1
00:00:00,000 --> 00:00:10,040
Welcome to Artificially Intelligent Marketing, a weekly podcast where we stay on top of the

2
00:00:10,040 --> 00:00:15,760
latest trends, tips and tools in the world of marketing AI, helping you get the best

3
00:00:15,760 --> 00:00:18,600
results from your marketing efforts.

4
00:00:18,600 --> 00:00:22,280
Now let's join our hosts, Paul Avery and Martin Broadhurst.

5
00:00:22,280 --> 00:00:25,760
Hello everyone, welcome to Artificially Intelligent Marketing.

6
00:00:25,760 --> 00:00:27,760
Lovely to have you back again with us.

7
00:00:27,760 --> 00:00:30,520
It is episode 15.

8
00:00:30,520 --> 00:00:31,520
How very exciting.

9
00:00:31,520 --> 00:00:32,520
How are you, Martin?

10
00:00:32,520 --> 00:00:34,600
What have you been up to?

11
00:00:34,600 --> 00:00:40,920
I've been putting together some new workshops that I'm delivering for businesses to help

12
00:00:40,920 --> 00:00:43,680
them get started on using chat GPT.

13
00:00:43,680 --> 00:00:48,000
Turns out there was a bit of demand for that and people wanted to know, how do I just get

14
00:00:48,000 --> 00:00:49,000
started?

15
00:00:49,000 --> 00:00:50,000
What can I do with this tool?

16
00:00:50,000 --> 00:00:56,120
A lot of chatter and as much as we might be thinking everybody is using this already,

17
00:00:56,120 --> 00:00:58,280
quite a lot of people just need a bit of a leg up.

18
00:00:58,280 --> 00:00:59,840
So yeah, that's me.

19
00:00:59,840 --> 00:01:00,840
How about you?

20
00:01:00,840 --> 00:01:01,840
What have you been up to?

21
00:01:01,840 --> 00:01:04,200
I have been out and about this last week.

22
00:01:04,200 --> 00:01:10,000
I was at SAMPS, which is Sales and Marketing Professionals in Science.

23
00:01:10,000 --> 00:01:16,040
It's a sort of networking community for those people, of which I am one.

24
00:01:16,040 --> 00:01:21,880
And they these days run really great events for sales and marketing people in the life

25
00:01:21,880 --> 00:01:22,880
sciences.

26
00:01:22,880 --> 00:01:27,360
I was at a meeting at their event in Glasgow this week and the topic was AI.

27
00:01:27,360 --> 00:01:31,760
And we were talking about all the different applications of chat GPT.

28
00:01:31,760 --> 00:01:35,240
We did some really great stuff and had a lot of fun with Mid Journey.

29
00:01:35,240 --> 00:01:36,240
I was very lucky.

30
00:01:36,240 --> 00:01:41,680
I met some incredibly smart people who themselves are doing some awesome things with AI.

31
00:01:41,680 --> 00:01:43,160
So I learned loads from that.

32
00:01:43,160 --> 00:01:48,680
But I also learned a real ton of from the really great questions that everybody was

33
00:01:48,680 --> 00:01:49,680
asking.

34
00:01:49,680 --> 00:01:54,800
I was lucky to be part of, sat on some lunchtime tables having some amazing conversations with

35
00:01:54,800 --> 00:01:55,800
people.

36
00:01:55,800 --> 00:01:59,280
Then when I was wandering around, I could just hear people having these fantastic conversations.

37
00:01:59,280 --> 00:02:06,000
And I think that is the critical piece here is we're still early on in the emergence of

38
00:02:06,000 --> 00:02:09,640
a lot of these tools for use in marketing.

39
00:02:09,640 --> 00:02:13,560
And if you think you're too late to the party, you're not.

40
00:02:13,560 --> 00:02:15,320
But you got to get in now, right?

41
00:02:15,320 --> 00:02:18,820
Like start understanding what's going on, start playing with the tools, start thinking

42
00:02:18,820 --> 00:02:21,880
about what you can do with them.

43
00:02:21,880 --> 00:02:23,880
There was a great guy at the event called Dan.

44
00:02:23,880 --> 00:02:25,520
I'm going to try and get him on the podcast.

45
00:02:25,520 --> 00:02:27,560
Actually, Martin is a bit of a ninja.

46
00:02:27,560 --> 00:02:30,880
And he said there are no real experts in this yet.

47
00:02:30,880 --> 00:02:32,760
And I think to a certain extent, he was right.

48
00:02:32,760 --> 00:02:36,880
I think there's many of us that have done a lot of deep diving as you and I have Martin

49
00:02:36,880 --> 00:02:41,160
and have perhaps been following AI for a couple of years.

50
00:02:41,160 --> 00:02:45,960
But everything's merging so quickly and things are changing so fast.

51
00:02:45,960 --> 00:02:49,880
It's important to just get in there and start playing with bits and pieces yourself.

52
00:02:49,880 --> 00:02:53,840
And obviously to subscribe to the Artificially Intelligent Marketing podcast so you can get

53
00:02:53,840 --> 00:02:55,920
a bit of a leg up and a bit of a head start.

54
00:02:55,920 --> 00:02:59,480
Well, I think that's an interesting way of putting it, isn't it?

55
00:02:59,480 --> 00:03:04,160
Anyone that's been doing it for a few years at the moment has a bit of a head start and

56
00:03:04,160 --> 00:03:05,400
that's it.

57
00:03:05,400 --> 00:03:11,320
But people that immerse themselves in this are going to be able to catch up pretty quickly

58
00:03:11,320 --> 00:03:13,000
tinkering around with these tools.

59
00:03:13,000 --> 00:03:15,640
You can learn a lot in a very short amount of time.

60
00:03:15,640 --> 00:03:19,600
Were there any big takeaways from the event that you think are worth sharing?

61
00:03:19,600 --> 00:03:26,200
I think the main one for me is there are still quite a few people who haven't really played

62
00:03:26,200 --> 00:03:27,200
with the tools that much.

63
00:03:27,200 --> 00:03:31,160
Although I must say, I thought the event somewhat self-selected for people who had an interest

64
00:03:31,160 --> 00:03:35,160
in this and have probably been playing a bit because I had some really cool use cases and

65
00:03:35,160 --> 00:03:38,560
I learned quite a lot from what people are doing.

66
00:03:38,560 --> 00:03:42,700
I think there's been a fair bit of dabbling with ChatGPT, but people are still just trying

67
00:03:42,700 --> 00:03:47,360
to figure out how to get the best out of it and what they can rely on it for.

68
00:03:47,360 --> 00:03:51,380
Especially in our industry, we showed a number of several examples.

69
00:03:51,380 --> 00:03:58,360
We have to do a lot of explaining complex scientific topics and justifying our conclusions

70
00:03:58,360 --> 00:04:00,000
with citations.

71
00:04:00,000 --> 00:04:06,560
ChatGPT is wondrous at just making up citations that sound absolutely on the nose, but they're

72
00:04:06,560 --> 00:04:07,800
not real.

73
00:04:07,800 --> 00:04:12,600
In many cases, we were showing live demos almost where that was happening and also live

74
00:04:12,600 --> 00:04:16,000
demos where it would combine citations.

75
00:04:16,000 --> 00:04:19,120
Maybe we'd get the title of the paper right, but the author in the year wrong.

76
00:04:19,120 --> 00:04:22,500
Of course, if you can't rely on it for that type of stuff, can you really do anything

77
00:04:22,500 --> 00:04:23,500
with it?

78
00:04:23,500 --> 00:04:25,440
I think that was a reasonable question.

79
00:04:25,440 --> 00:04:28,200
I think people are still trying to get their heads around that part.

80
00:04:28,200 --> 00:04:33,560
Then at the other end of the spectrum, I think the ninjas now are moving beyond getting half-decent

81
00:04:33,560 --> 00:04:36,640
at mid-journey and getting half-decent at ChatGPT.

82
00:04:36,640 --> 00:04:42,000
They're looking at tools like Zapier and other sort of like AutoGPT and they're starting

83
00:04:42,000 --> 00:04:48,640
to think, how can I chain actions together to get interesting outputs using Zapier?

84
00:04:48,640 --> 00:04:53,680
I know that, Martin, you've talked previously about your application for recording a bit

85
00:04:53,680 --> 00:04:59,000
of audio and then using Zapier to basically get that audio processed, transcribed through

86
00:04:59,000 --> 00:05:03,120
something like Whisper and then turned into something like a short blog post section or

87
00:05:03,120 --> 00:05:07,160
a social post using ChatGPT.

88
00:05:07,160 --> 00:05:11,600
I think that's where people who are on the front end of this curve in the marketing area

89
00:05:11,600 --> 00:05:15,760
at least, that's where they are right now, at least from what I could see.

90
00:05:15,760 --> 00:05:17,160
Yeah, interesting.

91
00:05:17,160 --> 00:05:19,760
Reflects what I think I'm seeing as well.

92
00:05:19,760 --> 00:05:26,440
Good to know that we're all in the same kind of boat figuring this stuff out together.

93
00:05:26,440 --> 00:05:28,880
We're all in it together.

94
00:05:28,880 --> 00:05:31,600
Talking about the things that we're seeing, here's some of the things that we're seeing

95
00:05:31,600 --> 00:05:32,600
this week.

96
00:05:32,600 --> 00:05:34,360
We're going to cover four main stories.

97
00:05:34,360 --> 00:05:40,400
We're going to talk about Boston Consulting Group reporting on 91% of CMOs seeing efficiency

98
00:05:40,400 --> 00:05:41,400
gains from AI.

99
00:05:41,400 --> 00:05:43,600
We're going to dive into that report.

100
00:05:43,600 --> 00:05:48,360
We're going to talk about some upgrades to ChatGPT's API, which for those of you who

101
00:05:48,360 --> 00:05:53,240
are playing with tools like Zapier could actually be important in terms of how much it costs

102
00:05:53,240 --> 00:05:55,520
you and what you can now do.

103
00:05:55,520 --> 00:05:59,440
We're going to look a little bit at some of the legal stuff around this, in particular,

104
00:05:59,440 --> 00:06:04,800
EU Parliament Act to vote through the EU AI Act.

105
00:06:04,800 --> 00:06:07,760
Crumbs don't say that after a cider.

106
00:06:07,760 --> 00:06:11,360
And then we're going to look at another report that said that not many people are using ChatGPT

107
00:06:11,360 --> 00:06:13,360
and that was a report from Morgan Stanley.

108
00:06:13,360 --> 00:06:15,640
So we're going to dive into those.

109
00:06:15,640 --> 00:06:19,880
We're going to do our short snippets in a minute and then we're going to focus on our tool

110
00:06:19,880 --> 00:06:22,000
of the week, which is Contea.

111
00:06:22,000 --> 00:06:26,200
And we're actually going to hear from one of Contea's founders in an interview this

112
00:06:26,200 --> 00:06:27,200
week.

113
00:06:27,200 --> 00:06:30,320
And that's how we're going to dive into our tool of the week.

114
00:06:30,320 --> 00:06:34,400
So with that, let's get cracking on the short snippets because there is a fair few of them

115
00:06:34,400 --> 00:06:36,480
to get through this week, isn't there Martin?

116
00:06:36,480 --> 00:06:44,600
So numero uno is Mercedes-Benz is partnering with Microsoft to integrate ChatGPT into its

117
00:06:44,600 --> 00:06:46,600
in-car voice assistant.

118
00:06:46,600 --> 00:06:50,960
So I didn't know this, but there is already like a sort of hey Mercedes type tool in a

119
00:06:50,960 --> 00:06:54,280
lot of Mercedes new cars.

120
00:06:54,280 --> 00:06:58,600
And now that they've got this integration with ChatGPT, you can have much more natural

121
00:06:58,600 --> 00:07:04,480
conversations with your car, certainly with your friends as well, if you want.

122
00:07:04,480 --> 00:07:09,200
But if they're not around with your car and it will be able to deal with contextual follow-up

123
00:07:09,200 --> 00:07:15,400
questions and really manage the natural language elements of that better stuff, quite interesting.

124
00:07:15,400 --> 00:07:25,980
Then we had stable diffusion generated QR codes have gone a bit viral this week using

125
00:07:25,980 --> 00:07:31,200
stable diffusion to create custom VR codes, which are basically also quite beautiful images.

126
00:07:31,200 --> 00:07:35,540
And this is a quite a big improvement on other attempts, certainly that I've seen to try

127
00:07:35,540 --> 00:07:40,320
and make QR codes look all branded or look more attractive.

128
00:07:40,320 --> 00:07:44,480
Because in essence, you have like a really gorgeous image, almost with like a QR code

129
00:07:44,480 --> 00:07:49,800
hidden in it that you can take a picture of and it will work as a QR code.

130
00:07:49,800 --> 00:07:53,360
My experience has been some of them work well and some of them break really badly, but that's

131
00:07:53,360 --> 00:07:54,840
what you see through our mind.

132
00:07:54,840 --> 00:07:55,840
Absolutely.

133
00:07:55,840 --> 00:07:56,840
Some of them work, some of them don't.

134
00:07:56,840 --> 00:08:03,640
But if they can get that working, that's really quite cool because the images look stunning.

135
00:08:03,640 --> 00:08:08,560
So it's a functional QR code embedded within them.

136
00:08:08,560 --> 00:08:13,240
If we can get that working consistently, I think whoever commercialises that fine-tuned

137
00:08:13,240 --> 00:08:16,560
stable diffusion model will do quite well for themselves.

138
00:08:16,560 --> 00:08:17,560
I agree.

139
00:08:17,560 --> 00:08:21,880
I bet there is a big old race going on to do just that.

140
00:08:21,880 --> 00:08:23,680
If you're interested in playing with this, you can.

141
00:08:23,680 --> 00:08:29,360
There are how-to's on the internet to get stable diffusion running locally off your

142
00:08:29,360 --> 00:08:33,840
machine, not the model, but your access to the model.

143
00:08:33,840 --> 00:08:38,800
And then you can get instructions on how to tailor your settings to be able to get half

144
00:08:38,800 --> 00:08:40,920
decent QR code type images out of it.

145
00:08:40,920 --> 00:08:45,800
But as Martin said, when we get a commercial version of this, all of those technical hoops

146
00:08:45,800 --> 00:08:49,880
will be removed and we'll all be able to pay $10 a month and crack out as many awesome

147
00:08:49,880 --> 00:08:51,800
looking QR codes as we like.

148
00:08:51,800 --> 00:08:55,080
So we're excited to see that when it comes around.

149
00:08:55,080 --> 00:09:00,200
Meta is making its open source model, Llama, that we've touched on before on the podcast,

150
00:09:00,200 --> 00:09:04,520
available for businesses to use and build on top of.

151
00:09:04,520 --> 00:09:09,120
So this is quite interesting because Llama is not quite as good as GPT-4, probably somewhere

152
00:09:09,120 --> 00:09:14,260
between 3.5 and 4, maybe closer to 3.5 to be honest.

153
00:09:14,260 --> 00:09:17,920
But it's interesting because A, it's going to provide businesses with another option

154
00:09:17,920 --> 00:09:20,240
to be able to build their tools on top of.

155
00:09:20,240 --> 00:09:24,320
But because it's open source, it probably gives businesses a bit more flex in terms

156
00:09:24,320 --> 00:09:29,040
of if they have the coding capabilities internally to really play with it.

157
00:09:29,040 --> 00:09:32,960
Plus there's lots of arguments floating around, some of which we've talked about on the podcast

158
00:09:32,960 --> 00:09:40,280
and with Martin in terms of the ability to see those tools develop really quickly because

159
00:09:40,280 --> 00:09:43,800
the open source community are playing with them, expanding their capabilities, making

160
00:09:43,800 --> 00:09:45,360
them better and all that stuff.

161
00:09:45,360 --> 00:09:46,720
That'd be an interesting one.

162
00:09:46,720 --> 00:09:55,320
With that one, do we know if you have to download the model and host it yourself or is it API

163
00:09:55,320 --> 00:09:56,320
access?

164
00:09:56,320 --> 00:10:00,360
Do we know anything about how these businesses are going to access it at this point?

165
00:10:00,360 --> 00:10:04,760
So the host of a well-researched podcast would absolutely be able to answer that question

166
00:10:04,760 --> 00:10:06,920
Martin, but I cannot.

167
00:10:06,920 --> 00:10:07,920
It is a great question.

168
00:10:07,920 --> 00:10:11,360
I will accept that answer.

169
00:10:11,360 --> 00:10:12,960
We go deep on the main stories.

170
00:10:12,960 --> 00:10:18,320
We don't go so deep on the short snippets, but if you're interested in this, folks, hit

171
00:10:18,320 --> 00:10:19,320
the Googlebot.

172
00:10:19,320 --> 00:10:24,880
I'm sure you're all barred or Jack GPT plus Bing plugin and I'm sure you can uncover that

173
00:10:24,880 --> 00:10:29,560
little bit of extra information, very important bit of information that Martin has just asked

174
00:10:29,560 --> 00:10:30,560
about.

175
00:10:30,560 --> 00:10:36,120
Then we have Accenture who has announced a $3 billion investment in AI over the next

176
00:10:36,120 --> 00:10:41,860
three years to enhance this team of AI pros so that they can provide AI-focused solutions

177
00:10:41,860 --> 00:10:43,360
for their clients.

178
00:10:43,360 --> 00:10:47,540
In other words, help companies improve their efficiency and effectiveness across their

179
00:10:47,540 --> 00:10:49,400
business using AI.

180
00:10:49,400 --> 00:10:50,600
That's all well and good.

181
00:10:50,600 --> 00:10:54,560
Personally, I'd probably just hire Martin because he's doing a lot of this stuff already.

182
00:10:54,560 --> 00:10:56,120
Martin, give Accenture a call.

183
00:10:56,120 --> 00:10:59,240
I think they need your help, pal.

184
00:10:59,240 --> 00:11:04,280
I'm not sure that their budget is vast enough for my services.

185
00:11:04,280 --> 00:11:06,640
They rates are going up by the moment.

186
00:11:06,640 --> 00:11:07,640
Absolutely.

187
00:11:07,640 --> 00:11:12,200
They're in their early doors, people, before he is at $3 billion a day.

188
00:11:12,200 --> 00:11:13,200
Yeah.

189
00:11:13,200 --> 00:11:15,400
I have an early bird access offer.

190
00:11:15,400 --> 00:11:16,400
A lifetime deal.

191
00:11:16,400 --> 00:11:17,400
Let's get him on AppSumo.

192
00:11:17,400 --> 00:11:18,400
He loves AppSumo.

193
00:11:18,400 --> 00:11:19,400
Yeah.

194
00:11:19,400 --> 00:11:25,840
It's $2.5 billion for a day.

195
00:11:25,840 --> 00:11:30,400
I saw that story and thought it was an interesting one from an AI-focused solutions guy.

196
00:11:30,400 --> 00:11:33,960
I posted about this on LinkedIn separately.

197
00:11:33,960 --> 00:11:39,320
I'm getting a lot of people speaking to me at the moment and saying, we want to do AI.

198
00:11:39,320 --> 00:11:44,320
I go, okay, well, to do what?

199
00:11:44,320 --> 00:11:45,320
What are we looking to achieve?

200
00:11:45,320 --> 00:11:47,600
Oh no, we just know that our business needs it.

201
00:11:47,600 --> 00:11:52,800
It's like, oh no, you've done that thing where we need AI.

202
00:11:52,800 --> 00:11:56,760
You've brought a solution looking for a problem.

203
00:11:56,760 --> 00:12:03,080
I think we just have to be mindful at any kind of scale business that we're not just

204
00:12:03,080 --> 00:12:06,800
going AI, AI, AI, shiny thing, shiny thing.

205
00:12:06,800 --> 00:12:09,760
No, not AI, AIO.

206
00:12:09,760 --> 00:12:10,760
AI, AIO.

207
00:12:10,760 --> 00:12:11,760
Isn't it all?

208
00:12:11,760 --> 00:12:15,160
I know that my two-year-old too long.

209
00:12:15,160 --> 00:12:19,600
I'd be willing to argue that your two-year-old is probably more socially dexterous than me

210
00:12:19,600 --> 00:12:23,800
and probably has a better sense of humor.

211
00:12:23,800 --> 00:12:25,320
You actually did a social post, didn't you?

212
00:12:25,320 --> 00:12:31,520
I saw this week on this as it's relating to have a business use case where you can see

213
00:12:31,520 --> 00:12:36,600
AI can provide a benefit rather than thinking, well, let's put AI in our business.

214
00:12:36,600 --> 00:12:42,040
Well, an example of that is someone came to me a few weeks ago.

215
00:12:42,040 --> 00:12:47,280
This was off the back of a conversation we'd had following a conference talk that I'd done

216
00:12:47,280 --> 00:12:53,040
and they said, we really want to introduce AI and sat down with them and they didn't

217
00:12:53,040 --> 00:12:54,040
need AI.

218
00:12:54,040 --> 00:12:58,080
They just needed to use the marketing automation tools that they already had and they needed

219
00:12:58,080 --> 00:13:02,520
to use them better and more effectively than they are doing.

220
00:13:02,520 --> 00:13:03,800
Didn't need AI.

221
00:13:03,800 --> 00:13:06,720
It needed some fairly simple automation.

222
00:13:06,720 --> 00:13:11,080
I think there's an education gap.

223
00:13:11,080 --> 00:13:12,080
I think so.

224
00:13:12,080 --> 00:13:13,080
I think you're right.

225
00:13:13,080 --> 00:13:17,040
I think there's probably a walk and then run aspect to it as well because of examples like

226
00:13:17,040 --> 00:13:18,040
you just gave.

227
00:13:18,040 --> 00:13:21,160
But right, let's crack on through these short snippets.

228
00:13:21,160 --> 00:13:28,880
The next one we have is GPT-4 outperforming humans in pitch deck effectiveness apparently.

229
00:13:28,880 --> 00:13:33,560
So Clarify Capital asked 500 investors and business owners to rate pitch decks created

230
00:13:33,560 --> 00:13:37,800
by humans on their own or with track GPT's help.

231
00:13:37,800 --> 00:13:43,680
And overall, the investors said that GPT generated pitch decks were on average about two times

232
00:13:43,680 --> 00:13:49,320
more convincing and that they were three times more likely to invest after reading a GPT-4

233
00:13:49,320 --> 00:13:53,720
pitch deck than a human one, which is interesting.

234
00:13:53,720 --> 00:13:59,200
My takeaway from there was GPT-4 augmentation is going to help you do a better job if you

235
00:13:59,200 --> 00:14:01,680
ask it the right questions and you use clever prompts.

236
00:14:01,680 --> 00:14:05,800
But I think the human in the loop, there's probably a little bit more critical than the

237
00:14:05,800 --> 00:14:08,120
report suggested.

238
00:14:08,120 --> 00:14:12,280
Martin is nodding for our listeners and that's because he likes to agree with me occasionally,

239
00:14:12,280 --> 00:14:14,240
but not that often.

240
00:14:14,240 --> 00:14:15,240
Right.

241
00:14:15,240 --> 00:14:23,640
So the next one is generative AI could add $4.4 trillion annually to the global economy,

242
00:14:23,640 --> 00:14:26,080
according to a McKinsey report.

243
00:14:26,080 --> 00:14:29,760
And according to, I think I found this in the Neuron, which is a great newsletter for

244
00:14:29,760 --> 00:14:36,560
those that love their AI, this is more than the annual GDP of Canada, Australia and Spain

245
00:14:36,560 --> 00:14:37,560
combined.

246
00:14:37,560 --> 00:14:45,200
So there's some large hope there that we're going to get some positive economic impact

247
00:14:45,200 --> 00:14:47,240
from using AI.

248
00:14:47,240 --> 00:14:50,900
Where exactly that sits would be a good question.

249
00:14:50,900 --> 00:14:55,000
Next little story is AI bringing John Lennon's voice back to life.

250
00:14:55,000 --> 00:14:59,480
So we've got a quote here from Paul McCartney.

251
00:14:59,480 --> 00:15:02,120
So this is not a someone in their bedroom thinking, I know what I'm going to do.

252
00:15:02,120 --> 00:15:06,120
I'm going to create a cool tune and we'll get John Lennon on it.

253
00:15:06,120 --> 00:15:09,480
It was actually because, and here's the quote, when we came to make what would be the last

254
00:15:09,480 --> 00:15:13,640
Beatles record, it was a demo that John had that we'd worked on and we'd just finished

255
00:15:13,640 --> 00:15:15,800
it up and it'll be released this year.

256
00:15:15,800 --> 00:15:20,160
We were able to take John's voice and get it pure through this AI so that then we could

257
00:15:20,160 --> 00:15:22,040
mix the record as you would normally do.

258
00:15:22,040 --> 00:15:25,840
And it's because they had this recording on like a really old cassette tape and it sounded

259
00:15:25,840 --> 00:15:28,380
awful and they were able to clean it up.

260
00:15:28,380 --> 00:15:32,240
So the inference here is it's not quite as sexy this story as it sounds.

261
00:15:32,240 --> 00:15:38,200
I think it's more of an AI audio cleanup story than a synthetic John Lennon story.

262
00:15:38,200 --> 00:15:41,940
But hey, that means we might get some more Beatles music soon, which is not something

263
00:15:41,940 --> 00:15:44,040
that perhaps we'd have ever thought we were going to get.

264
00:15:44,040 --> 00:15:45,680
Yeah, I think that's pretty cool.

265
00:15:45,680 --> 00:15:49,840
And if you think about all of the archive footage that exists across different mediums,

266
00:15:49,840 --> 00:15:55,160
whether it be movies, television, audio like this, there'll be loads of it that's just

267
00:15:55,160 --> 00:15:58,280
sat there waiting to be AI remastered.

268
00:15:58,280 --> 00:16:02,640
We've had digital remastering for ages now, but taking it to the next level.

269
00:16:02,640 --> 00:16:07,960
And yeah, think about all of those lost episodes of Doctor Who that all of the Doctor Who fanboys

270
00:16:07,960 --> 00:16:13,760
will be wanting to see in glorious AI enhanced technicolor.

271
00:16:13,760 --> 00:16:15,840
Oh, I never thought about that.

272
00:16:15,840 --> 00:16:21,800
We could use uncrop to turn Star Trek The Next Generation from 4-3 into widescreen.

273
00:16:21,800 --> 00:16:25,640
That is a neat use case.

274
00:16:25,640 --> 00:16:28,520
Could you imagine?

275
00:16:28,520 --> 00:16:30,080
I would love to see its interpretation.

276
00:16:30,080 --> 00:16:34,840
But yeah, that would be quite a good way of actually turning 4-3 content into 16-9, right?

277
00:16:34,840 --> 00:16:37,200
I never really thought about that.

278
00:16:37,200 --> 00:16:43,280
Moving swiftly on, but into a somewhat similar story, Meta has just open sourced the kings

279
00:16:43,280 --> 00:16:45,600
and queens of open source Meta.

280
00:16:45,600 --> 00:16:46,600
Thank you so much.

281
00:16:46,600 --> 00:16:50,680
Well, what they've open sourced this time is their AI powered music generator called

282
00:16:50,680 --> 00:16:56,320
MusicGen, which has been trained on over 20,000 hours of music and it can turn text descriptions

283
00:16:56,320 --> 00:17:01,360
into short-witted audio clips and it can even be guided by reference audio that you give

284
00:17:01,360 --> 00:17:02,360
it.

285
00:17:02,360 --> 00:17:05,720
Now we've talked previously about Google having done a fair bit of work in this space,

286
00:17:05,720 --> 00:17:11,160
but the key difference is Meta's tool is open source, so people can start playing with it,

287
00:17:11,160 --> 00:17:14,080
improving it, expanding it a la llama.

288
00:17:14,080 --> 00:17:15,800
Lama a la.

289
00:17:15,800 --> 00:17:18,000
Something like that.

290
00:17:18,000 --> 00:17:19,840
So that's pretty cool news as well.

291
00:17:19,840 --> 00:17:23,360
Then we've got LinkedIn introduces AI powered tools for marketers.

292
00:17:23,360 --> 00:17:27,940
So now you can get copy suggestions within your campaign manager when you're building

293
00:17:27,940 --> 00:17:33,560
out bits and pieces like headlines and copy for your ads and for your sponsored posts

294
00:17:33,560 --> 00:17:35,360
and those types of things.

295
00:17:35,360 --> 00:17:43,200
And then finally, we may be getting some more chat GPT new features because there's been

296
00:17:43,200 --> 00:17:47,860
a leaked screenshot which shows something called workspaces, which would allow users

297
00:17:47,860 --> 00:17:52,720
to create a profile for themselves that chat GPT can readily remember.

298
00:17:52,720 --> 00:17:59,320
And it also looks like it might hint at being able to upload files directly into chat GPT.

299
00:17:59,320 --> 00:18:04,080
I saw this was doing the rounds in some various discord groups and on Facebook groups and

300
00:18:04,080 --> 00:18:06,920
just in the AI enthusiast community.

301
00:18:06,920 --> 00:18:08,840
I couldn't track the source of it though.

302
00:18:08,840 --> 00:18:10,920
Have you any idea where it came from?

303
00:18:10,920 --> 00:18:12,360
I just saw it doing the rounds.

304
00:18:12,360 --> 00:18:14,960
Yeah, I've seen it floating like you've seen it floating.

305
00:18:14,960 --> 00:18:20,840
I mean, the screen grab itself looks pretty sort of real, but then of course that wouldn't

306
00:18:20,840 --> 00:18:23,840
be hard to knock up.

307
00:18:23,840 --> 00:18:26,760
It would seem a weird thing to try and fake.

308
00:18:26,760 --> 00:18:27,760
Like it's interesting.

309
00:18:27,760 --> 00:18:31,400
And I think if you use chat GPT, you might go, hmm, I can see that that might make a

310
00:18:31,400 --> 00:18:37,440
few things easier, but it's not like, you know, Donald Trump being wrestled to the ground

311
00:18:37,440 --> 00:18:39,480
by 20 police officers.

312
00:18:39,480 --> 00:18:42,640
You know, you can see why someone produced that fake image.

313
00:18:42,640 --> 00:18:43,640
But yeah, right.

314
00:18:43,640 --> 00:18:46,520
Let's get into our main stories.

315
00:18:46,520 --> 00:18:52,000
And Martin, first one's with you and it's Boston Consulting Group reporting 91% of global

316
00:18:52,000 --> 00:18:57,080
CMOC enhanced efficiency from generative AI for marketing.

317
00:18:57,080 --> 00:19:02,960
So what was it in this story that grabbed your eyeballs?

318
00:19:02,960 --> 00:19:08,280
I think it's interesting that we're starting to see the big consulting groups really make

319
00:19:08,280 --> 00:19:09,520
a lot of noise about this.

320
00:19:09,520 --> 00:19:16,240
We've got, I think, well, we've got several of them mentioned on this very episode because

321
00:19:16,240 --> 00:19:22,840
we've got McKinsey we've already mentioned, we've got BCG and actually later on another

322
00:19:22,840 --> 00:19:28,180
big player, maybe not a consulting group, but Morgan Stanley featuring as well.

323
00:19:28,180 --> 00:19:38,480
This one was interesting that it was 200 CMOs surveyed from across Europe, Asia and North

324
00:19:38,480 --> 00:19:39,480
America.

325
00:19:39,480 --> 00:19:47,920
91% said that they are seeing enhanced efficiency from generative AI and that it had positively

326
00:19:47,920 --> 00:19:48,920
impacted their work.

327
00:19:48,920 --> 00:19:51,920
That was 93%.

328
00:19:51,920 --> 00:19:57,120
90% have already adopted generative AI for marketing.

329
00:19:57,120 --> 00:20:00,000
And overall, there was some interest in takeouts.

330
00:20:00,000 --> 00:20:09,480
BCG, they said that generative AI will be key for CMOs to gain competitive advantage

331
00:20:09,480 --> 00:20:12,160
through agility, innovation and responsibility.

332
00:20:12,160 --> 00:20:20,820
In fact, there's a particular phrase they said, BCG research indicates that CMOs are

333
00:20:20,820 --> 00:20:27,600
overwhelmingly optimistic and confident about the power of generative AI to enhance productivity

334
00:20:27,600 --> 00:20:31,120
and create competitive advantage.

335
00:20:31,120 --> 00:20:38,440
So overall, CMOs are on the whole using this new technology.

336
00:20:38,440 --> 00:20:40,120
They're seeing efficiency gains.

337
00:20:40,120 --> 00:20:46,520
I think that reflects my experience of it and yours too, which is why we do this podcast

338
00:20:46,520 --> 00:20:50,840
because we're so excited by the potential of it.

339
00:20:50,840 --> 00:20:58,560
The report goes on to say that while optimism is high, so 91% seeing enhanced efficiency,

340
00:20:58,560 --> 00:21:07,760
the impact depends on how meaningfully and ethically the generative AI will be applied.

341
00:21:07,760 --> 00:21:14,280
So CMOs can't see this as a magic bullet to enhance human judgment and creativity.

342
00:21:14,280 --> 00:21:22,120
Again, I think reflecting what we say so frequently on this podcast, which is human in the loop,

343
00:21:22,120 --> 00:21:29,800
you can't just outsource this and expect that AI is going to get it right every time.

344
00:21:29,800 --> 00:21:32,440
So yeah, I thought there were just some interesting headlines there.

345
00:21:32,440 --> 00:21:38,920
The BCG did recommend a few recommendations in there, which could be summarized as start

346
00:21:38,920 --> 00:21:45,400
experimenting, which exactly reflects your experience of the conversations you were having

347
00:21:45,400 --> 00:21:48,080
with people at SAMPS.

348
00:21:48,080 --> 00:21:55,640
Seek transformational outcomes, establish enterprise-wide models, and implement responsible

349
00:21:55,640 --> 00:21:56,640
AI guidelines.

350
00:21:56,640 --> 00:21:59,720
That last point I think is critical here.

351
00:21:59,720 --> 00:22:05,420
At the moment, there are something of a kind of wild west.

352
00:22:05,420 --> 00:22:12,360
We see some companies banning people from using generative AI, like Samsung, for instance,

353
00:22:12,360 --> 00:22:16,920
was famously reported as banning people from using chat GPT.

354
00:22:16,920 --> 00:22:23,200
In fact, just before we started recording today, I saw a headline.

355
00:22:23,200 --> 00:22:24,800
I haven't delved into this at all.

356
00:22:24,800 --> 00:22:27,200
It was literally just a headline.

357
00:22:27,200 --> 00:22:32,680
Google are advising employees not to use BARD.

358
00:22:32,680 --> 00:22:35,600
For the same reason that...

359
00:22:35,600 --> 00:22:39,560
Yeah, I haven't looked into it, but I'm guessing it's for the...

360
00:22:39,560 --> 00:22:43,080
Well, that's a funny one to me because I'm like, well, surely you own the model and it's

361
00:22:43,080 --> 00:22:44,080
your staff.

362
00:22:44,080 --> 00:22:48,660
But anyway, we can dig into that one in next week's episode, I think.

363
00:22:48,660 --> 00:22:55,160
But back to the point in this recommendation, implement responsible AI guidelines.

364
00:22:55,160 --> 00:23:01,640
We've signposted it before, but the Marketing Artificial Intelligence Institute have published

365
00:23:01,640 --> 00:23:08,240
via Creative Commons, basically the responsible AI usage manifesto.

366
00:23:08,240 --> 00:23:12,920
They've published their version of it and they've said to people, look, take this as

367
00:23:12,920 --> 00:23:19,880
a template that you can play around with, you can edit, you can make your own.

368
00:23:19,880 --> 00:23:25,800
I would highly recommend that organizations do that because having some rules and some

369
00:23:25,800 --> 00:23:32,320
guardrails around what your organization will and will not do when it comes to using AI

370
00:23:32,320 --> 00:23:36,640
will be invaluable, particularly when all employees have got access to things like chat

371
00:23:36,640 --> 00:23:38,800
GPT, just at their fingertips.

372
00:23:38,800 --> 00:23:40,000
Yeah, I agree.

373
00:23:40,000 --> 00:23:45,600
I think my big fear with this, I put a post on the socials today about it, is that we

374
00:23:45,600 --> 00:23:51,560
don't just trigger an ever-growing avalanche of crap content because we can just produce

375
00:23:51,560 --> 00:23:56,400
things at massive scale without really thinking about them and we just churn, churn, churn,

376
00:23:56,400 --> 00:24:02,760
churn, churn because it's already getting to the point where if I need to see another

377
00:24:02,760 --> 00:24:10,860
blog post on the five ways to optimize the header of a landing page, I don't need any

378
00:24:10,860 --> 00:24:13,440
more of those in my life.

379
00:24:13,440 --> 00:24:17,360
We'll get to the point, I suspect, where a lot of these tools, I'd choose a marketing

380
00:24:17,360 --> 00:24:21,960
example there, but we'll be answering those types of practical questions perhaps in a

381
00:24:21,960 --> 00:24:24,320
lot of cases.

382
00:24:24,320 --> 00:24:30,880
I really hope people use this to lean into producing really good content at scale where

383
00:24:30,880 --> 00:24:34,360
there's still quite a bit of time and thinking that has to go into the actual planning of

384
00:24:34,360 --> 00:24:39,820
it and chat GPT and other tools just help with the execution of it once we've done it

385
00:24:39,820 --> 00:24:45,800
rather than the temptation to just churn, churn, churn, churn, churn because I think

386
00:24:45,800 --> 00:24:51,360
everyone's just going to be harder for any of us to get through with content if we over

387
00:24:51,360 --> 00:24:55,360
use it as marketers and just turn everyone off to the content we're producing.

388
00:24:55,360 --> 00:24:59,800
Yeah, and there's definite wins that in the near term organizations can have.

389
00:24:59,800 --> 00:25:08,000
Just this week I was speaking to an associate and they were telling me about an organization

390
00:25:08,000 --> 00:25:11,720
that they were working with, never done any content marketing in their life, literally

391
00:25:11,720 --> 00:25:16,880
overnight, I think it was 20 blogs appeared on their website after they'd been introduced

392
00:25:16,880 --> 00:25:25,440
to chat GPT and within a number of weeks, 25% of lift in traffic.

393
00:25:25,440 --> 00:25:29,160
There are near term wins and you can see when that incentive is there, people are going

394
00:25:29,160 --> 00:25:33,680
to do it, people are going to churn, churn, churn, but this is not a long term strategy.

395
00:25:33,680 --> 00:25:34,680
No, no.

396
00:25:34,680 --> 00:25:40,480
I think if you ask yourself the right questions, if you really understand your audience and

397
00:25:40,480 --> 00:25:44,400
you think about what information that they need that they can't easily find and chat

398
00:25:44,400 --> 00:25:49,920
GPT helps you formulate your ideas, package it into some coherent pros that then you edit

399
00:25:49,920 --> 00:25:56,000
and tweak so it's in your voice and doesn't carry through any errors that chat GPT's

400
00:25:56,000 --> 00:26:01,040
introduced or you put the extra time and effort into really expand on a point that you understand

401
00:26:01,040 --> 00:26:03,080
and chat GPT clearly didn't.

402
00:26:03,080 --> 00:26:04,320
I've got no problems with that.

403
00:26:04,320 --> 00:26:11,400
I just, yeah, marketers in my experience ruin everything through mass overuse and I think

404
00:26:11,400 --> 00:26:14,920
we got to be careful, but we shall see how that plays out.

405
00:26:14,920 --> 00:26:15,920
Right.

406
00:26:15,920 --> 00:26:21,000
Next story is big upgrades to chat GPT's API.

407
00:26:21,000 --> 00:26:30,520
So OpenAI have released updated versions of GPT 4 and GPT 3.5 turbo with new technical

408
00:26:30,520 --> 00:26:37,560
capabilities, so things like function calling, larger context windows and lower pricing.

409
00:26:37,560 --> 00:26:41,880
So function calling allows developers to describe functions to the models, which can then return

410
00:26:41,880 --> 00:26:43,720
JSON to call those functions.

411
00:26:43,720 --> 00:26:48,600
So this enables connecting the models to external tools and APIs to be much more reliable and

412
00:26:48,600 --> 00:26:50,800
easier to do.

413
00:26:50,800 --> 00:26:56,280
They were very excited about that and loads of developers are very, very excited about

414
00:26:56,280 --> 00:26:57,280
that.

415
00:26:57,280 --> 00:27:02,320
I saw one developer saying that people had been asking for a kind of watered down version

416
00:27:02,320 --> 00:27:06,040
of this and they've gone above and beyond the delivery.

417
00:27:06,040 --> 00:27:12,960
So yeah, expect some really cool innovations from app developers in this domain.

418
00:27:12,960 --> 00:27:13,960
Agreed.

419
00:27:13,960 --> 00:27:17,960
And one would imagine much faster development because they won't have to think about clever

420
00:27:17,960 --> 00:27:22,800
and complex ways to try and get their information in and out of chat GPT.

421
00:27:22,800 --> 00:27:25,760
It's going to actually be much easier to do now.

422
00:27:25,760 --> 00:27:29,880
Yeah, and I think actually from an end user's perspective, going back to a theme that we

423
00:27:29,880 --> 00:27:35,800
bang on about, we probably won't even notice when apps are using this now because they'll

424
00:27:35,800 --> 00:27:39,040
just be making all of these different API calls, connecting different tools and it'll

425
00:27:39,040 --> 00:27:40,680
just be happening in the background.

426
00:27:40,680 --> 00:27:41,680
Yeah.

427
00:27:41,680 --> 00:27:42,680
Yeah.

428
00:27:42,680 --> 00:27:47,280
So just expect the tools that you use every day to just extend in their functionality

429
00:27:47,280 --> 00:27:52,240
and you won't even notice that it's chat GPT doing its thing in the background.

430
00:27:52,240 --> 00:27:53,520
Yeah.

431
00:27:53,520 --> 00:28:00,080
So GPT 3.5 turbo is getting a greater context window.

432
00:28:00,080 --> 00:28:02,600
So up to 16 K context.

433
00:28:02,600 --> 00:28:05,640
I think it was, I think the standard versions at like 4K.

434
00:28:05,640 --> 00:28:07,740
So that's like a big leap.

435
00:28:07,740 --> 00:28:08,920
You do have to pay more.

436
00:28:08,920 --> 00:28:14,680
That's double the price, but you can do a lot more when you can send a lot more information.

437
00:28:14,680 --> 00:28:20,000
The classic use case for you and I Martin is pushing long files like large transcripts

438
00:28:20,000 --> 00:28:24,600
and other very long documents to it to maybe ask it to summarize or whatever.

439
00:28:24,600 --> 00:28:31,040
The token price as well is actually still pretty cheap on particularly 3.5.

440
00:28:31,040 --> 00:28:33,560
The token price is really cheap.

441
00:28:33,560 --> 00:28:36,600
Less so on GPT 4, that is still a bit pricey.

442
00:28:36,600 --> 00:28:37,600
Yep.

443
00:28:37,600 --> 00:28:44,960
But we've got 32 K context coming to GPT 4 more beyond, I know people have been struggling

444
00:28:44,960 --> 00:28:45,960
to get hold of this.

445
00:28:45,960 --> 00:28:47,920
I do think they've provided it to some developers.

446
00:28:47,920 --> 00:28:50,040
Very, very small amount.

447
00:28:50,040 --> 00:28:52,240
I think it's one of their priorities, isn't it?

448
00:28:52,240 --> 00:28:55,680
To get it rolled out very quickly in the coming weeks they're talking about.

449
00:28:55,680 --> 00:28:56,680
Yeah.

450
00:28:56,680 --> 00:28:57,680
I mean, and that is super cool, right?

451
00:28:57,680 --> 00:29:02,440
Because that's even double the massive increase that we just described for 3.5.

452
00:29:02,440 --> 00:29:09,000
And again, it will allow developers to send lots more information to have chat GPT to

453
00:29:09,000 --> 00:29:10,000
do summarizations.

454
00:29:10,000 --> 00:29:14,360
And we've got a workflow here at Bystratil where we summarize, we take the transcript

455
00:29:14,360 --> 00:29:17,440
from this podcast and we have to cut it up into multiple chunks so we can feed it into

456
00:29:17,440 --> 00:29:20,160
different tools to be able to get different summaries because it can't handle the whole

457
00:29:20,160 --> 00:29:22,720
transcript because you and I talk too much, Martin.

458
00:29:22,720 --> 00:29:24,600
So the transcript is too long.

459
00:29:24,600 --> 00:29:27,720
Whereas with this nice big context, that would be different.

460
00:29:27,720 --> 00:29:31,240
So there's a load of other stuff here as it relates to other little tweaks.

461
00:29:31,240 --> 00:29:37,000
But in essence, some more functions that are going to be really, really powerful developers,

462
00:29:37,000 --> 00:29:41,080
some more context so you can send more information in, in one go, which is super helpful for

463
00:29:41,080 --> 00:29:44,000
a number of applications we'll talk about in a moment.

464
00:29:44,000 --> 00:29:50,960
And also, in many cases, it's a bit cheaper to do certain aspects as well.

465
00:29:50,960 --> 00:29:52,480
Why is this important for marketers?

466
00:29:52,480 --> 00:29:57,560
So outside of all of the developer benefits, which then translate into benefits for us,

467
00:29:57,560 --> 00:30:01,000
and let's be honest, a huge amount of the tools that are being produced at the moment

468
00:30:01,000 --> 00:30:07,320
are a perfect fit for business people, especially marketers and sales folk, but probably marketers

469
00:30:07,320 --> 00:30:08,320
above all else.

470
00:30:08,320 --> 00:30:11,760
So I'm probably looking through my own very specific window on the world, but that's definitely

471
00:30:11,760 --> 00:30:13,320
what I see.

472
00:30:13,320 --> 00:30:18,480
The other thing is, we talked at the beginning of this podcast about where are people on

473
00:30:18,480 --> 00:30:25,200
the front end of this wave going, and they're going towards building their own integrations

474
00:30:25,200 --> 00:30:28,640
with tools like Zapier where they don't need to be able to code, but they've got an idea

475
00:30:28,640 --> 00:30:31,200
for something that they need to be able to do.

476
00:30:31,200 --> 00:30:36,400
And that means access to the OpenAI API.

477
00:30:36,400 --> 00:30:40,460
That means you're pushing information into Whisper to get transcripts, or you're pushing

478
00:30:40,460 --> 00:30:44,520
information into ChatGPT to do different things.

479
00:30:44,520 --> 00:30:50,480
And so whenever the context window gets bigger, or the cost gets less, or there becomes wider

480
00:30:50,480 --> 00:30:55,920
access to GPT-4 and its wider context window, that's cool things that those people who are

481
00:30:55,920 --> 00:31:00,280
experimenting with automation tools like Zapier, and there are others available, Zapier, you

482
00:31:00,280 --> 00:31:03,520
do want to sponsor the podcast, you'd be very welcome.

483
00:31:03,520 --> 00:31:07,180
They can do cool things, even cooler than what they're doing now.

484
00:31:07,180 --> 00:31:11,560
So if you're looking to get on the front of that wave, I think these are the types of

485
00:31:11,560 --> 00:31:15,360
things that perhaps you could get excited about and start playing with.

486
00:31:15,360 --> 00:31:21,880
Right, next story is EU Parliament votes through the EU AI Act.

487
00:31:21,880 --> 00:31:23,800
Martin, this caught your eye this week.

488
00:31:23,800 --> 00:31:24,800
What were your thoughts?

489
00:31:24,800 --> 00:31:28,640
Well, we've been following this one over a number of weeks now.

490
00:31:28,640 --> 00:31:32,040
We knew that the EU had brought forward the AI Act.

491
00:31:32,040 --> 00:31:37,800
They're trying to be a global leader in this domain.

492
00:31:37,800 --> 00:31:43,120
The EU Parliament has actually approved the draft for the AI Act.

493
00:31:43,120 --> 00:31:44,980
This was on June 14th.

494
00:31:44,980 --> 00:31:52,760
So now they are basically setting the negotiating position for the Parliament.

495
00:31:52,760 --> 00:31:56,160
A few key features from the Act.

496
00:31:56,160 --> 00:32:04,120
The law includes a ban on police use of live facial recognition in public places.

497
00:32:04,120 --> 00:32:09,440
So they can't be using cameras to tag who is in a crowd.

498
00:32:09,440 --> 00:32:15,780
MEPs will now negotiate with EU countries to finalise the law, which will come into

499
00:32:15,780 --> 00:32:20,800
place in three years time, 2026.

500
00:32:20,800 --> 00:32:26,000
The rules will regulate all sorts of different AI uses, not just marketing, like we talk

501
00:32:26,000 --> 00:32:32,420
about, but everything from medical AI, drones, deepfakes, chatbots, large language models

502
00:32:32,420 --> 00:32:34,160
and more.

503
00:32:34,160 --> 00:32:43,200
It really takes this risk-based approach so that an AI model that is making healthcare

504
00:32:43,200 --> 00:32:48,520
funding decisions for an individual will be higher risk versus an AI chatbot that tells

505
00:32:48,520 --> 00:32:52,440
you when the cinema times are in your local area, things like that.

506
00:32:52,440 --> 00:32:56,560
A couple of interesting observations in the Guardian reporting of this.

507
00:32:56,560 --> 00:33:00,320
So the Act passed very comfortably.

508
00:33:00,320 --> 00:33:05,160
I think it had 499 in favour of it and fewer than 100 against.

509
00:33:05,160 --> 00:33:13,020
But there were said to be some collection of MEPs from the far right who were going

510
00:33:13,020 --> 00:33:22,100
to vote against it because of the restrictions on it, such as being able to use live facial

511
00:33:22,100 --> 00:33:24,480
recognition for the police.

512
00:33:24,480 --> 00:33:27,520
They want more biometric surveillance.

513
00:33:27,520 --> 00:33:33,680
They want more capabilities to clamp down and identify people in a crowd.

514
00:33:33,680 --> 00:33:41,160
Yeah, so who would have thought that the far right would want to be overly aggressive and

515
00:33:41,160 --> 00:33:44,720
fascist in their approach to managing society?

516
00:33:44,720 --> 00:33:50,640
That's all I've got on that really.

517
00:33:50,640 --> 00:33:58,320
I was taken by the emotional recognition tech aspect of the story for workplaces and schools

518
00:33:58,320 --> 00:34:04,360
and how you're not allowed to use it and thinking, I don't really need that for this podcast

519
00:34:04,360 --> 00:34:07,600
because I can see how much fun you're having, Ma, and it's written all over your face.

520
00:34:07,600 --> 00:34:10,160
I don't need AI to tell me about that.

521
00:34:10,160 --> 00:34:14,960
Isn't that a creepy bit of tech though?

522
00:34:14,960 --> 00:34:26,360
It really makes me recoil the idea of an employer having emotion recognition technology in the

523
00:34:26,360 --> 00:34:27,360
workplace.

524
00:34:27,360 --> 00:34:32,840
Can you imagine this at BioStrata just coming in one morning and just doing a quick analysis

525
00:34:32,840 --> 00:34:37,040
on the state of the room, just on people's facial expressions?

526
00:34:37,040 --> 00:34:41,880
I know actually the technology itself is junk.

527
00:34:41,880 --> 00:34:48,480
Emotional recognition and emotional detection is so culturally specific that it's similar

528
00:34:48,480 --> 00:34:53,160
to the GPT detection tools that are on the market at the moment that have too many false

529
00:34:53,160 --> 00:34:55,480
positives and false negatives.

530
00:34:55,480 --> 00:34:57,000
It's very similar to that at the moment.

531
00:34:57,000 --> 00:35:02,760
The emotional detection is not in any way commercial ready.

532
00:35:02,760 --> 00:35:04,320
Cool.

533
00:35:04,320 --> 00:35:08,560
Sounds like my emotional intelligence as well to be honest, so that's commercially ready.

534
00:35:08,560 --> 00:35:10,000
Not ready for public consumption.

535
00:35:10,000 --> 00:35:13,720
No, no, but I am almost certain that you're having a good time because you've got a big

536
00:35:13,720 --> 00:35:18,040
smile on your face and I do believe it's real.

537
00:35:18,040 --> 00:35:22,080
Talking about something else that may or may not be real, a little segue that as you're

538
00:35:22,080 --> 00:35:29,800
going to see, dear listen, doesn't make any sense at all, is into our next news story.

539
00:35:29,800 --> 00:35:34,920
There is this report here from Morgan Stanley that there aren't actually that many people

540
00:35:34,920 --> 00:35:37,720
using chat GPT.

541
00:35:37,720 --> 00:35:40,680
This speaks to what we were talking about a bit earlier, Martin, in terms of when you're

542
00:35:40,680 --> 00:35:47,280
inside the bottle, can't read the label, when you're inside baseball, you know the inside

543
00:35:47,280 --> 00:35:51,720
bits of it, but I think sometimes you can forget what the rest of the world looks like.

544
00:35:51,720 --> 00:35:59,280
According to a Morgan Stanley survey of 2000 people, only 19% had used chat GPT and only

545
00:35:59,280 --> 00:36:02,680
9% had used Google Bard.

546
00:36:02,680 --> 00:36:06,160
The majority of people not using chat bots were unlikely to use them in the next six

547
00:36:06,160 --> 00:36:08,200
months, which is interesting, right?

548
00:36:08,200 --> 00:36:12,040
It's like, I haven't used them, not going to use them next.

549
00:36:12,040 --> 00:36:13,040
Which is fine.

550
00:36:13,040 --> 00:36:16,920
Chatbot services were mostly being used for learning about new topics.

551
00:36:16,920 --> 00:36:21,280
Early adopters commonly were using chat bots for researching products, comparing prices

552
00:36:21,280 --> 00:36:24,520
and shopping.

553
00:36:24,520 --> 00:36:26,680
Just on that point, I did the other day.

554
00:36:26,680 --> 00:36:38,080
I bought a new PC for things that people need PCs for and I've not bought a new computer,

555
00:36:38,080 --> 00:36:43,280
certainly not one that had any kind of these are spec on it, a desktop at least for ages.

556
00:36:43,280 --> 00:36:50,000
I had very little idea about what's the best GPU, blah, blah, blah.

557
00:36:50,000 --> 00:36:55,240
I used Google Bard to give me the rundown and it was brilliant.

558
00:36:55,240 --> 00:36:56,600
It was so effective.

559
00:36:56,600 --> 00:37:01,920
The whole experience of it was an absolute pleasure because I had just enough knowledge

560
00:37:01,920 --> 00:37:02,920
about the subject matter.

561
00:37:02,920 --> 00:37:08,440
I understood, I understand things like what a processor is for and what RAM is for and

562
00:37:08,440 --> 00:37:10,000
hard drives and GPUs.

563
00:37:10,000 --> 00:37:13,880
But I don't know if you tell me that this one versus this one and just give me the numbers

564
00:37:13,880 --> 00:37:18,840
and the names of the models, the Ryzen 7 versus the, I don't know.

565
00:37:18,840 --> 00:37:20,640
I haven't a clue.

566
00:37:20,640 --> 00:37:24,720
Being able to just stick in the specs for different computers and go, this is what I

567
00:37:24,720 --> 00:37:32,000
need it to do, which of these is going to be the best one to do it?

568
00:37:32,000 --> 00:37:33,000
It worked brilliantly.

569
00:37:33,000 --> 00:37:36,680
I was very impressed with my experience there.

570
00:37:36,680 --> 00:37:40,240
Sorry, that was just a little tangent about how I buy things now.

571
00:37:40,240 --> 00:37:43,080
But I think that's useful because I haven't really used it for that yet.

572
00:37:43,080 --> 00:37:48,000
And of course, that is eventually going to be probably what our normal search experience

573
00:37:48,000 --> 00:37:51,520
looks like, especially when it comes to product exploration.

574
00:37:51,520 --> 00:37:57,120
So I hadn't really thought about BARD as being a good way of testing a little bit what that

575
00:37:57,120 --> 00:37:58,120
would be like.

576
00:37:58,120 --> 00:38:03,080
Of course, it doesn't have a load of the other things that we expect to see in Google's generative

577
00:38:03,080 --> 00:38:06,440
search experience in terms of ads on the side and all that other stuff.

578
00:38:06,440 --> 00:38:10,480
But yeah, I think I'll have to have a play with that.

579
00:38:10,480 --> 00:38:15,040
So yeah, I mean, this is a data point in terms of a sea of data points.

580
00:38:15,040 --> 00:38:18,840
And it's a little bit like there are lies, damn lies and statistics because you can read

581
00:38:18,840 --> 00:38:22,360
one report and 90% of people have tried ChatGPT.

582
00:38:22,360 --> 00:38:27,700
You can have another one and only 20% odd have and don't intend to ever touch it again.

583
00:38:27,700 --> 00:38:33,680
So I'm sure there's a massive aspect of who you ask, what they do, how old they are, how

584
00:38:33,680 --> 00:38:37,320
tech savvy they are, etc, etc, to this.

585
00:38:37,320 --> 00:38:41,480
But certainly there's still a decent chunk of the population who are not as exposed to

586
00:38:41,480 --> 00:38:46,200
this as perhaps sometimes people like Martin and I might feel they are.

587
00:38:46,200 --> 00:38:47,200
What do you say, Nige?

588
00:38:47,200 --> 00:38:54,160
Yeah, I think that if you take the Boston Consulting Group one that we saw earlier with

589
00:38:54,160 --> 00:39:00,240
the 90-91% that said they'd used it, that's a self-selected audience, isn't it?

590
00:39:00,240 --> 00:39:05,800
Well, not self-selected, but it's a very specific audience that are living in this world of

591
00:39:05,800 --> 00:39:06,800
tech and data.

592
00:39:06,800 --> 00:39:11,600
If you're a CMO, you've got to be aware of how you're going to get that competitive advantage

593
00:39:11,600 --> 00:39:15,520
and people are going AI, AI, AI, left, right and center.

594
00:39:15,520 --> 00:39:19,520
So you're bound to have had a bit of a dabble with it because if you're not, your competitors

595
00:39:19,520 --> 00:39:20,520
are.

596
00:39:20,520 --> 00:39:29,560
If you take a snapshot of 2000 people in the world, just like a general, whoever you are,

597
00:39:29,560 --> 00:39:34,200
you could be a delivery driver for UPS, you could be a primary school teacher, you could

598
00:39:34,200 --> 00:39:36,120
be this, that or the other.

599
00:39:36,120 --> 00:39:40,000
Your world isn't looking for that competitive advantage.

600
00:39:40,000 --> 00:39:44,640
So where's the urgency to try these tools?

601
00:39:44,640 --> 00:39:50,040
I'm sure you'll get that they will get to people at some point, particularly when copilot

602
00:39:50,040 --> 00:39:52,360
and things like that roll out.

603
00:39:52,360 --> 00:39:55,400
But yeah, it didn't come as a massive surprise to me.

604
00:39:55,400 --> 00:40:00,760
I do the workshops with business owners.

605
00:40:00,760 --> 00:40:05,400
Increasingly over the past 12 months, more and more people, a higher percentage of people

606
00:40:05,400 --> 00:40:07,560
attending the sessions are using them.

607
00:40:07,560 --> 00:40:14,480
However, I did a workshop the other week, 20 people in the room and still four of them,

608
00:40:14,480 --> 00:40:20,340
even though the session was called Mastering ChatGPT and Generative AI, four of them had

609
00:40:20,340 --> 00:40:22,340
never used ChatGPT.

610
00:40:22,340 --> 00:40:27,920
They were looking to develop a lot over the course of that session.

611
00:40:27,920 --> 00:40:35,920
I think the other thing here is, and I haven't gone that deep into the BCG report, but I

612
00:40:35,920 --> 00:40:41,880
also can imagine a world in which your BCG, you assume that CMOs are all over technology

613
00:40:41,880 --> 00:40:47,480
developments.

614
00:40:47,480 --> 00:40:51,520
How do you think Generative AI will impact the efficiency of your organization from not

615
00:40:51,520 --> 00:40:55,040
very much through to massive enhancements?

616
00:40:55,040 --> 00:40:58,560
I don't see many CMOs going, sorry, Generative what?

617
00:40:58,560 --> 00:41:00,280
ChatGPT what?

618
00:41:00,280 --> 00:41:06,880
So I think there might even be some aspects of that, that by definition, assume some level

619
00:41:06,880 --> 00:41:08,600
of knowledge that may or may not be there.

620
00:41:08,600 --> 00:41:14,080
I'm sure it is there because I do think it's their job like you described, Martin, but

621
00:41:14,080 --> 00:41:17,960
surveys are only as good as the people that you survey and the questions that you ask.

622
00:41:17,960 --> 00:41:22,240
It probably sounds a little bit obvious, but when you actually dive deep into what happens

623
00:41:22,240 --> 00:41:26,760
when you get either of those things wrong, or you accidentally bias a certain research

624
00:41:26,760 --> 00:41:30,160
project in a certain way, you can pretty much prove whatever you like.

625
00:41:30,160 --> 00:41:37,640
But hey, it's good to get these bits of information and just get a feel for what people are saying

626
00:41:37,640 --> 00:41:42,040
and what people are seeing in the market and try and get a cross section of all these different

627
00:41:42,040 --> 00:41:43,040
things.

628
00:41:43,040 --> 00:41:45,240
Right, that's our stories for the week.

629
00:41:45,240 --> 00:41:50,480
Our tool of the week as we mentioned is Contea, but a little bit of a special one this week.

630
00:41:50,480 --> 00:41:54,720
We're going to hear directly from Sven, who's one of the co-founders of Contea, who caught

631
00:41:54,720 --> 00:42:00,840
up with Martin in an interview earlier this week.

632
00:42:00,840 --> 00:42:05,480
Hello Sven and welcome to Artificially Intelligent Marketing.

633
00:42:05,480 --> 00:42:08,680
Thank you for joining us today.

634
00:42:08,680 --> 00:42:09,680
Thank you for having me.

635
00:42:09,680 --> 00:42:12,400
It's a pleasure to have you on.

636
00:42:12,400 --> 00:42:20,180
We connected on LinkedIn after you introduced us to your app, which is, am I pronouncing

637
00:42:20,180 --> 00:42:21,180
it right?

638
00:42:21,180 --> 00:42:24,640
Contee.ai or have I got that wrong?

639
00:42:24,640 --> 00:42:29,400
Is it the British love of a cup of tea that is throwing me off there?

640
00:42:29,400 --> 00:42:31,760
It definitely is.

641
00:42:31,760 --> 00:42:33,160
We are thinking about that quite a lot.

642
00:42:33,160 --> 00:42:36,960
Obviously, we're both Germans, right, and we were searching for a name like, we're going

643
00:42:36,960 --> 00:42:38,620
to name this.

644
00:42:38,620 --> 00:42:43,840
We came up with the name Contea, which is like kind of a mixture between content and

645
00:42:43,840 --> 00:42:50,240
EA, which is like in our understanding, like emotional intelligence or like artificial

646
00:42:50,240 --> 00:42:52,120
intelligence, like a mixture of those two.

647
00:42:52,120 --> 00:42:56,360
So after we throw all of those three things together, we came up with the name Contea.

648
00:42:56,360 --> 00:43:01,280
We were thinking like, but how is a British speaker going to do it, like an English speaker

649
00:43:01,280 --> 00:43:02,660
and pronounce this?

650
00:43:02,660 --> 00:43:04,400
We didn't know, but now I know.

651
00:43:04,400 --> 00:43:07,040
We actually pronounced it out, Contee.

652
00:43:07,040 --> 00:43:08,040
So I did say Contee.

653
00:43:08,040 --> 00:43:10,760
I actually think it would be Contee.

654
00:43:10,760 --> 00:43:15,200
But Contee was the first thing that I saw when I logged into the system and you sent

655
00:43:15,200 --> 00:43:16,200
me the link.

656
00:43:16,200 --> 00:43:19,800
So one of the questions I had for you was, where did the name come from?

657
00:43:19,800 --> 00:43:20,960
So that makes a great deal of sense.

658
00:43:20,960 --> 00:43:24,200
I like the emotional intelligence side of that.

659
00:43:24,200 --> 00:43:26,280
We can dig into that in a moment.

660
00:43:26,280 --> 00:43:34,560
Before we get into that, tell us about the tool and what it does.

661
00:43:34,560 --> 00:43:40,400
I might start by saying that both Nick, my co-founder, and I, we are both copywriters.

662
00:43:40,400 --> 00:43:47,200
So we earn our bread, our money by kind of transporting ideas from people into written

663
00:43:47,200 --> 00:43:49,680
text.

664
00:43:49,680 --> 00:43:54,920
And when this whole JetJPT thing went down, like that was like a big move in going through

665
00:43:54,920 --> 00:43:59,000
the copywriting scene, everybody was like walking, running around like headless chicken

666
00:43:59,000 --> 00:44:05,240
running, oh my God, we're all going to be substituted by this AI thing, what not.

667
00:44:05,240 --> 00:44:11,280
So Rick Rada took a different approach and we were thinking, all right, so you can't

668
00:44:11,280 --> 00:44:16,040
stop this avalanche of technological advancements.

669
00:44:16,040 --> 00:44:19,160
So the only thing you can do is try to use it.

670
00:44:19,160 --> 00:44:23,560
So we started experimenting of how to use it to become better copywriters.

671
00:44:23,560 --> 00:44:26,660
And this is how we eventually ended up with Conteer.

672
00:44:26,660 --> 00:44:34,720
So what Conteer does is it helps you to create LinkedIn contents in up to 10 times of the

673
00:44:34,720 --> 00:44:37,960
speed of just writing it.

674
00:44:37,960 --> 00:44:42,520
And how it works is this, Conteer creates customized questions for you.

675
00:44:42,520 --> 00:44:46,800
So for example, you are already locked in and you tried it out, it generated questions

676
00:44:46,800 --> 00:44:49,840
about your specific business.

677
00:44:49,840 --> 00:44:58,000
Those questions then you get into your interface and you can answer them through voicemail.

678
00:44:58,000 --> 00:45:03,040
Because in our experience, talking about something is pretty easy, but writing it down, this

679
00:45:03,040 --> 00:45:04,900
is the difficult part.

680
00:45:04,900 --> 00:45:08,160
And this is exactly the part that Conteer takes over.

681
00:45:08,160 --> 00:45:14,420
So you talk something into it, you give it your input, then our tool takes your input

682
00:45:14,420 --> 00:45:19,360
and puts it into perfectly coherent and concise LinkedIn posts, which doesn't only help your

683
00:45:19,360 --> 00:45:25,840
expertise in it, but and more importantly, your tonality.

684
00:45:25,840 --> 00:45:29,000
So your specific words.

685
00:45:29,000 --> 00:45:33,840
Because many people now use JTPT to create LinkedIn contents, but this content is just

686
00:45:33,840 --> 00:45:36,320
very generic and doesn't sound like you at all.

687
00:45:36,320 --> 00:45:39,120
It doesn't have your words, it doesn't have your expertise.

688
00:45:39,120 --> 00:45:43,900
And with Conteer, we can help you to create content which actually sounds like you and

689
00:45:43,900 --> 00:45:45,800
transports your expertise.

690
00:45:45,800 --> 00:45:51,680
Yeah, I must say I tried the platform and that was one of the things that really stood

691
00:45:51,680 --> 00:45:56,000
out for me was, first of all, using it was very, very simple.

692
00:45:56,000 --> 00:45:58,800
You know, you hit record and then you start talking.

693
00:45:58,800 --> 00:46:03,120
But in the output that it gave me, it was very reflective of the language that I used,

694
00:46:03,120 --> 00:46:10,460
not kind of verbatim because in the way that I was the kind of chain of thought when you're

695
00:46:10,460 --> 00:46:13,680
talking is a bit all over the place, maybe a bit rambling.

696
00:46:13,680 --> 00:46:18,360
It tidied it up, but it kept the prescient points and I really liked that.

697
00:46:18,360 --> 00:46:25,460
So was it your experience as a copywriter through doing the kind of background research

698
00:46:25,460 --> 00:46:29,820
when you work with a client, you have to sit down, you interview them, you dig into that.

699
00:46:29,820 --> 00:46:35,160
Was that what kind of led you down this path then, just thinking about that workflow?

700
00:46:35,160 --> 00:46:36,880
100%.

701
00:46:36,880 --> 00:46:45,740
So as a copywriter, what we try to do is we want to sound like the clients, but better.

702
00:46:45,740 --> 00:46:49,480
This is basically what describes our job.

703
00:46:49,480 --> 00:46:56,280
And there are many artificial intelligence tools out there, but they never really did

704
00:46:56,280 --> 00:46:58,160
this job.

705
00:46:58,160 --> 00:47:03,960
And I think what qualified us to do so, to create a tool which does, as you said, kind

706
00:47:03,960 --> 00:47:10,160
of reflect your tonality is that we understand the foundations of copywriting and we were

707
00:47:10,160 --> 00:47:15,080
able to implement those foundations of copywriting through our prompts.

708
00:47:15,080 --> 00:47:18,680
So what you can't see obviously because it's all in the back ends and it's kind of like

709
00:47:18,680 --> 00:47:21,280
we try to keep it secret, right?

710
00:47:21,280 --> 00:47:26,360
Our prompts, because they're very complex and it's not just like write a LinkedIn post

711
00:47:26,360 --> 00:47:27,360
based on this transcript.

712
00:47:27,360 --> 00:47:28,360
They are huge.

713
00:47:28,360 --> 00:47:34,960
There's a huge prompts where we try to put all of our expertise of copywriting into them.

714
00:47:34,960 --> 00:47:39,840
Like for example, write it concise, but also try to keep it colloquial and so on and so

715
00:47:39,840 --> 00:47:40,940
on.

716
00:47:40,940 --> 00:47:46,480
So yeah, I think to come back to a question, if this is what, this is basically what we

717
00:47:46,480 --> 00:47:51,800
try to do, like to take our job, what we do as copywriters, trying to understand who the

718
00:47:51,800 --> 00:47:54,800
person is we're writing for.

719
00:47:54,800 --> 00:48:02,480
We tried our best to put this into a tool, which of course we didn't succeed completely.

720
00:48:02,480 --> 00:48:05,640
So like it sounds like you, but not 100%.

721
00:48:05,640 --> 00:48:10,480
It gets you 80% of the way, but the last 30% you still have to edit the post a little bit.

722
00:48:10,480 --> 00:48:15,040
You have to make it a little bit more personal, but we're trying to get as close to the 100%

723
00:48:15,040 --> 00:48:16,520
authenticity as possible.

724
00:48:16,520 --> 00:48:17,520
Yeah.

725
00:48:17,520 --> 00:48:23,720
And I think that's general good advice for anyone using any generative AI tool at the

726
00:48:23,720 --> 00:48:24,720
moment, right?

727
00:48:24,720 --> 00:48:29,960
That they take you so far, but they're never going to get you wholly over the line.

728
00:48:29,960 --> 00:48:31,400
You still need human in the loop.

729
00:48:31,400 --> 00:48:34,720
And that's something we talk about on the show quite a lot.

730
00:48:34,720 --> 00:48:39,040
This isn't something that we're fully outsourcing.

731
00:48:39,040 --> 00:48:42,760
So from a marketing perspective then, what are the marketing use cases?

732
00:48:42,760 --> 00:48:49,420
What could our listeners do with Contea that would help them in their marketing role?

733
00:48:49,420 --> 00:48:53,680
So this is a very interesting question because I can't give you a clear answer on that because

734
00:48:53,680 --> 00:48:58,140
we're still figuring that out or rather our customers.

735
00:48:58,140 --> 00:49:04,800
So the use case that we had in mind is for like small business owners, like Lee, for

736
00:49:04,800 --> 00:49:08,880
example, or like somebody who has like some kind of consulting, whatever, he doesn't have

737
00:49:08,880 --> 00:49:10,660
a big team.

738
00:49:10,660 --> 00:49:15,320
And that person wants to create awareness on LinkedIn, wants to create LinkedIn posts,

739
00:49:15,320 --> 00:49:19,320
but is struggling with it because first of all, you have to come up with the ideas.

740
00:49:19,320 --> 00:49:23,880
It is three or four a week and then you have to find the time to write them.

741
00:49:23,880 --> 00:49:26,480
And this is very difficult.

742
00:49:26,480 --> 00:49:31,160
So our use case that we had in mind is that those small business owners, they get the

743
00:49:31,160 --> 00:49:36,340
tool, they get their questions, they reply them through voicemail and within five minutes

744
00:49:36,340 --> 00:49:40,120
they have the post ready for a week.

745
00:49:40,120 --> 00:49:43,560
So this was our use case and this is what we thought that people can and what people

746
00:49:43,560 --> 00:49:45,720
actually can do with it.

747
00:49:45,720 --> 00:49:54,040
Now however, we get a lot of messages from agency owners or other copywriters say, look,

748
00:49:54,040 --> 00:50:01,400
I want to use this tool to scale my own work with my own clients.

749
00:50:01,400 --> 00:50:06,480
So what we're doing right now is we're building kind of like an surface level interface, like

750
00:50:06,480 --> 00:50:15,000
a kind of a login for the agencies where they have all of their clients in one interface.

751
00:50:15,000 --> 00:50:21,720
So this way, basically agencies can use our tool to automatize most of their work, which

752
00:50:21,720 --> 00:50:25,480
is their clients will get automatic questions.

753
00:50:25,480 --> 00:50:29,040
They get a first draft.

754
00:50:29,040 --> 00:50:32,060
As we said, like 80% of the way it gets you there.

755
00:50:32,060 --> 00:50:38,200
But then those agencies, they can still go into those generated posts of the clients,

756
00:50:38,200 --> 00:50:42,600
fit a little bit of them, change them a little bit to make them 100% authentic and then post

757
00:50:42,600 --> 00:50:44,140
them.

758
00:50:44,140 --> 00:50:46,800
So this is basically the second use case we are now working on.

759
00:50:46,800 --> 00:50:50,880
So either you are a small business owner and you want to create LinkedIn content for yourself

760
00:50:50,880 --> 00:50:55,200
or you are an agency and you want to use it to scale your work basically.

761
00:50:55,200 --> 00:50:58,000
Yeah, I love that second use case.

762
00:50:58,000 --> 00:51:05,520
The agency model having worked agency side and consulted on many, many B2B clients, quite

763
00:51:05,520 --> 00:51:10,680
technical, working with clients in the manufacturing and engineering space.

764
00:51:10,680 --> 00:51:16,160
I don't have that expertise, but I know the questions to ask so I can sit down, record

765
00:51:16,160 --> 00:51:21,200
an interview with them, hand that transcript over to a copywriter and get them to extract

766
00:51:21,200 --> 00:51:22,320
the detail.

767
00:51:22,320 --> 00:51:26,720
Having that in an interface is lovely, which kind of takes me to one of the big talking

768
00:51:26,720 --> 00:51:33,080
points that we have regularly on the podcast, which is AI and particularly generative AI

769
00:51:33,080 --> 00:51:40,680
and these models are now being baked into all of the tools that we use everywhere.

770
00:51:40,680 --> 00:51:49,680
The real difference maker for me in terms of what makes a tool really excel or gives

771
00:51:49,680 --> 00:51:56,760
a tool maybe the best chance of surviving and thriving in the kind of tech startup SaaS

772
00:51:56,760 --> 00:51:59,340
world is UX.

773
00:51:59,340 --> 00:52:06,840
If you can give somebody that user experience, which there were so many copywritings over

774
00:52:06,840 --> 00:52:13,520
the past couple of years since GPT-3 came out and made API access available.

775
00:52:13,520 --> 00:52:19,400
Many companies launched, some of them have reached unicorn status with basically putting

776
00:52:19,400 --> 00:52:24,400
a front end on to GPT-3 and DaVinci.

777
00:52:24,400 --> 00:52:27,500
That's kind of what they've done.

778
00:52:27,500 --> 00:52:29,440
It's not going to sustain, right?

779
00:52:29,440 --> 00:52:32,360
ChatGPT has pulled the rug from under them.

780
00:52:32,360 --> 00:52:36,640
You can go into ChatGPT and say, write me a meta description and it will write you a

781
00:52:36,640 --> 00:52:38,060
meta description.

782
00:52:38,060 --> 00:52:45,000
What you've done is interesting because you've taken multiple different AI models and technologies

783
00:52:45,000 --> 00:52:46,000
and modalities.

784
00:52:46,000 --> 00:52:53,960
So you've taken audio as well and put this into a very simple, easy to use workflow.

785
00:52:53,960 --> 00:52:58,120
Tell us a bit about the, without giving away the secret sauce, I'm sure you're not going

786
00:52:58,120 --> 00:53:00,360
to tell us what you're using and what have you.

787
00:53:00,360 --> 00:53:04,840
Tell us a bit about the AI technologies that you're using and how you've kind of built

788
00:53:04,840 --> 00:53:07,240
that stack.

789
00:53:07,240 --> 00:53:12,720
This is very interesting and is like in, right, it gets quite a lot.

790
00:53:12,720 --> 00:53:16,600
What you basically do is you're just using GPT-4 to write some copies.

791
00:53:16,600 --> 00:53:22,120
The interesting thing is, so first we try to build that whole thing no codes because

792
00:53:22,120 --> 00:53:23,700
we're both no developers.

793
00:53:23,700 --> 00:53:30,560
So I'm a copywriter, I know about words, but I have no idea what numbers.

794
00:53:30,560 --> 00:53:32,840
This is not my expertise at all.

795
00:53:32,840 --> 00:53:37,480
However, my co-founder, Nico, luckily he worked as a web developer so he has some ideas about

796
00:53:37,480 --> 00:53:42,480
like front-end development and stuff.

797
00:53:42,480 --> 00:53:46,360
But we tried to put that whole thing, like set it up, no code, which in the beginning

798
00:53:46,360 --> 00:53:52,480
we did through a five or six step process on Zapier, which is basically you have recorder

799
00:53:52,480 --> 00:53:58,000
that gets recorded that goes to a transcription tool, which we changed by now to whisper.

800
00:53:58,000 --> 00:54:00,600
We'll probably get into that later.

801
00:54:00,600 --> 00:54:07,460
And then basically you have to open AI like GPT-4 integration and write the post.

802
00:54:07,460 --> 00:54:13,180
By now that whole process is not on Zapier anymore, but on N8n, which is like an equivalent,

803
00:54:13,180 --> 00:54:19,320
but it gives us more variables, like you can actually code in the tool.

804
00:54:19,320 --> 00:54:23,600
So we have way more possibilities of integrations.

805
00:54:23,600 --> 00:54:28,880
And the whole step, which you don't see is like from hitting records to your finished

806
00:54:28,880 --> 00:54:35,440
posts is a 50 step process, which runs through in the backends.

807
00:54:35,440 --> 00:54:39,840
And open AI are only two of them.

808
00:54:39,840 --> 00:54:42,200
So like the whole process now is huge.

809
00:54:42,200 --> 00:54:45,280
And even for me, like it's so hard kind of like to figure something out.

810
00:54:45,280 --> 00:54:49,520
Like I'm just asking like a voice into Nico like, yeah, I just saw there's another step,

811
00:54:49,520 --> 00:54:51,520
like what the fuck is this actually doing?

812
00:54:51,520 --> 00:54:55,560
And it's like kind of going to D10, I have no idea what he's talking about.

813
00:54:55,560 --> 00:55:00,920
But basically most of the functionalities right now are like kind of on own service.

814
00:55:00,920 --> 00:55:06,560
For example, we tried to integrate Whispers AI into N8n and Zapier.

815
00:55:06,560 --> 00:55:11,240
And I know that you talked about it in previous episodes, but we couldn't get it running.

816
00:55:11,240 --> 00:55:12,240
It didn't work.

817
00:55:12,240 --> 00:55:14,280
We couldn't integrate the API.

818
00:55:14,280 --> 00:55:18,560
Maybe there was like a bug in the beginning and now it's possible.

819
00:55:18,560 --> 00:55:24,080
But basically all of those integrations have to run over an own server.

820
00:55:24,080 --> 00:55:29,480
So basically we have like the N8n process, but the recorder, for example, is an own code

821
00:55:29,480 --> 00:55:36,240
solution that recorded and sends in own like encodes all the information to the API of

822
00:55:36,240 --> 00:55:38,400
Whisper.

823
00:55:38,400 --> 00:55:44,800
This transcription then goes to Airtable, which is basically equivalent to Excel, right?

824
00:55:44,800 --> 00:55:46,880
Where all our data is stored.

825
00:55:46,880 --> 00:55:54,120
And then it goes back to OpenAI's GPT-4 to write the whole thing.

826
00:55:54,120 --> 00:55:57,080
But yeah, like basically it's like a huge process right now.

827
00:55:57,080 --> 00:56:03,000
Yeah, it's interesting you say that because me and Paul have both played with our own

828
00:56:03,000 --> 00:56:09,960
Zaps using Whisper and recording an mp3 file on your phone and it drops into Google Drive

829
00:56:09,960 --> 00:56:13,720
and then it turns into it.

830
00:56:13,720 --> 00:56:22,680
And I think what you've done in terms of that 50 step process and the layering in of the

831
00:56:22,680 --> 00:56:26,520
copywriting expertise is ultimately that's the secret sauce, isn't it?

832
00:56:26,520 --> 00:56:29,400
Like that's where the magic happens.

833
00:56:29,400 --> 00:56:32,720
So where do you see this space evolving?

834
00:56:32,720 --> 00:56:36,360
What's next?

835
00:56:36,360 --> 00:56:40,040
How reliant are you on new models?

836
00:56:40,040 --> 00:56:44,920
Are you interested to see where OpenAI goes with, you mentioned Whisper, that's clearly

837
00:56:44,920 --> 00:56:46,560
one that you're using.

838
00:56:46,560 --> 00:56:48,320
Are you looking at alternatives?

839
00:56:48,320 --> 00:56:52,400
Are you finding that reliable?

840
00:56:52,400 --> 00:56:58,080
So first of all, regarding Whisper, first we used other transcription modules.

841
00:56:58,080 --> 00:57:03,240
And the good thing about our tool is if we want to change from one integration to another

842
00:57:03,240 --> 00:57:07,080
one, the only thing we have to do is change the API.

843
00:57:07,080 --> 00:57:11,880
So for example, if we decide that the solution from Bing now soon will be better than JetJPT

844
00:57:11,880 --> 00:57:15,280
or GPT-4, this won't click to change for us.

845
00:57:15,280 --> 00:57:17,880
So we tried this out.

846
00:57:17,880 --> 00:57:19,600
It's not even close.

847
00:57:19,600 --> 00:57:25,280
So GPT-4 still is like above all of them, like far ahead.

848
00:57:25,280 --> 00:57:30,320
But we very quickly can change them to other solutions.

849
00:57:30,320 --> 00:57:35,760
So how do we see this evolving?

850
00:57:35,760 --> 00:57:40,600
Do you mean like with the technologies behind or you mean the whole copywriting space?

851
00:57:40,600 --> 00:57:46,240
Well, I guess either or the copywriting space is one that I think listeners will be interested

852
00:57:46,240 --> 00:57:51,600
in from a marketing perspective.

853
00:57:51,600 --> 00:57:57,480
How much of this is really going to disrupt copywriters?

854
00:57:57,480 --> 00:58:02,160
You have that experience yourself, you have the industry expertise.

855
00:58:02,160 --> 00:58:08,840
What do you think copywriters should be thinking right now?

856
00:58:08,840 --> 00:58:17,360
So as a roller coaster ride, so when JetJPT got launched, everybody was going crazy, was

857
00:58:17,360 --> 00:58:18,720
going nuts.

858
00:58:18,720 --> 00:58:22,160
Like some were saying, okay, this is the Armageddon, this is the end of copywriting.

859
00:58:22,160 --> 00:58:23,160
Others are like, oh, this is fabulous.

860
00:58:23,160 --> 00:58:27,560
We're going to create 10 posts in five minutes.

861
00:58:27,560 --> 00:58:33,000
And I think this very quickly cooled down and realization kind of settled in like, all

862
00:58:33,000 --> 00:58:40,280
right, this is not going to substitute copywriting soon because it's unprecedented.

863
00:58:40,280 --> 00:58:46,200
So I think most copywriters already realize that, but you can't outsource human copywriting

864
00:58:46,200 --> 00:58:48,060
just yet.

865
00:58:48,060 --> 00:58:53,120
So my advice for copywriters is don't rely on GPT to write some articles because first

866
00:58:53,120 --> 00:58:55,120
of all, they're going to be super generic.

867
00:58:55,120 --> 00:59:00,340
This won't give you any traction, any expertise, any thought leadership on LinkedIn.

868
00:59:00,340 --> 00:59:03,000
This won't build trust with the target audience.

869
00:59:03,000 --> 00:59:09,760
First of all, and second of all, going forward, I'm very sure if not, if they don't ever

870
00:59:09,760 --> 00:59:15,080
do that, that LinkedIn, the algorithm is going to punish you for it.

871
00:59:15,080 --> 00:59:22,200
So we already know that it is possible to track copies which were generated through

872
00:59:22,200 --> 00:59:23,200
AI.

873
00:59:23,200 --> 00:59:27,680
They're already like open source websites where you just copy paste one post inside

874
00:59:27,680 --> 00:59:32,320
and say, tell me the chance that this was created by a human being or an AI.

875
00:59:32,320 --> 00:59:34,560
And it's very, very accurate.

876
00:59:34,560 --> 00:59:35,800
We tried it out.

877
00:59:35,800 --> 00:59:37,740
We created some posts on JTPT.

878
00:59:37,740 --> 00:59:39,400
We put it into that tool.

879
00:59:39,400 --> 00:59:43,680
It gave us a 99% chance that this was created by JTPT.

880
00:59:43,680 --> 00:59:47,880
Because I'm very sure LinkedIn is going to implement it at one point because they have

881
00:59:47,880 --> 00:59:51,800
an interest of keeping their feet interested.

882
00:59:51,800 --> 00:59:54,200
And JTPT content is not interesting.

883
00:59:54,200 --> 00:59:55,200
It's generic.

884
00:59:55,200 --> 00:59:57,880
So it's going to like just like splash.

885
00:59:57,880 --> 01:00:00,960
Like it's going to be like boring to be on LinkedIn because everybody is now creating

886
01:00:00,960 --> 01:00:03,160
10 content pieces a day.

887
01:00:03,160 --> 01:00:09,680
So I'm very sure LinkedIn is going to find a way to stop this from happening.

888
01:00:09,680 --> 01:00:15,560
So relying on JTPT, I think is not smart for copyrights to do.

889
01:00:15,560 --> 01:00:17,920
What they can do is to make their work easier.

890
01:00:17,920 --> 01:00:22,280
For example, what I do with my clients, I get a transcription from a client and I put

891
01:00:22,280 --> 01:00:27,160
that in JTPT and tell it like, tell me what this post is about.

892
01:00:27,160 --> 01:00:28,500
What's the key message here?

893
01:00:28,500 --> 01:00:30,360
So this helps me to write a copy.

894
01:00:30,360 --> 01:00:36,740
But to actually write the copies, I would not advise to use JTPT.

895
01:00:36,740 --> 01:00:39,320
So this is part one of the answer.

896
01:00:39,320 --> 01:00:45,120
Don't rely on it, but use it.

897
01:00:45,120 --> 01:00:51,560
Because if you don't start now experimenting with it and finding use cases for yourself

898
01:00:51,560 --> 01:00:55,560
on how to use it to become better and quicker, others will do so.

899
01:00:55,560 --> 01:01:00,260
And in the end, they will do the same work that you do, but in 10 times the speed.

900
01:01:00,260 --> 01:01:04,920
So they are able to do it like in one tenth of the pricing.

901
01:01:04,920 --> 01:01:07,920
And then you are in big trouble.

902
01:01:07,920 --> 01:01:09,920
So I think those are the two things you have to be aware of.

903
01:01:09,920 --> 01:01:15,200
Don't rely on it because LinkedIn, very sure at one point, will integrate something like

904
01:01:15,200 --> 01:01:21,200
that, that it will filter out and punish your post with your generic JTPT post.

905
01:01:21,200 --> 01:01:22,960
But you have to stay on track.

906
01:01:22,960 --> 01:01:25,040
Otherwise you're really going to be in trouble.

907
01:01:25,040 --> 01:01:27,840
No, I think they're good tips.

908
01:01:27,840 --> 01:01:34,720
On the AI detection, there was an interesting article recently about OpenAI's own detection

909
01:01:34,720 --> 01:01:36,640
tool.

910
01:01:36,640 --> 01:01:39,840
And it's about 26% accurate.

911
01:01:39,840 --> 01:01:47,800
So even the open source ones, they throw out too many false positives and false negatives.

912
01:01:47,800 --> 01:01:55,440
So they miss too many and they add too many as being AI generated when they're not.

913
01:01:55,440 --> 01:01:58,360
And on the flip side, they miss too many for them to be reliable.

914
01:01:58,360 --> 01:02:03,440
So when I've tried them, similarly, I've had experiences where I put AI copy in and it's

915
01:02:03,440 --> 01:02:05,200
99% definite.

916
01:02:05,200 --> 01:02:09,800
But actually when you roll that out at scale, they're too unreliable.

917
01:02:09,800 --> 01:02:11,800
You can't actually trust them.

918
01:02:11,800 --> 01:02:13,280
They're actually quite bad.

919
01:02:13,280 --> 01:02:18,280
So I think we're safe from that perspective for the time being.

920
01:02:18,280 --> 01:02:23,880
But ultimately, this is going to be a big problem for the likes of anyone training a

921
01:02:23,880 --> 01:02:29,720
large language model because soon they're going to be training on content written by

922
01:02:29,720 --> 01:02:32,920
large language models and they're going to start cannibalizing their own content.

923
01:02:32,920 --> 01:02:37,040
So yeah, I think they have to be aware of it and they're going to have to start thinking

924
01:02:37,040 --> 01:02:39,080
about, well, I say start thinking.

925
01:02:39,080 --> 01:02:45,080
They're well ahead of the thinking on this, but finding a solution to identify is going

926
01:02:45,080 --> 01:02:46,080
to be interesting.

927
01:02:46,080 --> 01:02:48,640
I know they tried the watermarking system as well.

928
01:02:48,640 --> 01:02:53,880
And apparently, to mark the tech, basically minor edits will just get rid of the watermark

929
01:02:53,880 --> 01:02:56,460
completely.

930
01:02:56,460 --> 01:02:58,120
Which is also what we realized with Conteer.

931
01:02:58,120 --> 01:03:00,000
Like, are you saying very interesting what you just said?

932
01:03:00,000 --> 01:03:01,000
I didn't know that.

933
01:03:01,000 --> 01:03:06,960
Like we just tracked it all and said 99% certainty that it was generated by AI.

934
01:03:06,960 --> 01:03:12,440
When we did that with Conteer articles, it gave us like a 5% rating on it.

935
01:03:12,440 --> 01:03:17,840
So like I said, no, that was written by humans just because probably there were some words

936
01:03:17,840 --> 01:03:24,120
which were in those copies because they are coming from transcripts, which kind of got

937
01:03:24,120 --> 01:03:25,440
rid of that watermark already.

938
01:03:25,440 --> 01:03:31,520
So yeah, that's very easy to undergo.

939
01:03:31,520 --> 01:03:34,720
So about time to wrap it up.

940
01:03:34,720 --> 01:03:36,720
Where can people go to sign up?

941
01:03:36,720 --> 01:03:39,720
The product is still in beta, isn't it?

942
01:03:39,720 --> 01:03:41,720
It is still in beta, but now we've rolled it out.

943
01:03:41,720 --> 01:03:46,320
Just today, we kind of opened it now for everybody for the open market.

944
01:03:46,320 --> 01:03:47,320
So it's very easy to find.

945
01:03:47,320 --> 01:03:48,320
You just have to go to get.conteer.ai.

946
01:03:48,320 --> 01:03:54,760
Conteer with EA in the end.

947
01:03:54,760 --> 01:03:55,760
There you can sign up.

948
01:03:55,760 --> 01:03:57,680
It's very easy.

949
01:03:57,680 --> 01:03:59,840
And this whole thing is a beta version.

950
01:03:59,840 --> 01:04:01,440
So everything works already.

951
01:04:01,440 --> 01:04:04,600
This is a 0% working product.

952
01:04:04,600 --> 01:04:09,560
But like every two weeks, we come up with a new functionality or we tweak a little bit

953
01:04:09,560 --> 01:04:10,560
to make it a little bit better.

954
01:04:10,560 --> 01:04:15,480
So there are constant changes going to happen over the next weeks and months for sure.

955
01:04:15,480 --> 01:04:16,480
Great.

956
01:04:16,480 --> 01:04:19,760
And is that available with a one month trial?

957
01:04:19,760 --> 01:04:22,520
What's the pricing of it?

958
01:04:22,520 --> 01:04:23,520
This is a good question.

959
01:04:23,520 --> 01:04:31,760
We thought about providing a free trial and we consciously decided not to do it.

960
01:04:31,760 --> 01:04:36,400
Because of AI tools, the problem is that there are so many AI tools right now.

961
01:04:36,400 --> 01:04:40,560
You could try a different AI tool every single day.

962
01:04:40,560 --> 01:04:46,320
And we kind of wanted to avoid getting a lot of clients to just use it out of pure interest.

963
01:04:46,320 --> 01:04:49,500
Like, oh, let's check this out, what this is doing.

964
01:04:49,500 --> 01:04:50,920
But we really want to filter out.

965
01:04:50,920 --> 01:04:56,720
The people we want are really people who see the benefit of it before using it.

966
01:04:56,720 --> 01:04:58,040
So there is no free trial.

967
01:04:58,040 --> 01:05:04,920
It is 75 euros, dollar, whatever it is, the same a month.

968
01:05:04,920 --> 01:05:09,400
However, there is a 100% money back guarantee.

969
01:05:09,400 --> 01:05:14,720
So after one month, you use it and you say, look, this is not doing what it is supposed

970
01:05:14,720 --> 01:05:15,720
to do.

971
01:05:15,720 --> 01:05:17,280
It didn't deliver.

972
01:05:17,280 --> 01:05:22,800
Then you just have to contact us in support and you get the money back instantly.

973
01:05:22,800 --> 01:05:23,800
Fantastic.

974
01:05:23,800 --> 01:05:24,800
Very generous.

975
01:05:24,800 --> 01:05:27,680
Well, thank you for joining us on the podcast this week.

976
01:05:27,680 --> 01:05:33,680
And we look forward to getting hands on with the tool in the future.

977
01:05:33,680 --> 01:05:34,680
Thank you very much for having me.

978
01:05:34,680 --> 01:05:40,080
I'm looking forward to have further conversation maybe with you, right down the line.

979
01:05:40,080 --> 01:05:43,480
So thanks very much there to Sven for coming on the podcast.

980
01:05:43,480 --> 01:05:44,480
Really appreciate that.

981
01:05:44,480 --> 01:05:48,480
It was a fascinating conversation with Martin there.

982
01:05:48,480 --> 01:05:54,280
I learned loads more than I was expecting to, not just about the tool, which is fantastic.

983
01:05:54,280 --> 01:05:59,000
So yeah, go and have a little play with Conteo and see if it can help you with some of the

984
01:05:59,000 --> 01:06:01,160
things that you're trying to do in your work.

985
01:06:01,160 --> 01:06:03,720
I've really enjoyed today's session.

986
01:06:03,720 --> 01:06:06,040
Martin, I hope you have as well.

987
01:06:06,040 --> 01:06:07,200
It was a bit of a good one.

988
01:06:07,200 --> 01:06:08,200
We got into a few good bits.

989
01:06:08,200 --> 01:06:12,280
Yeah, I like the way you describe it as a session, almost like therapy.

990
01:06:12,280 --> 01:06:13,280
Yeah.

991
01:06:13,280 --> 01:06:16,000
And maybe we've got a couple of tinnies on the go here as well.

992
01:06:16,000 --> 01:06:18,480
A couple of Budweiser to get us through.

993
01:06:18,480 --> 01:06:21,960
I think the listeners are thinking, yeah, I probably am going to need to turn to drink

994
01:06:21,960 --> 01:06:25,600
if I'm going to listen to any more episodes of this.

995
01:06:25,600 --> 01:06:30,340
This is not sponsored, but I'm currently drinking BrewDog Elvis juice, but I'm drinking the

996
01:06:30,340 --> 01:06:35,200
alcohol free version, not the 6.5 or fat version.

997
01:06:35,200 --> 01:06:39,800
I like it because you've got to keep it lucid when you're getting deep into the AI topics,

998
01:06:39,800 --> 01:06:40,800
Martin.

999
01:06:40,800 --> 01:06:41,800
Absolutely.

1000
01:06:41,800 --> 01:06:46,440
So if you have enjoyed this podcast and it's your first time here, where have you been?

1001
01:06:46,440 --> 01:06:49,400
But also please do consider hitting the subscribe button.

1002
01:06:49,400 --> 01:06:53,680
If there are other people in the world of marketing or just business folks that you

1003
01:06:53,680 --> 01:06:58,980
think might benefit from this, please forward it onto them and see if they fancy subscribing.

1004
01:06:58,980 --> 01:07:04,600
You can catch up with us on the Twitters if you go to our handle, which is AIMarketingPod.

1005
01:07:04,600 --> 01:07:05,600
Correct.

1006
01:07:05,600 --> 01:07:06,600
Yes.

1007
01:07:06,600 --> 01:07:11,960
15 episodes and he has nailed it.

1008
01:07:11,960 --> 01:07:13,560
And that's all it took.

1009
01:07:13,560 --> 01:07:16,340
Imagine what I'll learn over the next 100 episodes.

1010
01:07:16,340 --> 01:07:17,520
How exciting.

1011
01:07:17,520 --> 01:07:25,080
You can also subscribe at ArtificiallyIntelligentMarketing.com where all of these podcasts also live and

1012
01:07:25,080 --> 01:07:29,360
you'll get an automatic email when there is a new episode live so that you don't even

1013
01:07:29,360 --> 01:07:30,360
have to remember.

1014
01:07:30,360 --> 01:07:35,800
And of course you can subscribe to these on all of your favorite podcasting platforms.

1015
01:07:35,800 --> 01:07:37,280
Thank you so much for your time, Martin.

1016
01:07:37,280 --> 01:07:38,280
I've really enjoyed it.

1017
01:07:38,280 --> 01:07:41,260
I hope you have a great weekend and I'll see you next week.

1018
01:07:41,260 --> 01:07:42,560
See you next week.

1019
01:07:42,560 --> 01:07:43,560
Bye.

1020
01:07:43,560 --> 01:07:50,520
Thank you for listening to Artificially Intelligent Marketing.

1021
01:07:50,520 --> 01:07:56,560
To stay on top of the latest trends, tips and tools in the world of marketing AI, be

1022
01:07:56,560 --> 01:07:58,320
sure to subscribe.

1023
01:07:58,320 --> 01:08:12,320
We look forward to seeing you again next week.