consume content.

228
00:13:36,520 --> 00:13:37,840
And we don't know the answers.

229
00:13:37,840 --> 00:13:42,480
And then just sort of went, now everyone, this is your problem.

230
00:13:42,480 --> 00:13:44,040
Go figure out.

231
00:13:44,040 --> 00:13:48,040
We appreciate that this is a new method of interacting with the web and welcome feedback

232
00:13:48,040 --> 00:13:52,180
on additional ways to drive traffic back to your sources and add to the overall health

233
00:13:52,180 --> 00:13:53,600
of the ecosystem.

234
00:13:53,600 --> 00:13:54,600
Right.

235
00:13:54,600 --> 00:13:57,720
So no solutions whatsoever there then?

236
00:13:57,720 --> 00:14:03,320
As you said in the chat, and grenade dropped into the into the content marketing and SEO

237
00:14:03,320 --> 00:14:04,320
world.

238
00:14:04,320 --> 00:14:05,320
Yeah.

239
00:14:05,320 --> 00:14:06,320
Good luck, guys.

240
00:14:06,320 --> 00:14:07,320
Yeah.

241
00:14:07,320 --> 00:14:15,680
I mean, to be deliberately provocative, written content marketing in its current form could

242
00:14:15,680 --> 00:14:17,960
be about to die.

243
00:14:17,960 --> 00:14:20,240
Right.

244
00:14:20,240 --> 00:14:27,920
Because if your primary goal is to drive traffic to your website, gain subscribers, etc.

245
00:14:27,920 --> 00:14:31,120
This might make that significantly harder.

246
00:14:31,120 --> 00:14:36,000
If your primary goal is to be seen as a thought leader and chat GPT ends up somewhat surfacing

247
00:14:36,000 --> 00:14:40,840
information or giving credit to where the insights came from and names your company

248
00:14:40,840 --> 00:14:46,440
or your brand or you as a person, you could probably still get some of the value from

249
00:14:46,440 --> 00:14:47,440
it.

250
00:14:47,440 --> 00:14:52,920
But I think it puts a real pressure on that some of those content marketing models, because

251
00:14:52,920 --> 00:14:57,480
if you can't get people on your site, you can't get them to convert and subscribe or

252
00:14:57,480 --> 00:15:01,760
become a lead or all the other things that typically content marketers would do.

253
00:15:01,760 --> 00:15:07,240
I still think things like people talking nonsense on podcasts, just like this one, can still

254
00:15:07,240 --> 00:15:10,920
be a very useful way of just bringing some extra insight to things that you might not

255
00:15:10,920 --> 00:15:16,760
find even chat GPT can't find because it's coming out live and clearly unfiltered out

256
00:15:16,760 --> 00:15:18,560
of our heads, Martin.

257
00:15:18,560 --> 00:15:26,360
But yeah, it's OpenAI shaking the tree rather aggressively here and I think I heard a creak.

258
00:15:26,360 --> 00:15:28,800
Yeah, absolutely.

259
00:15:28,800 --> 00:15:32,200
So I think we've probably covered plugins in enough depth there.

260
00:15:32,200 --> 00:15:33,200
Agreed.

261
00:15:33,200 --> 00:15:35,360
Tell us about Nvidia, what they've been up to this week.

262
00:15:35,360 --> 00:15:36,360
Right.

263
00:15:36,360 --> 00:15:37,360
Yeah.

264
00:15:37,360 --> 00:15:38,920
So Nvidia Foundations, a big announcement from Nvidia.

265
00:15:38,920 --> 00:15:39,920
They've been doing a lot of work.

266
00:15:39,920 --> 00:15:47,040
If you don't know this company that we might familiarly think of as being a kind of graphics-based

267
00:15:47,040 --> 00:15:52,760
business with interests in graphics cards and things like that, they are heavily invested

268
00:15:52,760 --> 00:15:58,920
in AI and they made a big announcement this week about Nvidia Foundations, which is basically

269
00:15:58,920 --> 00:16:02,880
their cloud services that you can then build on top of.

270
00:16:02,880 --> 00:16:11,440
So you can do generative AI, they've got a thing called Bionimo or Bio-Nemo, Picasso,

271
00:16:11,440 --> 00:16:12,640
which is a suite of tools.

272
00:16:12,640 --> 00:16:16,980
They've got their own custom large language model, visual models, which can be tailored

273
00:16:16,980 --> 00:16:18,920
for domain specific tasks.

274
00:16:18,920 --> 00:16:24,600
So where we were talking about Bio-GPT the other day, they've got this Bionimo model,

275
00:16:24,600 --> 00:16:28,520
which I think we'll cover in more detail in a future week once we've had a bit more time

276
00:16:28,520 --> 00:16:31,260
to digest exactly what it means.

277
00:16:31,260 --> 00:16:36,400
It's available and accessible via the browser and it's simple API plugins.

278
00:16:36,400 --> 00:16:40,520
So again, they're pretty much doing what OpenAI is doing and saying, hey, we've got all of

279
00:16:40,520 --> 00:16:43,080
these tools, developers, please come and use them.

280
00:16:43,080 --> 00:16:46,960
You can access these tools, you can build your products on top of them.

281
00:16:46,960 --> 00:16:52,080
In terms of their launch partners, Adobe, Getty Images, Morningstar, Shutterstock, they're

282
00:16:52,080 --> 00:16:55,820
all collaborating on the product and they're saying they can be used for things like chat

283
00:16:55,820 --> 00:17:02,060
bots, customer service, content creation and digital simulation as well, which is an interesting

284
00:17:02,060 --> 00:17:03,060
use case.

285
00:17:03,060 --> 00:17:06,600
Obviously very big on the graphics and visual side of things.

286
00:17:06,600 --> 00:17:11,100
They're not surprising given Nvidia's background.

287
00:17:11,100 --> 00:17:13,760
So why is this interesting to marketers?

288
00:17:13,760 --> 00:17:16,840
I mean, most of this is about content production, right?

289
00:17:16,840 --> 00:17:22,840
So this is about streamlining content creation and very much focused on enhancing and making

290
00:17:22,840 --> 00:17:25,560
more efficient the kind of creative workflows.

291
00:17:25,560 --> 00:17:30,300
So if that's your space, I think there's lots of interesting things going on there.

292
00:17:30,300 --> 00:17:34,120
What they are saying though, and I think what we need to think about is this is just further

293
00:17:34,120 --> 00:17:37,200
accelerating AI adoption, right?

294
00:17:37,200 --> 00:17:42,400
There's new ways to use AI, new tools are going to be coming out, new products that

295
00:17:42,400 --> 00:17:46,300
have got more AI baked into them.

296
00:17:46,300 --> 00:17:50,720
If you have a product that isn't using AI tech, if you've got a digital product and

297
00:17:50,720 --> 00:17:56,320
you're not looking at how this can be used to improve or enhance your digital offer,

298
00:17:56,320 --> 00:17:57,560
you will get left behind.

299
00:17:57,560 --> 00:17:59,600
You will be a laggard.

300
00:17:59,600 --> 00:18:05,040
So yeah, I think this is just, you know, new set of tools that developers can access and

301
00:18:05,040 --> 00:18:06,480
that's really the headline.

302
00:18:06,480 --> 00:18:11,440
Go check it out, particularly if you're making a product in the creative industry space because

303
00:18:11,440 --> 00:18:18,960
Nvidia is looking to help you make better use of your product and tech.

304
00:18:18,960 --> 00:18:20,080
Absolutely.

305
00:18:20,080 --> 00:18:26,200
So many things to dig in there, work for Google and just see if any of those particular cloud

306
00:18:26,200 --> 00:18:32,880
services are going to be relevant for perhaps the niches that you work in.

307
00:18:32,880 --> 00:18:38,960
Mark mentioned BioLimo, which is obviously a very life science specific and something

308
00:18:38,960 --> 00:18:43,240
I will be having a look at and a play with over the next week and maybe we can return

309
00:18:43,240 --> 00:18:44,480
to next week as you said.

310
00:18:44,480 --> 00:18:47,640
But yes, go and have a play, go and have a look.

311
00:18:47,640 --> 00:18:48,640
Right.

312
00:18:48,640 --> 00:18:49,680
Let's talk about Bard.

313
00:18:49,680 --> 00:18:53,320
For long, because honestly it doesn't warrant it.

314
00:18:53,320 --> 00:18:56,800
But Google went, oh no, let's get involved quick.

315
00:18:56,800 --> 00:18:58,960
We can do this chat bot thing too.

316
00:18:58,960 --> 00:19:02,960
And so they announced Bard, which is their chat cheapy T equivalent, and they made it

317
00:19:02,960 --> 00:19:07,960
available by a waiting list and actually quite a few people I know got approved quite quickly.

318
00:19:07,960 --> 00:19:09,240
Not me.

319
00:19:09,240 --> 00:19:13,600
If you're listening, Google, so I can't say that I've played with it directly, but a lot

320
00:19:13,600 --> 00:19:17,480
of the feedback that I've had from people I know who are playing with it is yeah.

321
00:19:17,480 --> 00:19:21,520
I mean, this is kind of old news at this point.

322
00:19:21,520 --> 00:19:26,280
I think some of its advantages is that it will retrieve info from the web a little bit

323
00:19:26,280 --> 00:19:29,360
like Bing, not just its training data.

324
00:19:29,360 --> 00:19:32,480
So three chat GPT plugins.

325
00:19:32,480 --> 00:19:35,280
So 48 hours ago, that was kind of cool.

326
00:19:35,280 --> 00:19:37,800
Less cool now.

327
00:19:37,800 --> 00:19:43,360
But I, yeah, I think in essence here, what's important for marketers is the Google is clearly

328
00:19:43,360 --> 00:19:49,160
trying to play a bit of catch up here, maybe not doing a great job of it, to be honest.

329
00:19:49,160 --> 00:19:52,520
It's probably still worth getting yourself on the wait list and having a play if you

330
00:19:52,520 --> 00:19:57,080
get approved so that you can see how you might use it for content generation, brainstorming

331
00:19:57,080 --> 00:19:58,920
ideas, that type of thing.

332
00:19:58,920 --> 00:20:02,600
Anything that you might use chat GPT for really, just to see if you can get better results

333
00:20:02,600 --> 00:20:05,260
or different types of results with Bard.

334
00:20:05,260 --> 00:20:12,280
It's also a general business use of certainly Google apps users maybe gives us a slight

335
00:20:12,280 --> 00:20:17,440
insight into what we can expect when Google launches its suite of tools for workspace.

336
00:20:17,440 --> 00:20:21,960
And so what it might be like to have a chat system across Gmail, Doc Sheets, slides, that

337
00:20:21,960 --> 00:20:22,960
type of thing.

338
00:20:22,960 --> 00:20:29,400
But in essence, all you really need to know is Google is slowly but surely trying to edge

339
00:20:29,400 --> 00:20:33,720
its way into this story, but still continually feeling like it's being left behind, to be

340
00:20:33,720 --> 00:20:34,880
honest.

341
00:20:34,880 --> 00:20:36,040
They are playing catch up.

342
00:20:36,040 --> 00:20:39,840
I got accepted on after I got onto the chat wait list.

343
00:20:39,840 --> 00:20:45,200
It was about 48 hours, I think from what I understand they are prioritising Google One

344
00:20:45,200 --> 00:20:46,200
subscribers.

345
00:20:46,200 --> 00:20:50,520
So if you have a Google One subscription, I think you get faster access.

346
00:20:50,520 --> 00:20:51,520
I do.

347
00:20:51,520 --> 00:20:52,520
Oh, okay.

348
00:20:52,520 --> 00:20:53,520
Interesting.

349
00:20:53,520 --> 00:20:54,520
Yeah.

350
00:20:54,520 --> 00:20:55,520
Well, mine was pretty quick.

351
00:20:55,520 --> 00:20:58,120
It came through, like I say, within 48 hours.

352
00:20:58,120 --> 00:21:04,920
It's been interesting to have Bard open as well as Bing chat open as well as chat GPT

353
00:21:04,920 --> 00:21:08,900
and just running the prompt side by side.

354
00:21:08,900 --> 00:21:19,120
At this moment in time, I don't think there's much in it between Bard and the Bing chat.

355
00:21:19,120 --> 00:21:25,240
I still think GPT-4, even though Bing has GPT-4 integrated into it, I think when you

356
00:21:25,240 --> 00:21:27,840
use it in chat GPT, it just feels different.

357
00:21:27,840 --> 00:21:28,840
That's my...

358
00:21:28,840 --> 00:21:30,600
I think Bing's been shackled.

359
00:21:30,600 --> 00:21:35,080
When I ask it certain questions, it's like deliberately not trying that hard and being

360
00:21:35,080 --> 00:21:39,800
super safe with what it tells me to the point where it's like inane what it tells me at

361
00:21:39,800 --> 00:21:40,800
times.

362
00:21:40,800 --> 00:21:43,600
Yeah, I think that's probably a fair assessment of it.

363
00:21:43,600 --> 00:21:47,960
My immediate impressions, and this was kind of knee-jerk, I'd been playing with it for

364
00:21:47,960 --> 00:21:49,800
ten or 15 minutes.

365
00:21:49,800 --> 00:21:58,040
Bard is kind of chat GPT, GPT-3.5 with a few links.

366
00:21:58,040 --> 00:22:01,920
And at this point, given how fast things have accelerated, that is old.

367
00:22:01,920 --> 00:22:05,200
I mean, we're talking, that's November.

368
00:22:05,200 --> 00:22:09,040
Yeah, that's pretty real at this point.

369
00:22:09,040 --> 00:22:12,160
It feels like a different world and they are playing catch-up heavily.

370
00:22:12,160 --> 00:22:18,080
So yeah, it's worth more and more conversation, I would say.

371
00:22:18,080 --> 00:22:19,640
Agreed, Amundo.

372
00:22:19,640 --> 00:22:20,640
Let's get into creative.

373
00:22:20,640 --> 00:22:26,880
A lot of people will be listening to this, marketers where creative strategy and execution

374
00:22:26,880 --> 00:22:28,520
will be a part of their jobs.

375
00:22:28,520 --> 00:22:33,720
There's a raft of creative driven launches this week.

376
00:22:33,720 --> 00:22:37,160
Why don't you tell us a little bit about what Canva had to say?

377
00:22:37,160 --> 00:22:43,320
Yeah, so Canva had been teasing these 10 new gifts and they announced them at their Canva

378
00:22:43,320 --> 00:22:49,360
Create, their annual conference showcasing all of their big product updates.

379
00:22:49,360 --> 00:22:54,840
And in total, 10, well, 10 new gifts as the title suggested it would be.

380
00:22:54,840 --> 00:23:01,360
Magic design, magic draw, translate, magic eraser and edit, beat sync, presentation generator,

381
00:23:01,360 --> 00:23:07,160
magic write text to image or an evolution of, plus create an animation, a new brand

382
00:23:07,160 --> 00:23:11,280
hub, bonus features, including all sorts of kind of professional things that you get in

383
00:23:11,280 --> 00:23:17,840
more Adobe quality products, stuff that people have been frustrated about with Canva for

384
00:23:17,840 --> 00:23:18,840
ages.

385
00:23:18,840 --> 00:23:21,580
But in terms of the AI element, there were some really interesting ones.

386
00:23:21,580 --> 00:23:28,520
One in particular that stood out for me, and I think this just goes to show how fast the

387
00:23:28,520 --> 00:23:29,920
scene is moving.

388
00:23:29,920 --> 00:23:35,880
So when Google launched their Pixel smartphone last year, I don't know if you remember the

389
00:23:35,880 --> 00:23:42,280
adverts for the Google Pixel 7, I think it was, the big feature that they were really

390
00:23:42,280 --> 00:23:44,120
pushing on was the magic eraser.

391
00:23:44,120 --> 00:23:48,080
You take a photo and you can scrub out an item.

392
00:23:48,080 --> 00:23:49,600
And that was like their lead thing.

393
00:23:49,600 --> 00:23:52,640
Like you want to do this thing, you can do this right on the phone.

394
00:23:52,640 --> 00:23:53,640
It's amazing.

395
00:23:53,640 --> 00:24:00,000
Well, Canva just announced that feature as like one of 10 or 15 different features and

396
00:24:00,000 --> 00:24:02,480
you can do it like, there you go.

397
00:24:02,480 --> 00:24:03,480
Yeah.

398
00:24:03,480 --> 00:24:07,040
The main reason to buy that smartphone, which is the best part of a thousand pound, you

399
00:24:07,040 --> 00:24:10,360
could just do on Canva for like a 10 pound a month subscription and they'll probably

400
00:24:10,360 --> 00:24:13,480
have it on the app as well and anyone can do it.

401
00:24:13,480 --> 00:24:14,480
I mean, that's amazing.

402
00:24:14,480 --> 00:24:16,980
They've also integrated magic edit.

403
00:24:16,980 --> 00:24:25,920
So if you've done in-painting with Darley on OpenAI, where you can color in a bit of

404
00:24:25,920 --> 00:24:30,640
the, or kind of mask off a bit of an image and then describe what you want to see there

405
00:24:30,640 --> 00:24:36,120
instead, that's now baked into Canva powered by stable diffusion.

406
00:24:36,120 --> 00:24:38,480
So they're clearly working with stability AI there.

407
00:24:38,480 --> 00:24:39,480
It looks cool.

408
00:24:39,480 --> 00:24:42,320
It looks really interesting.

409
00:24:42,320 --> 00:24:44,400
They've expanded the magic write option.

410
00:24:44,400 --> 00:24:53,680
So last year they introduced AI text generation, which was specifically for the Canva docs.

411
00:24:53,680 --> 00:24:57,920
So like Google docs, the word processing element of Canva.

412
00:24:57,920 --> 00:25:00,680
They had it in the long form editor.

413
00:25:00,680 --> 00:25:03,000
They've now made that available across the suite.

414
00:25:03,000 --> 00:25:08,040
So anywhere there's a text box, you can now use AI text generation.

415
00:25:08,040 --> 00:25:16,640
And cool feature that I really did like is that they've now integrated 19 languages with

416
00:25:16,640 --> 00:25:20,480
this as well as an in-situ translation tool.

417
00:25:20,480 --> 00:25:25,440
So you can just, whatever you've written, translate it on the fly and it translates

418
00:25:25,440 --> 00:25:27,040
it in your design there and then.

419
00:25:27,040 --> 00:25:32,000
So for marketers, particularly those that are working globally, they have different

420
00:25:32,000 --> 00:25:37,260
audiences, you can within one suite, without having to jump and change to different places,

421
00:25:37,260 --> 00:25:40,320
you can do it all in one streamlined workflow.

422
00:25:40,320 --> 00:25:41,320
Yeah.

423
00:25:41,320 --> 00:25:45,440
So, I mean, there's too much to go into in terms of the announcement.

424
00:25:45,440 --> 00:25:49,800
There's a whole raft of things I haven't really touched on.

425
00:25:49,800 --> 00:25:55,040
The beat sync, which is a video editing thing where you drop your audio and it will automatically

426
00:25:55,040 --> 00:26:00,960
cut your video in line with the beat of the music and video and synchronizing.

427
00:26:00,960 --> 00:26:04,400
And I can't even really get the words out to describe it properly, Paul.

428
00:26:04,400 --> 00:26:05,400
It's just so much.

429
00:26:05,400 --> 00:26:12,320
It makes me want to put a backing track on this video and then just have it cut between

430
00:26:12,320 --> 00:26:19,600
you and me insanely fast to some crazy techno beat.

431
00:26:19,600 --> 00:26:22,800
Not because it has any real value.

432
00:26:22,800 --> 00:26:27,640
The poor listeners are at this point of thinking, surely there can't be any more announcements.

433
00:26:27,640 --> 00:26:28,920
Oh, but there are.

434
00:26:28,920 --> 00:26:36,160
But before we get to them, you were thinking, I think, through some of the sort of second

435
00:26:36,160 --> 00:26:42,360
order effects that you get from this explosion of AI powering creative tools.

436
00:26:42,360 --> 00:26:47,720
We'll look at Adobe Firefly and Runway ML in a moment, but what are your thoughts here?

437
00:26:47,720 --> 00:26:48,720
Yeah.

438
00:26:48,720 --> 00:26:54,440
So I think it's great that everyone has these new capabilities, but actually when you think

439
00:26:54,440 --> 00:26:58,120
about workflow, we're going to touch on this later on in the podcast in terms of the impact

440
00:26:58,120 --> 00:26:59,520
on the labor market.

441
00:26:59,520 --> 00:27:05,080
This just enables people that didn't have skills in a particular area before to suddenly

442
00:27:05,080 --> 00:27:06,080
have those skills.

443
00:27:06,080 --> 00:27:12,800
A really nice example that they show in the kind of presentation is they've got this animation

444
00:27:12,800 --> 00:27:13,800
tool.

445
00:27:13,800 --> 00:27:19,680
Now, previously you'd need an animator or someone that knows their way around any of

446
00:27:19,680 --> 00:27:22,960
the kind of video production tools to be able to add that on.

447
00:27:22,960 --> 00:27:28,440
But now your marketing exec who has had no video production training can create that

448
00:27:28,440 --> 00:27:30,920
little graphic that adds that little animation.

449
00:27:30,920 --> 00:27:34,120
So it reduces the need for those specialized design skills.

450
00:27:34,120 --> 00:27:35,720
You don't need to outsource as much.

451
00:27:35,720 --> 00:27:37,520
One person can do much more.

452
00:27:37,520 --> 00:27:40,160
I also like the fact that the translations are baked in.

453
00:27:40,160 --> 00:27:45,560
So again, just wider reach opens up more possibilities for your marketing.

454
00:27:45,560 --> 00:27:47,600
You don't have to...

455
00:27:47,600 --> 00:27:50,600
There's still always concerns around if that translation accurate.

456
00:27:50,600 --> 00:27:53,360
It's always worth getting a sense check.

457
00:27:53,360 --> 00:27:57,400
In terms of workflow, all of that piece is just sped up.

458
00:27:57,400 --> 00:28:04,560
I think there's much more efficiency gains from having everything just in one suite of

459
00:28:04,560 --> 00:28:05,600
tools.

460
00:28:05,600 --> 00:28:10,240
We didn't touch on some of the branding pieces as well.

461
00:28:10,240 --> 00:28:12,520
There's a really nice feature that they use.

462
00:28:12,520 --> 00:28:16,760
I don't know about you, Paul, but you've probably undergone during your time in business or

463
00:28:16,760 --> 00:28:22,360
helped to manage a couple of rebrands or redesigns of things.

464
00:28:22,360 --> 00:28:24,240
Sometimes they're big, sometimes they're small.

465
00:28:24,240 --> 00:28:30,000
Sometimes a small one might be that a logo has been updated, but that logo sits on hundreds

466
00:28:30,000 --> 00:28:31,560
of documents.

467
00:28:31,560 --> 00:28:35,960
So now you've got to go through and update hundreds of documents.

468
00:28:35,960 --> 00:28:39,840
These might be templates, brochures, whatever.

469
00:28:39,840 --> 00:28:44,120
Canva has done it so that that's just a one click process now and it rolls it out across

470
00:28:44,120 --> 00:28:45,120
all of your templates.

471
00:28:45,120 --> 00:28:47,120
There you go, done.

472
00:28:47,120 --> 00:28:49,960
Yeah, that's insane.

473
00:28:49,960 --> 00:28:54,820
In leveraging the power that comes with AI to do that in a much more intelligent way

474
00:28:54,820 --> 00:29:00,040
than would have been done before, you'd have to even one click tools from a year ago would

475
00:29:00,040 --> 00:29:04,600
get it right for a few templates and completely mess up how that should have been applied

476
00:29:04,600 --> 00:29:08,000
to a bunch of them that you then have to do manually.

477
00:29:08,000 --> 00:29:13,680
So the context awareness of a lot of these tools has improved a lot, hasn't it?

478
00:29:13,680 --> 00:29:16,160
I think you're right and it's interesting.

479
00:29:16,160 --> 00:29:20,280
Let's talk about Adobe Firefly briefly because it's kind of similar.

480
00:29:20,280 --> 00:29:25,840
So there's a waitlist opened up for this where it will create images from a prompt, so very

481
00:29:25,840 --> 00:29:29,960
dually, journey-esque, replace sections of an image.

482
00:29:29,960 --> 00:29:33,640
We're now starting to talk about these things like they're so passe, we're like, well, of

483
00:29:33,640 --> 00:29:34,640
course, isn't that easy?

484
00:29:34,640 --> 00:29:38,160
And it's like, if someone said a year ago, you'd even be able to do one or two of these

485
00:29:38,160 --> 00:29:41,480
things, you'd be like, oh, wow, that sounds amazing.

486
00:29:41,480 --> 00:29:44,440
Some crazy wizardry is involved in that.

487
00:29:44,440 --> 00:29:49,680
But no, these are just a bunch of bullet points now that come as expected.

488
00:29:49,680 --> 00:29:53,680
One of the things I liked about this, and I haven't got access to the BT yet, so I haven't

489
00:29:53,680 --> 00:29:57,400
played with it, but some of the demo videos were, of course, the cherry pick things that

490
00:29:57,400 --> 00:30:01,720
look amazing rather than things that don't work out very well, was generating effects

491
00:30:01,720 --> 00:30:04,120
and designs to lay on top of text.

492
00:30:04,120 --> 00:30:08,680
So in essence, I think in one of the videos, they're going to write the word yum, but they

493
00:30:08,680 --> 00:30:13,400
want it to look like a baked cake with chocolate dripping off of it.

494
00:30:13,400 --> 00:30:15,400
And it looks really good.

495
00:30:15,400 --> 00:30:18,720
And then if you don't quite like the way that the chocolate's dripping, it gives you multiple

496
00:30:18,720 --> 00:30:20,920
options of how the chocolate might drip off those letters.

497
00:30:20,920 --> 00:30:25,120
And you can basically write anything you like in a whole host of whatever creative styles

498
00:30:25,120 --> 00:30:27,960
you can imagine, which is kind of really cool.

499
00:30:27,960 --> 00:30:31,920
Then there's the generation of custom vectors, brushes, and textures.

500
00:30:31,920 --> 00:30:37,840
So in essence, I think of the example is there's like a dog, like a poodle maybe, and they

501
00:30:37,840 --> 00:30:42,880
select the dog's hair because it's got quite a big fluffy head and turn it into a brush,

502
00:30:42,880 --> 00:30:46,320
like a context-aware brush that they can basically expand the hairstyle out and give it this

503
00:30:46,320 --> 00:30:52,120
crazy massive hairstyle that looks exactly like it should because this smart AI driven

504
00:30:52,120 --> 00:30:56,560
brush doesn't just mimic that piece of hair it was trained on.

505
00:30:56,560 --> 00:31:00,680
It understands how that hair would look over on the other side of the head or if it was

506
00:31:00,680 --> 00:31:04,160
made to be much bigger, which is really, really cool.

507
00:31:04,160 --> 00:31:08,560
And then there's a bunch of stuff in there as well around editing video and 3D models

508
00:31:08,560 --> 00:31:11,620
and all those other things.

509
00:31:11,620 --> 00:31:12,620
So it's really cool.

510
00:31:12,620 --> 00:31:17,040
I recommend going to the firefly.adaby.com website to have a look at some of those things

511
00:31:17,040 --> 00:31:19,320
because they are amazing.

512
00:31:19,320 --> 00:31:21,360
Recoloring vectors, just imagine, right?

513
00:31:21,360 --> 00:31:27,360
You get a load of icons off of adaby stock, I would assume, but maybe shuttlestock or

514
00:31:27,360 --> 00:31:28,360
whatever.

515
00:31:28,360 --> 00:31:32,720
And then you just type in the box what your brand colors are and then it intelligently

516
00:31:32,720 --> 00:31:38,280
does that, not just to simple icons, but complex icons or even full vector images would be

517
00:31:38,280 --> 00:31:40,000
really quite cool and fun.

518
00:31:40,000 --> 00:31:45,240
Again, my only real fear with some of these things is because I completely take your point

519
00:31:45,240 --> 00:31:50,960
and I think it will have a big impact, Martin, that people without design training can now

520
00:31:50,960 --> 00:31:53,240
get really good design results.

521
00:31:53,240 --> 00:31:57,720
But I think, A, it will funnel us all into a very templated world.

522
00:31:57,720 --> 00:32:01,720
Oh, there's another word with where the word looks like it's got chocolate dripping off

523
00:32:01,720 --> 00:32:07,000
it, but they all look a bit like that because now everybody just uses Firefly to do them,

524
00:32:07,000 --> 00:32:08,000
right?

525
00:32:08,000 --> 00:32:12,520
That's just the text example, but you could definitely imagine the types of banner images

526
00:32:12,520 --> 00:32:16,440
and social images and brochure templates and stuff that Canvas got and everybody's just

527
00:32:16,440 --> 00:32:20,760
looking the same because they do it for speed instead of differentiation.

528
00:32:20,760 --> 00:32:24,160
Let's not forget, creative branding is about being memorable.

529
00:32:24,160 --> 00:32:26,160
It's about standing out.

530
00:32:26,160 --> 00:32:29,760
So taking these shortcuts is probably going to make sense in some cases, but in other

531
00:32:29,760 --> 00:32:33,200
cases, it's going to actually achieve the opposite of what you're trying to do.

532
00:32:33,200 --> 00:32:39,280
So I think in those cases, being able to really go through a proper creative concepting process

533
00:32:39,280 --> 00:32:45,360
with creative directors coming up with a novel idea that is the essence of your brand, doing

534
00:32:45,360 --> 00:32:48,960
the research to make sure it's really going to differentiate you from your competitors

535
00:32:48,960 --> 00:32:53,160
and it's aligned with who your brand is and the benefits you bring and what the personality

536
00:32:53,160 --> 00:32:55,680
of your brand is and all that stuff.

537
00:32:55,680 --> 00:33:02,080
And then executing that with a level of finesse and customizability that these tools can't

538
00:33:02,080 --> 00:33:04,000
give very easily, right?

539
00:33:04,000 --> 00:33:09,440
You can't go in to these tools at the moment and make minor tweaks to the output.

540
00:33:09,440 --> 00:33:13,600
I think that will come very quickly with Firefly, if I'm honest, because of course, it can give

541
00:33:13,600 --> 00:33:18,120
you your creations, port them straight into Illustrator or Photoshop and you can tweak

542
00:33:18,120 --> 00:33:20,800
away to your heart's content.

543
00:33:20,800 --> 00:33:26,160
So I do think there is still a space for that level of design excellence.

544
00:33:26,160 --> 00:33:30,840
It's just the quick and easy things are now even quicker and even easier and probably

545
00:33:30,840 --> 00:33:32,880
maybe designer in the loop, not so required.

546
00:33:32,880 --> 00:33:36,240
I think it's all about the application, isn't it?

547
00:33:36,240 --> 00:33:37,800
Where are you using this?

548
00:33:37,800 --> 00:33:42,960
If you're putting together an internal presentation for the board, you don't need to spend hours

549
00:33:42,960 --> 00:33:44,720
crafting that presentation anymore.

550
00:33:44,720 --> 00:33:51,760
You can stick your data in and Canva or Microsoft Copilot or Google's AI workspace will just

551
00:33:51,760 --> 00:33:53,440
create the deck for you.

552
00:33:53,440 --> 00:33:57,640
So that job that would have taken you a couple of hours to do is now you've got the data,

553
00:33:57,640 --> 00:34:00,840
stick it in, tell it the story you want to tell, it's done.

554
00:34:00,840 --> 00:34:05,160
You don't need it to be brand creative, workshopped.

555
00:34:05,160 --> 00:34:07,480
No, it's just, it's functional.

556
00:34:07,480 --> 00:34:14,960
So functional design like from Earth, but it doesn't need to be memorable, it needs

557
00:34:14,960 --> 00:34:18,720
to be work that will be made much easier I think.

558
00:34:18,720 --> 00:34:23,720
Often within deliberately restricted brand guidelines to keep stuff on brand.

559
00:34:23,720 --> 00:34:27,600
It's almost like probably the tempted approach in the use case you just described is actually

560
00:34:27,600 --> 00:34:29,960
useful for keeping things restricted.

561
00:34:29,960 --> 00:34:31,040
I think you're right.

562
00:34:31,040 --> 00:34:36,840
I think they'll find their place and it will be human in the loop where humans bring unique

563
00:34:36,840 --> 00:34:41,960
value and then AI in the loop where speed efficiency and automation and to be honest,

564
00:34:41,960 --> 00:34:45,760
operating within a set of rules is valuable.

565
00:34:45,760 --> 00:34:46,760
Cool beans.

566
00:34:46,760 --> 00:34:50,760
Let's do one last design driven one.

567
00:34:50,760 --> 00:34:52,760
Like this is a throwaway piece of news.

568
00:34:52,760 --> 00:34:55,080
It's like, oh my goodness.

569
00:34:55,080 --> 00:34:56,240
We've talked so much about design.

570
00:34:56,240 --> 00:34:57,760
Let's just do this one quickly.

571
00:34:57,760 --> 00:35:04,200
RunwayML teases the launch of text to video with a really brilliant teaser video of what

572
00:35:04,200 --> 00:35:05,200
you're going to be able to do.

573
00:35:05,200 --> 00:35:09,640
So in essence, for those that don't know, RunwayML is like a suite of video and image

574
00:35:09,640 --> 00:35:13,360
driven AI tools that can do all the standard things we've been describing, but some also

575
00:35:13,360 --> 00:35:20,080
pretty cool stuff in between like some of the object removal that Mark was talking about

576
00:35:20,080 --> 00:35:21,400
for Campbell, but with video.

577
00:35:21,400 --> 00:35:26,960
So you do a video and there's somebody in it you don't want, you can take them out.

578
00:35:26,960 --> 00:35:30,320
But what they did a couple of weeks ago, maybe a month ago as they launched Gen 1, which

579
00:35:30,320 --> 00:35:35,080
was you would do a video on your phone and then you could stylize it and really make

580
00:35:35,080 --> 00:35:36,080
it look pretty neat.

581
00:35:36,080 --> 00:35:40,440
And I think we talked about the Indiana Jones example in a previous episode with the launch

582
00:35:40,440 --> 00:35:42,120
of Gen 2.

583
00:35:42,120 --> 00:35:45,480
Now you can do something similar, but you don't have to give it a video to make a stylized

584
00:35:45,480 --> 00:35:46,480
video.

585
00:35:46,480 --> 00:35:49,040
You can do video to video like that.

586
00:35:49,040 --> 00:35:50,040
You can do image to video.

587
00:35:50,040 --> 00:35:53,720
So you give it a stylized image and say, make a video of this, but you could do text to

588
00:35:53,720 --> 00:35:57,540
video, text to video.

589
00:35:57,540 --> 00:36:04,040
So we're talking earlier about giving non creatively designed people, videographer, non

590
00:36:04,040 --> 00:36:07,640
videographers, non designers access to tools where they can get reasonable results where

591
00:36:07,640 --> 00:36:10,320
they happen to have all that skill or training.

592
00:36:10,320 --> 00:36:14,520
And what Gen 2 is suggesting is going to be able to do is give us the opportunity to be

593
00:36:14,520 --> 00:36:18,600
able to produce interesting video content without actually having to capture any video,

594
00:36:18,600 --> 00:36:21,520
which is right what we want.

595
00:36:21,520 --> 00:36:24,200
The examples on the announcement were brilliant.

596
00:36:24,200 --> 00:36:30,760
I love the one of, there was someone had some books and they just did like a very short

597
00:36:30,760 --> 00:36:36,040
clip of the camera kind of zooming in, getting a closeup of these books that are stacked

598
00:36:36,040 --> 00:36:37,220
side by side.

599
00:36:37,220 --> 00:36:42,360
And then there was a prompt, which is something like turn this into an aerial shot of the

600
00:36:42,360 --> 00:36:44,040
New York city skyline or something.

601
00:36:44,040 --> 00:36:45,520
It was along those lines.

602
00:36:45,520 --> 00:36:51,720
And that video was then magically transformed into a skyline of the modern metropolitan

603
00:36:51,720 --> 00:36:53,000
city with skyscraper.

604
00:36:53,000 --> 00:36:54,280
Yeah, that was cool.

605
00:36:54,280 --> 00:36:55,280
That was cool.

606
00:36:55,280 --> 00:36:59,920
There was an example of show a surfer catching a wave as well, which was a text based prompt

607
00:36:59,920 --> 00:37:02,240
that produced a pretty interesting video.

608
00:37:02,240 --> 00:37:08,960
I think one thing that I still know is that the videos are far from real production quality,

609
00:37:08,960 --> 00:37:09,960
right?

610
00:37:09,960 --> 00:37:11,520
Like they look a bit rough.

611
00:37:11,520 --> 00:37:16,880
They remind me of my, I would love to say worst outputs from Dually, but probably my

612
00:37:16,880 --> 00:37:21,160
average and near best outputs Dually where it like looks kind of cool, but there's enough

613
00:37:21,160 --> 00:37:26,480
clunky bits in the image that like don't look that good.

614
00:37:26,480 --> 00:37:28,000
So that's where this technology is at.

615
00:37:28,000 --> 00:37:30,400
So I think it's more of an example of what's coming.

616
00:37:30,400 --> 00:37:35,040
This is like being a production ready tool, you know, never need to use a videographer

617
00:37:35,040 --> 00:37:37,080
again, just explain to runway what you want.

618
00:37:37,080 --> 00:37:42,280
And you get this thing that you could, that Nike could create a commercial off the back

619
00:37:42,280 --> 00:37:43,920
of.

620
00:37:43,920 --> 00:37:46,800
I think it's interesting as well.

621
00:37:46,800 --> 00:37:50,720
Was it, was it, was mid journey five just released as well?

622
00:37:50,720 --> 00:37:52,720
Yeah, mid journey five.

623
00:37:52,720 --> 00:37:55,200
The image quality there looks really good.

624
00:37:55,200 --> 00:37:56,200
Portrait looks stunning.

625
00:37:56,200 --> 00:38:00,120
And look at they're incredible.

626
00:38:00,120 --> 00:38:04,880
And there's been some examples going around, I think is it Obama and Angela Merkel like

627
00:38:04,880 --> 00:38:09,720
having on the beach, having tea on the beach, looking pretty photorealistic.

628
00:38:09,720 --> 00:38:14,160
So so we're getting there on the quality of image front.

629
00:38:14,160 --> 00:38:15,520
And do you see the Trump one?

630
00:38:15,520 --> 00:38:20,560
Sorry to the Trump getting arrested one, which this week of all weeks has caused a bit of

631
00:38:20,560 --> 00:38:25,560
a, you know, talk about fake news and deep fakes like that one's actually got people

632
00:38:25,560 --> 00:38:26,960
a bit riled up.

633
00:38:26,960 --> 00:38:28,420
And thinking it's real, right?

634
00:38:28,420 --> 00:38:32,000
Because these were drifting into deep fake territory with how good the quality is getting.

635
00:38:32,000 --> 00:38:39,600
If you look at how to be firefly, the text examples I described, they are production

636
00:38:39,600 --> 00:38:40,600
ready.

637
00:38:40,600 --> 00:38:44,800
Like you could drop those on a big glossy billboard on London underground and people

638
00:38:44,800 --> 00:38:47,960
would think they're professional designer put them together.

639
00:38:47,960 --> 00:38:52,480
If you contrast that with even as little as three to six months ago, people with eight

640
00:38:52,480 --> 00:38:55,800
fingers faces that look like ghouls.

641
00:38:55,800 --> 00:38:59,000
And the only real way that you could get a decent image is deliberately doing something

642
00:38:59,000 --> 00:39:03,040
impressionist like the best results I've got from Dually to still today.

643
00:39:03,040 --> 00:39:08,780
When I say give me a painting in the style of Monet, because it's impressionist by default.

644
00:39:08,780 --> 00:39:12,520
And so the fact that it can't, it kind of doesn't really understand what the shapes

645
00:39:12,520 --> 00:39:14,440
are doesn't really matter.

646
00:39:14,440 --> 00:39:20,320
But yes, with some of these examples that we've seen from from mid journey five, we're

647
00:39:20,320 --> 00:39:23,720
getting there, that's kind of scary in itself as well really.

648
00:39:23,720 --> 00:39:27,600
Do you know, I would say mine, I'm a bit of a tech nerd.

649
00:39:27,600 --> 00:39:30,160
So when all this stuff is launching, I'm like, oh, so excited.

650
00:39:30,160 --> 00:39:31,160
Oh, so excited.

651
00:39:31,160 --> 00:39:32,160
Oh, so excited.

652
00:39:32,160 --> 00:39:38,000
This week is the first week where I've been like, a little bit afraid.

653
00:39:38,000 --> 00:39:44,920
Because this train is starting to like, like, it's flying, it's going to come off the tracks

654
00:39:44,920 --> 00:39:47,040
if it keeps going at this speed, right?

655
00:39:47,040 --> 00:39:53,080
And either, I try and process it in my brain, either all of these companies have been working

656
00:39:53,080 --> 00:39:57,740
on these things for months, which I think is highly likely and that everything needing

657
00:39:57,740 --> 00:40:02,560
to be launched at once because the opportunity is going to be gone is driving so many launches.

658
00:40:02,560 --> 00:40:04,480
I think that's a big part of it.

659
00:40:04,480 --> 00:40:10,360
But you've got to wonder how much having access to these tools a year ahead of where we are

660
00:40:10,360 --> 00:40:16,600
now because of course, in the labs are going to be ahead of us is enabling faster development.

661
00:40:16,600 --> 00:40:19,080
And we're really hitting that exponential curve, right?

662
00:40:19,080 --> 00:40:24,440
We talked earlier about chat, GPT being able to basically do half the lifting of building

663
00:40:24,440 --> 00:40:27,160
a plugin for you by just explaining what you want.

664
00:40:27,160 --> 00:40:31,680
Well, if you work internally at OpenAI and you've got access to these tools, how much

665
00:40:31,680 --> 00:40:36,480
quicker can you develop and ship updates to your software because you've got all these

666
00:40:36,480 --> 00:40:37,480
assistants.

667
00:40:37,480 --> 00:40:46,440
So we, I tried to imagine what the exponential curve of AI would look like the last two weeks,

668
00:40:46,440 --> 00:40:49,560
like what's next week going to bring?

669
00:40:49,560 --> 00:40:53,480
Where are we going to be by the end of 2023 if we're on that curve?

670
00:40:53,480 --> 00:40:59,640
Just look back at the tweet from, I forget, was it Sam Oldman or was it Greg Brockman?

671
00:40:59,640 --> 00:41:05,560
One of the leadership team at OpenAI when chat GPT came out around Christmas time, I

672
00:41:05,560 --> 00:41:11,280
might have even said it in the first podcast, they said, we will look back in a year's time,

673
00:41:11,280 --> 00:41:18,880
we will look back at the AI that we have today, I think it was basically quaint.

674
00:41:18,880 --> 00:41:24,840
And already, we're three months out from that tweet and already we're dismissing Bard as

675
00:41:24,840 --> 00:41:26,440
if it was a complete non-entity.

676
00:41:26,440 --> 00:41:27,440
Right.

677
00:41:27,440 --> 00:41:28,440
Yeah.

678
00:41:28,440 --> 00:41:33,560
And all these technical wizardry and all these design programs as being table stakes.

679
00:41:33,560 --> 00:41:34,560
Yeah.

680
00:41:34,560 --> 00:41:35,560
Yeah.

681
00:41:35,560 --> 00:41:36,560
It's insane.

682
00:41:36,560 --> 00:41:37,560
It's insane.

683
00:41:37,560 --> 00:41:42,520
And probably, this is the last set of stories for today, a good point to actually switch

684
00:41:42,520 --> 00:41:47,440
into speaking to some of these research papers that have come out, Martin, that I know you've

685
00:41:47,440 --> 00:41:49,560
been diving deeper into.

686
00:41:49,560 --> 00:41:50,560
What do you want to do first?

687
00:41:50,560 --> 00:41:54,040
Do you want to do OpenAI's impact on the labor market or what do you want to look at the

688
00:41:54,040 --> 00:41:55,040
AI?

689
00:41:55,040 --> 00:42:00,480
I think the AGI one is a good one because this one concludes with some of the labor

690
00:42:00,480 --> 00:42:01,480
market stuff.

691
00:42:01,480 --> 00:42:02,480
Okay.

692
00:42:02,480 --> 00:42:08,160
So the Microsoft research team obviously had access to GPT-4 for a while while they were

693
00:42:08,160 --> 00:42:10,000
integrating it with Bing.

694
00:42:10,000 --> 00:42:14,160
They've had access to it and they said, actually, this is probably a good time for us to write

695
00:42:14,160 --> 00:42:19,600
a paper on this because this is quite an incredible piece of technology.

696
00:42:19,600 --> 00:42:25,960
So they've written a report saying that this is the spark, well, these are the first sparks

697
00:42:25,960 --> 00:42:29,020
of artificial general intelligence.

698
00:42:29,020 --> 00:42:36,640
And then proceeded to write a paper that explains why they think that is the case and why this

699
00:42:36,640 --> 00:42:41,920
is beyond just a simple artificial intelligence, which can do a narrow range of tasks that

700
00:42:41,920 --> 00:42:42,920
it's trained to do.

701
00:42:42,920 --> 00:42:46,240
But actually, we're heading much more into something that has an understanding of the

702
00:42:46,240 --> 00:42:51,120
world beyond what it's kind of trained to do.

703
00:42:51,120 --> 00:42:55,840
And there are some fascinating little insights in it.

704
00:42:55,840 --> 00:43:01,960
So just as a, in fact, today we got as a side note to that.

705
00:43:01,960 --> 00:43:11,840
The first article today that I've read confirming GPT-4 as a one trillion parameter model.

706
00:43:11,840 --> 00:43:14,560
So I saw this article shared.

707
00:43:14,560 --> 00:43:17,400
I forget the source now.

708
00:43:17,400 --> 00:43:21,900
There was an article about Elon Musk's relationship with OpenAI, but within that is clearly a

709
00:43:21,900 --> 00:43:26,880
well-sourced article that mentions that GPT-4 is a one trillion parameter model.

710
00:43:26,880 --> 00:43:32,720
And I think this really shows if you just play around with GPT-4 and GPT-3.5, you see

711
00:43:32,720 --> 00:43:34,840
big differences very quickly.

712
00:43:34,840 --> 00:43:39,600
So going back to the researchers, they've laid out a series of different tests and experiments

713
00:43:39,600 --> 00:43:43,240
that they've done to demonstrate the capabilities.

714
00:43:43,240 --> 00:43:51,160
And one of the things that they're uncovering is that the text model has a very solid understanding

715
00:43:51,160 --> 00:43:54,080
of spatial reasoning.

716
00:43:54,080 --> 00:43:56,460
And so there's a couple of examples of that.

717
00:43:56,460 --> 00:44:03,880
One is that they asked it to draw a unicorn using TXE code, which is something I'm not

718
00:44:03,880 --> 00:44:08,480
familiar with, but it's a kind of diagramming code.

719
00:44:08,480 --> 00:44:09,720
And it did it.

720
00:44:09,720 --> 00:44:15,900
If you look at the GPT-3 version of it, it was just basically a hexagon with a square

721
00:44:15,900 --> 00:44:17,900
on the corner.

722
00:44:17,900 --> 00:44:21,440
The GPT-4 one does a pretty good example of it.

723
00:44:21,440 --> 00:44:22,440
Then they edited it.

724
00:44:22,440 --> 00:44:29,360
They took that graph, took away the horn, and then asked it to add the horn onto this

725
00:44:29,360 --> 00:44:32,160
and it mapped it on again.

726
00:44:32,160 --> 00:44:33,260
That wasn't a visual input.

727
00:44:33,260 --> 00:44:36,920
They just put the code in and said, add a horn for a unicorn onto this.

728
00:44:36,920 --> 00:44:40,040
And it was able to do that.

729
00:44:40,040 --> 00:44:43,200
There was a really interesting example with navigating a map.

730
00:44:43,200 --> 00:44:48,920
So using a text chat, so back and forth, they said, you are basically walking through a

731
00:44:48,920 --> 00:44:53,400
building, go straight ahead, now enter the room.

732
00:44:53,400 --> 00:44:58,000
There are four doors, turn right, you now enter another room.

733
00:44:58,000 --> 00:45:02,840
There are two doors, turn left, and so on and so forth.

734
00:45:02,840 --> 00:45:08,600
At the end of this chat, which saw it walk through a room taking, I think it was 11 or

735
00:45:08,600 --> 00:45:15,680
12 moves through this space, at the end of the chat, they asked it to summarize and describe

736
00:45:15,680 --> 00:45:17,600
the building that it was in.

737
00:45:17,600 --> 00:45:21,960
And it gave an example of where it was talking about, it started off in the living room,

738
00:45:21,960 --> 00:45:25,560
the living room leads onto the hallway and onto the kitchen.

739
00:45:25,560 --> 00:45:29,720
We went into the hallway, the hallway had this, this, this, and this, and it perfectly

740
00:45:29,720 --> 00:45:32,080
describes that room.

741
00:45:32,080 --> 00:45:42,080
And they asked it to plot a map using, again, using code, using some sort of mapping model,

742
00:45:42,080 --> 00:45:44,400
which I'm unfamiliar with again.

743
00:45:44,400 --> 00:45:47,160
And it wrote the code for this map.

744
00:45:47,160 --> 00:45:50,280
And then when you put the code in, it had perfectly plotted.

745
00:45:50,280 --> 00:45:51,840
So that's a different modality.

746
00:45:51,840 --> 00:45:57,920
It could take a text description and then turn that text description into a map.

747
00:45:57,920 --> 00:46:03,880
So it was a kind of, again, a different kind of modality, understood space and time.

748
00:46:03,880 --> 00:46:08,360
Separately to that, there was an interesting thread on Twitter where someone said, they

749
00:46:08,360 --> 00:46:12,400
asked it, there's a diamond in a thimble inside a cup.

750
00:46:12,400 --> 00:46:16,760
The cup, you take the cup into the bathroom, you take the cup out of the bathroom, put

751
00:46:16,760 --> 00:46:22,800
the cup on the bed, you turn the cup upside down, you take the cup, put it on the table.

752
00:46:22,800 --> 00:46:24,960
Where's the diamond?

753
00:46:24,960 --> 00:46:28,040
And GPT-4 says it's on the bed.

754
00:46:28,040 --> 00:46:30,040
GPT-3 is like, oh, it's on the table.

755
00:46:30,040 --> 00:46:34,840
So GPT-4 has much better reasoning of the world around it.

756
00:46:34,840 --> 00:46:39,040
A couple of other things, there was a Plato's dialogues.

757
00:46:39,040 --> 00:46:47,400
They got it to do, to explain the dangers of autoregressive language models.

758
00:46:47,400 --> 00:46:52,120
So basically explain the dangers of something like a text language model.

759
00:46:52,120 --> 00:46:53,120
Of itself.

760
00:46:53,120 --> 00:46:54,120
Exactly.

761
00:46:54,120 --> 00:47:00,280
And they did it with ChatGPT and GPT-4 and got it to do it in the dialogues format of

762
00:47:00,280 --> 00:47:02,400
Plato's dialogues.

763
00:47:02,400 --> 00:47:07,200
And then got it to critique its own chat out with everything.

764
00:47:07,200 --> 00:47:08,200
As a teacher.

765
00:47:08,200 --> 00:47:11,080
Like if you were the teacher and then it was like, gave feedback.

766
00:47:11,080 --> 00:47:15,200
Needed to improve, even though it produced it itself.

767
00:47:15,200 --> 00:47:16,680
It was kind of mad.

768
00:47:16,680 --> 00:47:17,680
Yeah.

769
00:47:17,680 --> 00:47:20,320
And then they did a similar thing with patient notes.

770
00:47:20,320 --> 00:47:26,360
So they did it in a medical setting, kind of described a situation and then got it to

771
00:47:26,360 --> 00:47:30,240
write some patient notes and then got it to critique the patient notes.

772
00:47:30,240 --> 00:47:36,480
And it, basically in the first patient notes, it had hallucinated something about BMI.

773
00:47:36,480 --> 00:47:41,120
But the only way you can get the BMI of a patient is by knowing the height of a weight.

774
00:47:41,120 --> 00:47:44,080
And one of those factors wasn't known in this text.

775
00:47:44,080 --> 00:47:49,360
So when it was critiquing the patient notes, it said, well, it shouldn't have given BMI

776
00:47:49,360 --> 00:47:52,920
because it didn't have all the relevant information.

777
00:47:52,920 --> 00:47:56,960
So really the subtle little thing like that.

778
00:47:56,960 --> 00:48:01,160
And there was an interesting example with sketch generation and possible applications

779
00:48:01,160 --> 00:48:02,720
for sketch generation.

780
00:48:02,720 --> 00:48:06,400
Again, using plotting, map plotting software.

781
00:48:06,400 --> 00:48:12,000
So it can't, it's not an image generator, but you can get it through using code and

782
00:48:12,000 --> 00:48:15,000
plotting graphs, that kind of language.

783
00:48:15,000 --> 00:48:16,880
You can get it to draw sketches.

784
00:48:16,880 --> 00:48:22,560
And it was asked to draw a sketch of a city building game, like SimCity or something like

785
00:48:22,560 --> 00:48:23,560
that.

786
00:48:23,560 --> 00:48:25,720
So draw a sketch of that kind of environment.

787
00:48:25,720 --> 00:48:31,280
So it plotted out that, and then they took that sketch, put it into stable diffusion

788
00:48:31,280 --> 00:48:34,360
where you can turn a sketch into image.

789
00:48:34,360 --> 00:48:40,520
And then they had an AI version of a Sim building, city building game, created off the back of

790
00:48:40,520 --> 00:48:43,160
a sketch.

791
00:48:43,160 --> 00:48:45,440
And it was brilliant.

792
00:48:45,440 --> 00:48:49,720
And it was much more detailed and realistic.

793
00:48:49,720 --> 00:48:54,320
The final output from stable diffusion was much more realistic than when it didn't have

794
00:48:54,320 --> 00:49:00,060
the sketch or if you just asked a text to image generator to create the model itself.

795
00:49:00,060 --> 00:49:05,680
So I've got a question, Mark, because this is just being trained on a load of information.

796
00:49:05,680 --> 00:49:08,520
These are, are these somewhat emerging capabilities?

797
00:49:08,520 --> 00:49:11,860
They're not being, it's not like they trained it and they went, you know what I mean, Ray,

798
00:49:11,860 --> 00:49:16,440
as if it could create a sketch by basically plotting things like a graph and then so we

799
00:49:16,440 --> 00:49:17,440
can give it to me.

800
00:49:17,440 --> 00:49:18,440
Like this isn't designed.

801
00:49:18,440 --> 00:49:21,240
These are all, a lot of these are emergent behaviors.

802
00:49:21,240 --> 00:49:22,240
Exactly.

803
00:49:22,240 --> 00:49:29,280
The things that the model has, it's created an understanding of the world by having read

804
00:49:29,280 --> 00:49:36,800
enough, a big enough corpus of text, which is basically everything ever, all of the time.

805
00:49:36,800 --> 00:49:41,080
And it's now understood all of that.

806
00:49:41,080 --> 00:49:42,920
This bar me the way that it can do it.

807
00:49:42,920 --> 00:49:48,040
It developed a complex 3D game in HTML and JavaScript where there were these defenders

808
00:49:48,040 --> 00:49:50,960
that kind of come out and block you as you try and move through.

809
00:49:50,960 --> 00:49:53,960
And it was able to, it was asked to create this game and it did it.

810
00:49:53,960 --> 00:50:00,160
If you tried to get GPT-3 to do that, GPT-3, the response to GPT-3 is basically, I'm a

811
00:50:00,160 --> 00:50:01,480
large language model.

812
00:50:01,480 --> 00:50:03,120
No, I'm not doing that.

813
00:50:03,120 --> 00:50:04,120
What do you think I am?

814
00:50:04,120 --> 00:50:05,120
Intelligent.

815
00:50:05,120 --> 00:50:07,280
And then GPT-4 is like, yeah, I'll do that.

816
00:50:07,280 --> 00:50:08,280
You know, I'll do it.

817
00:50:08,280 --> 00:50:12,280
Rub his hands together, I'd love a challenge mate.

818
00:50:12,280 --> 00:50:15,080
Last one before we move on, because we're going to run out of time, but tell us a bit

819
00:50:15,080 --> 00:50:18,200
about the stack in the book.

820
00:50:18,200 --> 00:50:20,120
Oh yes.

821
00:50:20,120 --> 00:50:27,640
So it was asked to stack a book, nine eggs, a laptop, a bottle and a nail on top of each

822
00:50:27,640 --> 00:50:28,640
other.

823
00:50:28,640 --> 00:50:37,240
And they got chat GPT to identify how it would do it and GPT-4 to perform that same task.

824
00:50:37,240 --> 00:50:42,880
And GPT-4 gives quite a reasonable answer where it says you would put the book down,

825
00:50:42,880 --> 00:50:48,360
you would put the eggs three by three grid spaced apart, then put the laptop on top of

826
00:50:48,360 --> 00:50:49,360
that.

827
00:50:49,360 --> 00:50:51,520
So it's load bearing and it spreads the way evenly.

828
00:50:51,520 --> 00:50:55,160
And then you put the bottle and nail on top of that.

829
00:50:55,160 --> 00:50:58,840
And then you say, you think, wow, the reasoning, the ability to understand physics of the new

830
00:50:58,840 --> 00:50:59,840
environment, brilliant.

831
00:50:59,840 --> 00:51:03,520
But actually it was the chat GPT response, which I thought was brilliant because the

832
00:51:03,520 --> 00:51:09,400
chat GPT said place the eggs on top of the nail, making sure they are balanced and not

833
00:51:09,400 --> 00:51:11,040
tilting to one side.

834
00:51:11,040 --> 00:51:13,600
Like you idiot.

835
00:51:13,600 --> 00:51:18,000
This is absolutely just borked his understanding of the world there.

836
00:51:18,000 --> 00:51:22,720
Nine eggs on top of a nail.

837
00:51:22,720 --> 00:51:29,120
It does make me wonder if now I just want to go and ask it a load of questions from

838
00:51:29,120 --> 00:51:30,600
Monty Python and the Holy Grail.

839
00:51:30,600 --> 00:51:33,960
Like I want to say, what is the airspeed velocity of an unladen swallow?

840
00:51:33,960 --> 00:51:39,760
And I want it to come back and say, is that an African swallow or European swallow?

841
00:51:39,760 --> 00:51:43,400
So I can say, I don't know that and go flying off into the bridge of despair.

842
00:51:43,400 --> 00:51:49,280
But yeah, it's kind of insane how its reasoning abilities are emerging, right?

843
00:51:49,280 --> 00:51:50,280
Yeah.

844
00:51:50,280 --> 00:51:54,320
And this leads them in the paper to look at the societal impact and it concludes with

845
00:51:54,320 --> 00:51:55,320
this.

846
00:51:55,320 --> 00:52:01,000
We actually have a section as well talking about being able to plug it into other tech.

847
00:52:01,000 --> 00:52:06,120
So giving it connections via APIs and at least there's a day before the plugin announcement.

848
00:52:06,120 --> 00:52:10,120
So I think if you look at the paper, you might have seen the plugin announcement coming out.

849
00:52:10,120 --> 00:52:11,120
I like that.

850
00:52:11,120 --> 00:52:12,120
Yeah.

851
00:52:12,120 --> 00:52:16,120
But yeah, the societal impact, they do talk about AI could displace human workers and

852
00:52:16,120 --> 00:52:23,640
reduce particularly the high skilled workers because when GPT-4 can do as well or better

853
00:52:23,640 --> 00:52:28,760
than most humans on professional exams, we saw this with the bar, scores in the top 10%

854
00:52:28,760 --> 00:52:36,240
of humans for the legal bar exam and it can start to diagnose diseases at a rate that's

855
00:52:36,240 --> 00:52:42,520
as good or better than the average doctor, the status of these jobs could be diminished.

856
00:52:42,520 --> 00:52:50,800
So that's an interesting thing for us to consider, which leads us to OpenAI's research paper

857
00:52:50,800 --> 00:52:52,760
on the impact on the labour market.

858
00:52:52,760 --> 00:53:02,840
So this came out on 17th of March and they basically looked at what GPT's generative

859
00:53:02,840 --> 00:53:07,720
free-terrain transformers will do to the labour market.

860
00:53:07,720 --> 00:53:12,480
They specifically looked at the US labour market and the headline figures are 80% of

861
00:53:12,480 --> 00:53:17,760
the workforce could have 10% of their tasks impacted by GPT.

862
00:53:17,760 --> 00:53:20,360
80%.

863
00:53:20,360 --> 00:53:24,720
Most people will probably see their jobs impacted at least a little bit.

864
00:53:24,720 --> 00:53:25,720
Yes.

865
00:53:25,720 --> 00:53:26,720
Yes.

866
00:53:26,720 --> 00:53:31,560
And 19% of workers may see 50% of tasks affected.

867
00:53:31,560 --> 00:53:37,920
So a fifth of people, half your jobs is being done by a computer basically.

868
00:53:37,920 --> 00:53:38,920
Yep.

869
00:53:38,920 --> 00:53:43,480
And the more interesting note from all of this is that it's the higher wage jobs that

870
00:53:43,480 --> 00:53:46,360
face the higher, greatest exposure.

871
00:53:46,360 --> 00:53:53,040
So legal, medical, scientific, you name it.

872
00:53:53,040 --> 00:53:55,080
Very interesting there.

873
00:53:55,080 --> 00:53:59,160
So it's going to have significant economic, social and policy implications, potentially

874
00:53:59,160 --> 00:54:05,600
widening inequality, one of the big things that they've identified there.

875
00:54:05,600 --> 00:54:09,280
So why does this matter?

876
00:54:09,280 --> 00:54:12,560
Well, I mean, it's hard to argue why doesn't it matter?

877
00:54:12,560 --> 00:54:19,400
This is going to change the way that we all work, the productivity that we get, but it's

878
00:54:19,400 --> 00:54:23,320
also going to displace jobs and it is going to cause great inequality.

879
00:54:23,320 --> 00:54:28,000
They did list in there actually as well, jobs that aren't going to be impacted.

880
00:54:28,000 --> 00:54:37,200
And unsurprisingly, it's things like builders and people that work on oil rigs and yeah,

881
00:54:37,200 --> 00:54:38,840
it's lots of manual work.

882
00:54:38,840 --> 00:54:43,480
They lay that out and say, we're always going to need those jobs.

883
00:54:43,480 --> 00:54:46,080
They're not going to be impacted.

884
00:54:46,080 --> 00:54:53,520
Well yes, I agree with that, except for you've got Elon Musk and Tesla's Optimus robot that

885
00:54:53,520 --> 00:54:54,520
they're working on.

886
00:54:54,520 --> 00:54:59,680
And then there was a big load of buzz around them figure and it's an AI driven robots.

887
00:54:59,680 --> 00:55:05,400
Can I just chime in on the Tesla hype cycle?

888
00:55:05,400 --> 00:55:14,960
There was a great video that I saw the other day of, it was from 2014 onwards, Elon Musk

889
00:55:14,960 --> 00:55:20,600
saying every year we are one year away from fully autonomous driving vehicles.

890
00:55:20,600 --> 00:55:27,520
He has made that statement for the best part of a decade and Tesla are still no further

891
00:55:27,520 --> 00:55:29,240
along than level two autonomy.

892
00:55:29,240 --> 00:55:31,720
In fact, another car company has got to level three.

893
00:55:31,720 --> 00:55:35,380
I think Mercedes is the first to get to level three autonomy.

894
00:55:35,380 --> 00:55:44,280
So I take everything that Elon Musk says with a absolute mountain of salt.

895
00:55:44,280 --> 00:55:48,520
Enough to give you the type of heart disease that you'd ideally want an AI's ability to

896
00:55:48,520 --> 00:55:52,240
detect nice and early if you're going to be eating that much salt.

897
00:55:52,240 --> 00:55:56,520
Now I take that point, Martin, we'll have to see how all that shakes out.

898
00:55:56,520 --> 00:56:00,920
Maybe some of his promises will actually start to see some delivery.

899
00:56:00,920 --> 00:56:01,920
Right.

900
00:56:01,920 --> 00:56:02,920
Wow.

901
00:56:02,920 --> 00:56:04,240
That's a lot to go through.

902
00:56:04,240 --> 00:56:06,320
We do do tools of the week and bits and pieces.

903
00:56:06,320 --> 00:56:08,440
So we're going to cram them on the end.

904
00:56:08,440 --> 00:56:12,160
If you're on a, if you're in the gym, do an extra 10 minutes.

905
00:56:12,160 --> 00:56:13,160
You're welcome.

906
00:56:13,160 --> 00:56:14,660
You're going to be even fitter.

907
00:56:14,660 --> 00:56:16,880
If you walk in the dog, the dog deserves it.

908
00:56:16,880 --> 00:56:19,480
A nice, nice extra bit of sniffing.

909
00:56:19,480 --> 00:56:21,040
So let's look at the tool of the week.

910
00:56:21,040 --> 00:56:22,720
We'll be really quick here.

911
00:56:22,720 --> 00:56:24,440
So we've been playing with a couple of tools.

912
00:56:24,440 --> 00:56:28,880
I've been playing with ask your PDF and chat shape this week.

913
00:56:28,880 --> 00:56:35,060
So in essence, these are chat GPT enabled chat bots that allow you to basically query

914
00:56:35,060 --> 00:56:37,600
information from web pages and PDF.

915
00:56:37,600 --> 00:56:40,040
So we'll ask you PDF.

916
00:56:40,040 --> 00:56:44,640
You basically upload a PDF, like an ebook or a peer review paper or something.

917
00:56:44,640 --> 00:56:48,400
And then it briefly summarizes it for you and suggests some questions you might ask

918
00:56:48,400 --> 00:56:51,520
of it, but you can ask whatever questions you want.

919
00:56:51,520 --> 00:56:56,200
And this includes being able to summarize the PDF and it's been working particularly

920
00:56:56,200 --> 00:56:57,200
well.

921
00:56:57,200 --> 00:56:58,280
I've been actually quite impressed with it.

922
00:56:58,280 --> 00:57:03,000
I've read it somewhere by Stratas eBooks and asked some questions about it.

923
00:57:03,000 --> 00:57:05,760
And I think the information was reasonable that it was surfacing.

924
00:57:05,760 --> 00:57:10,480
So that was kind of cool, but I've mostly been using it to summarize PDFs instead of

925
00:57:10,480 --> 00:57:13,080
reading the whole things at this point.

926
00:57:13,080 --> 00:57:16,700
So I think it'd be down to the marketers who are listening to think about what other creative

927
00:57:16,700 --> 00:57:18,200
things they might do with that.

928
00:57:18,200 --> 00:57:22,320
The other thing I was playing with is chat shape, which is a Chrome extension that at

929
00:57:22,320 --> 00:57:27,060
the moment reads data on a webpage and then allows you to query it with natural language

930
00:57:27,060 --> 00:57:29,640
through a chat interface.

931
00:57:29,640 --> 00:57:34,400
So in essence, you could go to Wikipedia and then you could just ask questions of the Wikipedia

932
00:57:34,400 --> 00:57:37,040
article without actually having to read it.

933
00:57:37,040 --> 00:57:41,800
I think this is interesting proof of principle, but I think when it's going to get really

934
00:57:41,800 --> 00:57:48,040
exciting is when it can access all of the content on a given website, which at the moment

935
00:57:48,040 --> 00:57:51,320
they don't offer, but they say is coming soon.

936
00:57:51,320 --> 00:57:55,880
So I quite want to get hold of it when that happens so that I can give it access to all

937
00:57:55,880 --> 00:58:00,560
of our resources section and blog section on our website and then allow people to ask

938
00:58:00,560 --> 00:58:05,800
questions about life science marketing, where the information will be gathered from our

939
00:58:05,800 --> 00:58:11,160
blog specifically and from all the different pieces of content that we've prepared.

940
00:58:11,160 --> 00:58:15,760
Which is one of the use cases that OpenAI have talked about integrating knowledge bases.

941
00:58:15,760 --> 00:58:20,600
They're going to be making that really, I think they're building their own tool for

942
00:58:20,600 --> 00:58:22,720
that as well actually.

943
00:58:22,720 --> 00:58:27,960
Which is what we'd be able to do because if you've got like an internal wiki or something

944
00:58:27,960 --> 00:58:31,480
like that, it should absolutely be able to do that as well.

945
00:58:31,480 --> 00:58:36,560
But I think your point is really interesting, because ultimately there's so many third party

946
00:58:36,560 --> 00:58:41,400
tools that are being developed that Copilot released by Microsoft, Google's doing what

947
00:58:41,400 --> 00:58:47,080
it does now, this plugin on top of ChatGPT, a lot of well-funded startups just went pop

948
00:58:47,080 --> 00:58:49,480
because it's like, oh crumbs.

949
00:58:49,480 --> 00:58:54,040
Now it turns out Microsoft's already doing that, well now you can do it in a weekend

950
00:58:54,040 --> 00:58:57,520
using the plugin ecosystem, the ChatGPT.

951
00:58:57,520 --> 00:59:02,360
So you mentioned Ask Your PDF, is that available for free?

952
00:59:02,360 --> 00:59:03,600
Can we sign up today?

953
00:59:03,600 --> 00:59:11,560
Yeah, so you can trial the chat shape and Ask Your PDF is free, or at least like the

954
00:59:11,560 --> 00:59:13,320
use level I've been using it for.

955
00:59:13,320 --> 00:59:16,160
I haven't had any pop-ups or anything telling me I've got to pay or anything now.

956
00:59:16,160 --> 00:59:20,040
I didn't have to register, you go to Ask Your PDF and it's just got a window where you can

957
00:59:20,040 --> 00:59:21,160
drop your PDF.

958
00:59:21,160 --> 00:59:25,760
I would imagine, I personally wouldn't put any sensitive commercial information into

959
00:59:25,760 --> 00:59:30,080
it that wasn't publicly available and could be found by other means, because who knows

960
00:59:30,080 --> 00:59:33,280
where that PDF is going.

961
00:59:33,280 --> 00:59:36,880
For obvious reasons, you're going to install the Chrome extension, you should know what

962
00:59:36,880 --> 00:59:40,680
you're giving that Chrome extension access to, which is basically everywhere you go on

963
00:59:40,680 --> 00:59:41,680
the web.

964
00:59:41,680 --> 00:59:44,720
But yes, I think they're worth having a play with.

965
00:59:44,720 --> 00:59:51,600
Right, last little bit, we're going to talk about this Ninja marketing application for

966
00:59:51,600 --> 00:59:52,600
AI.

967
00:59:52,600 --> 00:59:56,520
It's kind of a sales application, I guess really, but that was sent to me by a friend

968
00:59:56,520 --> 00:59:58,600
of mine and it's really kind of cool.

969
00:59:58,600 --> 01:00:01,960
Let me just, let me down there and just disappear.

970
01:00:01,960 --> 01:00:03,680
Let me just dig this up.

971
01:00:03,680 --> 01:00:05,080
Where's my link?

972
01:00:05,080 --> 01:00:07,600
There it is.

973
01:00:07,600 --> 01:00:12,480
So this comes from Spencer Scott on Twitter describing a strategy that was applied by

974
01:00:12,480 --> 01:00:14,880
his friend, Mohammed Ammar.

975
01:00:14,880 --> 01:00:18,480
So here's what, I'm just going to read it to you because I don't think I could do it

976
01:00:18,480 --> 01:00:21,120
justice trying to paraphrase it.

977
01:00:21,120 --> 01:00:25,800
So he says, my buddy Mohammed Ammar runs a software development shop.

978
01:00:25,800 --> 01:00:29,360
Like every business owner, he wants more customers and historically some of his best customers

979
01:00:29,360 --> 01:00:32,840
are Shopify owners that he builds custom apps for.

980
01:00:32,840 --> 01:00:34,800
This is where it gets crazy.

981
01:00:34,800 --> 01:00:41,840
So in essence, what Mohammed does is he pulls a list of a hundred thousand Shopify sites

982
01:00:41,840 --> 01:00:44,360
with name, email, website, metadata, et cetera.

983
01:00:44,360 --> 01:00:47,840
Maybe that's a bit too much, there you go.

984
01:00:47,840 --> 01:00:52,400
Normally he would send some half, this is not my words, half baked shitty email that

985
01:00:52,400 --> 01:00:55,360
would result in a 0.01% reply rate.

986
01:00:55,360 --> 01:00:58,840
Well instead of doing that, he built the most insane tool I've ever seen.

987
01:00:58,840 --> 01:01:03,320
He built a tool that takes all the data from Shopify store, trains GPT-4 on what the site

988
01:01:03,320 --> 01:01:09,400
does in sales and then has GPT-4 come up with an idea for an app to generate all revenue

989
01:01:09,400 --> 01:01:14,520
for the store and then pitch the app in a cold email.

990
01:01:14,520 --> 01:01:16,480
He says, Spencer says, I'm not making this up.

991
01:01:16,480 --> 01:01:21,400
The first email he showed me was GPT-4 pitching a bath bomb company.

992
01:01:21,400 --> 01:01:26,060
The idea of an app that allows users to build their own bath bombs on the site.

993
01:01:26,060 --> 01:01:27,060
The email was insane.

994
01:01:27,060 --> 01:01:33,520
It gets even crazy because the owner responds with an email saying, hi, I love this idea,

995
01:01:33,520 --> 01:01:37,600
but I'm not sure how it would work at scale with manufacturing.

996
01:01:37,600 --> 01:01:43,760
Of which Spencer's friend doesn't know about manufacturing, so he has GPT-4 respond knowledgeably

997
01:01:43,760 --> 01:01:48,000
saying I understand that you might have some concerns, but our app could look up your inventory

998
01:01:48,000 --> 01:01:51,720
you have in stock and only allow customers to build products you already have.

999
01:01:51,720 --> 01:01:56,120
So now GPT-4 is an expert in these challenges.

1000
01:01:56,120 --> 01:01:59,960
The person doing this doesn't even have to know or think about it.

1001
01:01:59,960 --> 01:02:05,160
The customer thought it was an incredible idea and wasn't ready to do it right now,

1002
01:02:05,160 --> 01:02:07,160
but really wanted to circle back to it.

1003
01:02:07,160 --> 01:02:15,200
Then there was another one that I have to tell you as well about for a socks company.

1004
01:02:15,200 --> 01:02:21,420
So it's an email that says, yo, some of this is blanked out for obvious reasons.

1005
01:02:21,420 --> 01:02:24,760
Can I just say I'm absolutely loving brand socks.

1006
01:02:24,760 --> 01:02:28,180
Your mission to support these people with every purchase is freaking awesome.

1007
01:02:28,180 --> 01:02:32,640
So I've got this amazeball idea for a Shopify app that will boost your sock game even more.

1008
01:02:32,640 --> 01:02:35,240
This is GPT-4 writing this.

1009
01:02:35,240 --> 01:02:39,800
Look at this, Socksess Stories, where your customers can share their coolest experiences

1010
01:02:39,800 --> 01:02:43,440
rocking their brand socks right on your store.

1011
01:02:43,440 --> 01:02:47,700
Think about it, happy customers, heartwarming stories, maybe even some selfies.

1012
01:02:47,700 --> 01:02:51,000
What better way to spread the love and ramp up sales?

1013
01:02:51,000 --> 01:02:55,880
Just thinking out loud here says GPT-4, which is not doing that.

1014
01:02:55,880 --> 01:02:59,560
But if you ever want to chat more about this or bounce ideas around, I'm all ears.

1015
01:02:59,560 --> 01:03:00,760
Keep doing what you're doing.

1016
01:03:00,760 --> 01:03:03,480
Your socks rock and so do you.

1017
01:03:03,480 --> 01:03:05,880
So get to reply.

1018
01:03:05,880 --> 01:03:07,200
Hi, person.

1019
01:03:07,200 --> 01:03:11,960
I'm not sure if this is another spam email that I typically get, but you had me at Socksess

1020
01:03:11,960 --> 01:03:13,240
Stories.

1021
01:03:13,240 --> 01:03:17,840
So I'd definitely be up for hearing more.

1022
01:03:17,840 --> 01:03:25,560
He sent this pitch to 3000 Shopify owners and is getting these incredible bits of feedback

1023
01:03:25,560 --> 01:03:31,160
because of how customized the pitch is, how interesting it is, and how creatively and

1024
01:03:31,160 --> 01:03:34,240
novel novelly the pitch is delivered.

1025
01:03:34,240 --> 01:03:39,520
All of it is being automated by GPT-4.

1026
01:03:39,520 --> 01:03:44,200
And the ad production will be automated by people who go, I like the sound of this.

1027
01:03:44,200 --> 01:03:45,200
Can you make it happen?

1028
01:03:45,200 --> 01:03:47,680
And then GPT-4 will make it happen.

1029
01:03:47,680 --> 01:03:51,000
And then he'll just be sat there going, this is printing money.

1030
01:03:51,000 --> 01:03:54,200
But I mean, it's insane, that story.

1031
01:03:54,200 --> 01:03:57,760
So I wanted to share that with everyone because I think as marketers, and we talk about a

1032
01:03:57,760 --> 01:04:05,000
lot of different things on this podcast Martin, there's still a massive gap that should be

1033
01:04:05,000 --> 01:04:08,640
filled by you, dear listeners, imagination.

1034
01:04:08,640 --> 01:04:10,800
What problems are you facing?

1035
01:04:10,800 --> 01:04:11,960
What do you need?

1036
01:04:11,960 --> 01:04:14,220
What do you wish that you could do?

1037
01:04:14,220 --> 01:04:20,840
Because if it's interrogating data systems, asking questions of things, getting summaries,

1038
01:04:20,840 --> 01:04:29,280
with the new plugin marketplace or platform from OpenAI, it's really going to come down

1039
01:04:29,280 --> 01:04:30,440
to what do you want to do?

1040
01:04:30,440 --> 01:04:32,140
What can you imagine that you might do?

1041
01:04:32,140 --> 01:04:35,400
What ninja use cases can you think of?

1042
01:04:35,400 --> 01:04:39,080
Because you're going to be able to potentially build things that are very, very customized

1043
01:04:39,080 --> 01:04:40,360
just to what you want to do.

1044
01:04:40,360 --> 01:04:43,080
And I found it's really exciting.

1045
01:04:43,080 --> 01:04:45,040
I agree.

1046
01:04:45,040 --> 01:04:46,220
It's a brave new world.

1047
01:04:46,220 --> 01:04:47,600
It is indeed.

1048
01:04:47,600 --> 01:04:48,600
It is indeed.

1049
01:04:48,600 --> 01:04:53,360
And with that, Martin, we'll probably give people's ears a rest and probably their brains

1050
01:04:53,360 --> 01:04:54,360
as well.

1051
01:04:54,360 --> 01:04:57,960
I hope you took a billion notes, everyone, for cool stuff you want to do with all those

1052
01:04:57,960 --> 01:04:58,960
tools.

1053
01:04:58,960 --> 01:05:04,420
And if you like this podcast, please do subscribe, share it with all your marketing friends because

1054
01:05:04,420 --> 01:05:06,640
maybe they would enjoy it as well.

1055
01:05:06,640 --> 01:05:08,440
We still got loads of downloads, which is amazing.

1056
01:05:08,440 --> 01:05:11,800
We're blown away by how people have been receiving this so far.

1057
01:05:11,800 --> 01:05:14,600
If you do have any feedback, we'd love to hear it.

1058
01:05:14,600 --> 01:05:16,400
What do you want to see on the podcast?

1059
01:05:16,400 --> 01:05:18,280
What topics do you want to focus on?

1060
01:05:18,280 --> 01:05:23,040
If you are yourself a ninja marketing or salesperson who's come up with an awesome interest in

1061
01:05:23,040 --> 01:05:26,960
application for AI and your own workflow, we'd love to get you on.

1062
01:05:26,960 --> 01:05:28,720
We'd love you to tell us about what you've done.

1063
01:05:28,720 --> 01:05:32,820
Or if you're not too keen to come on the podcast, just send us an email or a LinkedIn direct

1064
01:05:32,820 --> 01:05:36,520
message about what you've been doing and we'll get you featured and tell your story and help

1065
01:05:36,520 --> 01:05:39,560
inspire people for their own ideas.

1066
01:05:39,560 --> 01:05:41,480
So I think with that, we'll go into our weekend, Martin.

1067
01:05:41,480 --> 01:05:42,480
What do you think?

1068
01:05:42,480 --> 01:05:43,480
Yeah, it sounds good.

1069
01:05:43,480 --> 01:05:48,880
So this weekend, I'm off to Peterborough to watch the mighty Rams play Peterborough.

1070
01:05:48,880 --> 01:05:51,600
I haven't given my Derby County update for the week.

1071
01:05:51,600 --> 01:05:52,720
So I know that's important.

1072
01:05:52,720 --> 01:05:53,720
People want to hear it.

1073
01:05:53,720 --> 01:05:54,720
So yeah, that's fun.

1074
01:05:54,720 --> 01:06:00,760
But it might be a revolt actually that we didn't do a Derby County update at the beginning

1075
01:06:00,760 --> 01:06:03,320
because we claimed there was too much AI stuff going on.

1076
01:06:03,320 --> 01:06:06,680
But there's never going to be that much AI stuff that we don't want to know what Derby

1077
01:06:06,680 --> 01:06:08,520
County is up to.

1078
01:06:08,520 --> 01:06:13,240
I am looking at the sunshine through my window right now thinking I've got to get me some

1079
01:06:13,240 --> 01:06:14,240
of that.

1080
01:06:14,240 --> 01:06:19,640
So I think I might take the dog out for a little walkie and yeah, try and enjoy a little

1081
01:06:19,640 --> 01:06:22,280
bit of sunshine before the weekend gets here.

1082
01:06:22,280 --> 01:06:23,280
Well enjoy.

1083
01:06:23,280 --> 01:06:24,280
Cheers, buddy.

1084
01:06:24,280 --> 01:06:25,280
I'll speak to you soon.

1085
01:06:25,280 --> 01:06:26,280
Bye bye.

1086
01:06:26,280 --> 01:06:31,480
Thank you for listening to Artificially Intelligent Marketing.

1087
01:06:31,480 --> 01:06:37,540
To stay on top of the latest trends, tips and tools in the world of marketing AI, be

1088
01:06:37,540 --> 01:06:39,300
sure to subscribe.

1089
01:06:39,300 --> 01:06:43,360
We look forward to seeing you again next week.