1
00:00:00,000 --> 00:00:09,960
Welcome to Artificially Intelligent Marketing, a weekly podcast where we stay on top of the

2
00:00:09,960 --> 00:00:15,700
latest trends, tips and tools in the world of marketing AI, helping you get the best

3
00:00:15,700 --> 00:00:18,540
results from your marketing efforts.

4
00:00:18,540 --> 00:00:23,320
Now let's join our hosts, Paul Avery and Martin Broadhurst.

5
00:00:23,320 --> 00:00:27,820
Welcome to Episode 26 of Artificially Intelligent Marketing.

6
00:00:27,820 --> 00:00:32,560
We are here to bring you all the news that you need to know about if you're a marketer

7
00:00:32,560 --> 00:00:34,840
interested in the power of AI.

8
00:00:34,840 --> 00:00:38,400
As usual, I am joined by my very good friend, Martin Broadhurst.

9
00:00:38,400 --> 00:00:41,400
Martin, how are you today?

10
00:00:41,400 --> 00:00:47,600
The jet lag from post-inbound has finally subsided, but I'm good.

11
00:00:47,600 --> 00:00:51,000
Yeah, Friday afternoon, lots of news to get through this week.

12
00:00:51,000 --> 00:00:54,480
It's been an absolute bummer week in terms of AI.

13
00:00:54,480 --> 00:00:59,160
And we were saying on the WhatsApp earlier this week that it felt like a relatively slow

14
00:00:59,160 --> 00:01:02,480
week actually earlier on in the week.

15
00:01:02,480 --> 00:01:06,960
Once we gathered all the stories up and we realized how much we had to get through, we

16
00:01:06,960 --> 00:01:11,280
were glad that we didn't have an interview to go live with this week because we honestly

17
00:01:11,280 --> 00:01:12,840
wouldn't have had the time.

18
00:01:12,840 --> 00:01:19,000
So with that in mind, we should probably quit the preamble and get into the productive discussions.

19
00:01:19,000 --> 00:01:22,160
So Martin, why don't you take us through our first story this week?

20
00:01:22,160 --> 00:01:27,240
This one's a story that's been bubbling under the surface for a couple of weeks now, but

21
00:01:27,240 --> 00:01:34,560
Google is gearing up to launch Gemini, its own suite of conversational AI software.

22
00:01:34,560 --> 00:01:42,280
And this is big because it's actually intended to take on GPT-4, but aims to be a real formidable

23
00:01:42,280 --> 00:01:45,760
contender in the generative AI market.

24
00:01:45,760 --> 00:01:52,600
Currently, Google is allowing a select group of companies to test an early version, offering

25
00:01:52,600 --> 00:02:00,000
functionalities that range from chatbots to text summarization and, of course, text generation.

26
00:02:00,000 --> 00:02:06,160
Gemini has broader applications too, so it can assist software engineers with writing

27
00:02:06,160 --> 00:02:09,320
code and generating images as well.

28
00:02:09,320 --> 00:02:16,120
It is rumored to be, or expected to be at least, a multimodal model.

29
00:02:16,120 --> 00:02:22,920
So all of this is going to be delivered through Google's Cloud Vertex AI service.

30
00:02:22,920 --> 00:02:32,640
And the new suite is part of the overall Google strategy to really catch up in the AI realm,

31
00:02:32,640 --> 00:02:39,960
which obviously OpenAI has taken quite a lead ahead of Google in this area.

32
00:02:39,960 --> 00:02:47,060
So Gemini opens up all sorts of possibilities for enhanced customer engagement through chatbots

33
00:02:47,060 --> 00:02:50,320
and content generation as well.

34
00:02:50,320 --> 00:02:58,740
I think it's an exciting space or an exciting story because there was talk of it coming

35
00:02:58,740 --> 00:03:01,320
out in December.

36
00:03:01,320 --> 00:03:03,640
Now it sounds like it's being brought forward.

37
00:03:03,640 --> 00:03:09,960
I think it was initially reported in, we picked it up on, in Reuters.

38
00:03:09,960 --> 00:03:12,760
So it could be coming out very soon.

39
00:03:12,760 --> 00:03:17,920
It would make sense that they would do that alongside Google's Duet AI launch, which has

40
00:03:17,920 --> 00:03:21,160
obviously just come to market.

41
00:03:21,160 --> 00:03:25,320
Rumors are that Meta are entering this space as well.

42
00:03:25,320 --> 00:03:33,240
So Meta is also supposed to be working on a new model that will match or supersede GPT-4.

43
00:03:33,240 --> 00:03:39,880
There was the release of Falcon, which was the 180 billion parameter open source model.

44
00:03:39,880 --> 00:03:42,800
That was the past week or two.

45
00:03:42,800 --> 00:03:49,360
So if you look at this in the round, it would make sense that Google would want to push

46
00:03:49,360 --> 00:03:53,640
ahead with the launch of Gemini and get this out to the market quickly.

47
00:03:53,640 --> 00:04:01,440
Yeah, we're going to look at Google Duet in a bit more detail later, but Bard and other

48
00:04:01,440 --> 00:04:06,600
large language models from Google have lagged behind in terms of the qualities of the outputs

49
00:04:06,600 --> 00:04:13,000
of Claude, ChachiPT, arguably even Llama2.

50
00:04:13,000 --> 00:04:17,880
So they do need to try and be ahead of the pack in terms of pushing the envelope for

51
00:04:17,880 --> 00:04:19,320
the next step, don't they?

52
00:04:19,320 --> 00:04:27,080
So I guess one question that we might consider in mind is, does this feel like a too late

53
00:04:27,080 --> 00:04:29,680
an entry for Google bringing Gemini here?

54
00:04:29,680 --> 00:04:33,960
You know, was it a good idea because they could see how competitors would position themselves

55
00:04:33,960 --> 00:04:35,760
and how tools would be used?

56
00:04:35,760 --> 00:04:41,560
Or do we think this is a misstep being laid to market with their most powerful models?

57
00:04:41,560 --> 00:04:44,840
Yeah, very good question.

58
00:04:44,840 --> 00:04:49,960
I think it all depends on the model, right, and how they bring it to market.

59
00:04:49,960 --> 00:04:53,320
There's a lot of hype and in certain places.

60
00:04:53,320 --> 00:04:59,040
So I saw one Reddit discussion thread that someone did a thought experiment based on

61
00:04:59,040 --> 00:05:06,080
the amount of data that Google has available through all of its, well, through every channel

62
00:05:06,080 --> 00:05:08,200
that it has access to, right?

63
00:05:08,200 --> 00:05:16,240
And they said based on their kind of mental maths of this model, it could be, if they

64
00:05:16,240 --> 00:05:20,880
were to utilise all of that, it could be five times the size of GPT-4.

65
00:05:20,880 --> 00:05:27,760
Now, I don't think Gemini is going to be five times the size of GPT-4.

66
00:05:27,760 --> 00:05:32,680
The compute power would just be like crazy.

67
00:05:32,680 --> 00:05:38,480
But they have the capabilities to create a model and train a model that is bigger than

68
00:05:38,480 --> 00:05:39,480
GPT-4.

69
00:05:39,480 --> 00:05:45,320
So if it is, if they bring something to market and they can deploy it in a way that is accessible,

70
00:05:45,320 --> 00:05:49,100
easy to use, I think it's a real strong play.

71
00:05:49,100 --> 00:05:54,720
If however they play it safe and take the somewhat cautionary approach that they've

72
00:05:54,720 --> 00:06:01,280
taken today, I think Microsoft and OpenAI will continue to steal a march on them.

73
00:06:01,280 --> 00:06:08,760
An interesting observation I saw from the past couple of weeks is that Kassi Kasikov,

74
00:06:08,760 --> 00:06:18,880
the chief decision scientist at Google, she left Google this week and she did a keynote

75
00:06:18,880 --> 00:06:29,040
at MAKON back in July where she spoke about all AI models are built upon the decisions

76
00:06:29,040 --> 00:06:30,800
of people.

77
00:06:30,800 --> 00:06:39,800
And there was insinuation in some of the articles that were published about her departure that

78
00:06:39,800 --> 00:06:43,800
she didn't agree with some of the decisions that Google leadership were making in terms

79
00:06:43,800 --> 00:06:51,920
of how they're going to deploy AI in the future and what they're going to do with it next.

80
00:06:51,920 --> 00:07:00,360
So maybe Google's more reserved approach and more cautious approach is going to be

81
00:07:00,360 --> 00:07:05,200
thrown out of the, I think they're going to put in the bin, get rid of that, because they

82
00:07:05,200 --> 00:07:06,840
need to be more competitive.

83
00:07:06,840 --> 00:07:12,260
And they've seen how disruptive OpenAI has been by giving people access to the models

84
00:07:12,260 --> 00:07:14,320
just like, there you go.

85
00:07:14,320 --> 00:07:19,520
And there's a competitive advantage to just letting people have out it with the models

86
00:07:19,520 --> 00:07:20,520
themselves.

87
00:07:20,520 --> 00:07:22,280
Yeah, I think it's going to be interesting.

88
00:07:22,280 --> 00:07:31,480
You and I have been playing with Google Duet in the 14 day free trial with limited success,

89
00:07:31,480 --> 00:07:32,480
let's say.

90
00:07:32,480 --> 00:07:33,480
That's polite.

91
00:07:33,480 --> 00:07:34,480
You had a take home.

92
00:07:34,480 --> 00:07:35,480
Yeah.

93
00:07:35,480 --> 00:07:38,680
What was your take home about what's going to happen with your Google Duet account when

94
00:07:38,680 --> 00:07:39,680
your trial ends?

95
00:07:39,680 --> 00:07:40,680
Yeah.

96
00:07:40,680 --> 00:07:47,680
So my trial ends in about 24 hours and I'm categorically not renewing that trial.

97
00:07:47,680 --> 00:07:53,640
It's not a product that is worth paying close to what they're trying to charge for at the

98
00:07:53,640 --> 00:07:59,920
moment to the extent that today when I was in a Google Doc and I thought, oh, now that

99
00:07:59,920 --> 00:08:07,080
I've got this Google Doc open, I'll just use the built in, you know, barred text generator

100
00:08:07,080 --> 00:08:09,120
to knock something up for me.

101
00:08:09,120 --> 00:08:10,640
I couldn't find out how to do it.

102
00:08:10,640 --> 00:08:15,720
I couldn't see it on the interface to do the damn thing I wanted to do.

103
00:08:15,720 --> 00:08:19,280
The outputs aren't great in the Gmail composer.

104
00:08:19,280 --> 00:08:22,200
I think they're very meh.

105
00:08:22,200 --> 00:08:31,480
It feels like using GPT-3, which, you know, is fine, but I can probably write a better

106
00:08:31,480 --> 00:08:32,480
email myself.

107
00:08:32,480 --> 00:08:36,680
Everything doesn't need to have this exaggerated tone to it.

108
00:08:36,680 --> 00:08:37,680
I don't know.

109
00:08:37,680 --> 00:08:41,000
I just, the whole thing, I've totally been unimpressed by it.

110
00:08:41,000 --> 00:08:47,760
And until they work out how to deliver on the promise that they showed us in the preview

111
00:08:47,760 --> 00:08:52,800
trailer, this is not a product that is worth investing in at this moment in time.

112
00:08:52,800 --> 00:08:53,800
Yeah.

113
00:08:53,800 --> 00:08:57,880
I mean, we talked in the last episode or a couple of episodes ago about feeling like

114
00:08:57,880 --> 00:09:01,480
they rushed this to market as a half-baked tool.

115
00:09:01,480 --> 00:09:06,560
And I think our tests and the tests of other people that we know would confirm that, right?

116
00:09:06,560 --> 00:09:10,160
The Google Slides integration with it, pretty sucky at the moment.

117
00:09:10,160 --> 00:09:14,120
You could basically just generate images with it, which you could probably get better images

118
00:09:14,120 --> 00:09:16,960
out of Mid-Journey or even PipDrop, right?

119
00:09:16,960 --> 00:09:21,320
On stable stability AI models.

120
00:09:21,320 --> 00:09:24,640
The Gmail is really hard to get anything usable out of.

121
00:09:24,640 --> 00:09:25,640
I would agree with you.

122
00:09:25,640 --> 00:09:28,040
I just don't, it doesn't do what I want it to do.

123
00:09:28,040 --> 00:09:32,840
I want it to know the emails of the thread.

124
00:09:32,840 --> 00:09:36,380
They're already in the thread and then make a sensible recommendation on what the next

125
00:09:36,380 --> 00:09:37,440
email should be.

126
00:09:37,440 --> 00:09:41,480
I want it to know what my email style is and draft things in my style.

127
00:09:41,480 --> 00:09:46,760
That seems to have the whole point of having a generative AI tool baked into where all

128
00:09:46,760 --> 00:09:50,240
your other data sits that the systems know about you.

129
00:09:50,240 --> 00:09:52,480
And none of that's here yet.

130
00:09:52,480 --> 00:09:59,840
The Google Sheets integration to come up with project plan type stuff, very interesting.

131
00:09:59,840 --> 00:10:02,680
Not really that usable on a day-to-day basis.

132
00:10:02,680 --> 00:10:05,520
Kind of interesting, not commercially usable.

133
00:10:05,520 --> 00:10:07,720
Then we can analyze data and other stuff with it.

134
00:10:07,720 --> 00:10:08,720
Yeah, that'll be cool.

135
00:10:08,720 --> 00:10:11,900
So yeah, I'm with you.

136
00:10:11,900 --> 00:10:16,840
If you're thinking about trialing it, maybe do, but I wouldn't pay for it.

137
00:10:16,840 --> 00:10:21,220
I think the other thing is when this Gemini tool comes out, let's say it now gets baked

138
00:10:21,220 --> 00:10:23,840
in to, it's much better.

139
00:10:23,840 --> 00:10:27,080
Maybe it's better than GPT-4 and it's powering Google Duet.

140
00:10:27,080 --> 00:10:30,160
Well, now you've got to go back and trial it again, right?

141
00:10:30,160 --> 00:10:33,200
Because realistically that could be a leap forward in performance.

142
00:10:33,200 --> 00:10:36,560
The other thing is Google has promised things in a number of areas, including the trailer

143
00:10:36,560 --> 00:10:41,020
you mentioned, that there's going to be greater functionality to this.

144
00:10:41,020 --> 00:10:46,440
So we're not recommending you take your eye off of Google or Google Duet and you now put

145
00:10:46,440 --> 00:10:49,320
it to one side and say, that doesn't work for me.

146
00:10:49,320 --> 00:10:52,680
I need a different tool because it's probably going to evolve quite quickly.

147
00:10:52,680 --> 00:10:56,400
But as it stands, yeah, not something to pay for.

148
00:10:56,400 --> 00:10:59,980
Right, let's get into our next story then.

149
00:10:59,980 --> 00:11:02,680
So next story is really interesting one.

150
00:11:02,680 --> 00:11:08,280
The Vanity AI has entered the audio generation market with stable audio, which is a diffusion

151
00:11:08,280 --> 00:11:13,280
model that allows for fine-tune control over generated audio.

152
00:11:13,280 --> 00:11:18,720
So if you go to the stable audio website, you can use it to produce audio files that

153
00:11:18,720 --> 00:11:23,560
have been built based on prompts, text-based prompts that you give them, where you explain

154
00:11:23,560 --> 00:11:29,880
the type of music that you want, how long you want the music to play for, et cetera,

155
00:11:29,880 --> 00:11:30,880
et cetera.

156
00:11:30,880 --> 00:11:38,240
It's really quite impressive in terms of its ability to render high-quality audio pretty

157
00:11:38,240 --> 00:11:46,040
quickly for you to play with it like you would play with mid-journey for images.

158
00:11:46,040 --> 00:11:51,560
You can imagine the stable audio has a range of applications for marketing, from generating

159
00:11:51,560 --> 00:11:58,560
customized background music for your videos, to creating audio cues for live events, to

160
00:11:58,560 --> 00:12:03,240
jingles and other use cases.

161
00:12:03,240 --> 00:12:08,280
And as it stands, its licensing model also encourages you to use it for commercial use

162
00:12:08,280 --> 00:12:13,320
if you are a pro user who's paying, I think it's about $12 a month to access the tool

163
00:12:13,320 --> 00:12:18,480
and you can download things as MP3s, et cetera, et cetera, et cetera.

164
00:12:18,480 --> 00:12:22,400
There are actually, as it stands, three subscription tiers depending on what you need.

165
00:12:22,400 --> 00:12:24,680
There's a free version so you can go play with it.

166
00:12:24,680 --> 00:12:28,140
There's the professional and then there's the enterprise.

167
00:12:28,140 --> 00:12:35,600
So with free, you can generate up to 20 tracks of 45 seconds each per month for non-commercial

168
00:12:35,600 --> 00:12:36,600
use.

169
00:12:36,600 --> 00:12:40,600
And then if you're going to likely be a bit more of a power user, the pro plan will give

170
00:12:40,600 --> 00:12:44,600
you 500 tracks with 90 second duration each.

171
00:12:44,600 --> 00:12:48,760
And importantly, you can use these in commercial projects.

172
00:12:48,760 --> 00:12:53,640
If you've got larger scale and more specialized needs, that's where the enterprise plan comes

173
00:12:53,640 --> 00:12:59,960
in for you to basically negotiate with Stable Audio and the team there for what you need.

174
00:12:59,960 --> 00:13:02,920
Now you've been having a bit of a play with this, haven't you, Martin?

175
00:13:02,920 --> 00:13:06,440
And you've generated some little audio snippets.

176
00:13:06,440 --> 00:13:08,760
So we're going to hear them on the podcast.

177
00:13:08,760 --> 00:13:09,760
Tell us what you've been up to.

178
00:13:09,760 --> 00:13:11,920
Martin So yeah, so they've put a really good guide

179
00:13:11,920 --> 00:13:18,360
together on the Stable Audio website to explain what information went into the training model

180
00:13:18,360 --> 00:13:21,920
and what makes a good prompt really.

181
00:13:21,920 --> 00:13:27,400
So I took some of that content and I put it into ChatGPT and said, hey, there's this new

182
00:13:27,400 --> 00:13:28,400
model.

183
00:13:28,400 --> 00:13:29,400
This is what it's about.

184
00:13:29,400 --> 00:13:30,940
Can you help me create some prompts?

185
00:13:30,940 --> 00:13:32,000
And it did.

186
00:13:32,000 --> 00:13:39,980
So I asked ChatGPT to give me a prompt for Stable Audio that would create an introduction

187
00:13:39,980 --> 00:13:43,520
track or this hodcast.

188
00:13:43,520 --> 00:13:57,560
And here it is.

189
00:13:57,560 --> 00:14:07,920
So that's pretty cool.

190
00:14:07,920 --> 00:14:10,520
Quite electronic, but like very melodic.

191
00:14:10,520 --> 00:14:17,000
If I heard that at the beginning of a podcast, I think it wouldn't strike me to think, oh,

192
00:14:17,000 --> 00:14:19,920
where did they get that terrible sample from?

193
00:14:19,920 --> 00:14:21,880
Or, it sounds awful.

194
00:14:21,880 --> 00:14:26,080
And it even has that little sort of ending, right?

195
00:14:26,080 --> 00:14:28,180
It's the right type of length that you would want.

196
00:14:28,180 --> 00:14:32,520
And it ends with that feeling, which then you can lead into speech.

197
00:14:32,520 --> 00:14:34,360
So that was pretty interesting.

198
00:14:34,360 --> 00:14:36,200
It was very electronic though.

199
00:14:36,200 --> 00:14:39,720
Did you really push for electronic music in your prompt?

200
00:14:39,720 --> 00:14:46,400
Well, the prompt that was input ended up having that in it as a style.

201
00:14:46,400 --> 00:14:53,120
And I think that came from the context given to ChatGPT in the first place about this being

202
00:14:53,120 --> 00:14:54,440
a podcast about AI.

203
00:14:54,440 --> 00:14:59,520
So it's taken that as a kind of reference point and then put that into the Stable Audio

204
00:14:59,520 --> 00:15:00,520
prompt.

205
00:15:00,520 --> 00:15:01,520
Interesting.

206
00:15:01,520 --> 00:15:07,380
Listening to it on my headphones, I noticed I've done quite a lot of music production

207
00:15:07,380 --> 00:15:12,560
over the years in tools like Ableton Live and producing music tracks.

208
00:15:12,560 --> 00:15:18,840
And I did notice listening to it in a decent set of headphones, quite a muddy bass area.

209
00:15:18,840 --> 00:15:24,320
So I can imagine that proper audio files and music producers with much more talent and

210
00:15:24,320 --> 00:15:29,440
knowledge than me would probably have quite a bit to say, I think, about the audio mix.

211
00:15:29,440 --> 00:15:31,880
Like it's not professionally mixed audio.

212
00:15:31,880 --> 00:15:36,320
And to professionally mix audio, you need each of the parts of the audio split out,

213
00:15:36,320 --> 00:15:38,520
which of course you don't really get here.

214
00:15:38,520 --> 00:15:42,600
So I think that could be a bit of pushback from professional communities.

215
00:15:42,600 --> 00:15:48,760
But checking it on our podcast, for example, is something that we're able to craft using

216
00:15:48,760 --> 00:15:51,280
AI would probably be quite a fun thing to do.

217
00:15:51,280 --> 00:15:55,720
But I can imagine there'll be some pushback in those types of cores.

218
00:15:55,720 --> 00:15:57,560
Yeah, absolutely.

219
00:15:57,560 --> 00:16:05,200
And believe me, when I asked it to create a lullaby with loops and soft, gentle sounds

220
00:16:05,200 --> 00:16:10,320
for a baby or a toddler, it was discordant.

221
00:16:10,320 --> 00:16:13,720
It was the lullaby of nightmares.

222
00:16:13,720 --> 00:16:18,000
It was not the sort of thing that you would want to be playing as you put your child to

223
00:16:18,000 --> 00:16:19,080
sleep.

224
00:16:19,080 --> 00:16:21,720
I did try some other techniques with it, though.

225
00:16:21,720 --> 00:16:28,800
I thought, well, if we're creating a podcast intro, I would quite like a bit of audio branding.

226
00:16:28,800 --> 00:16:34,800
So I again asked ChatGPT to give us a prompt that would enable us to get something like

227
00:16:34,800 --> 00:16:43,240
the Netflix Tudum or the Intel Dun Dun Dun Dun audio branding, audio snippet, whatever

228
00:16:43,240 --> 00:16:46,480
you want to call it for the podcast.

229
00:16:46,480 --> 00:16:49,600
And you can listen to the output here.

230
00:16:49,600 --> 00:16:53,000
So that's pretty cool.

231
00:16:53,000 --> 00:16:56,440
Again, it's like it's understood what you were looking for.

232
00:16:56,440 --> 00:16:58,560
It did produce like 30 seconds of audio.

233
00:16:58,560 --> 00:17:01,040
It knew it needed to be a little snippet.

234
00:17:01,040 --> 00:17:06,400
It's interesting in that it's somewhat close to the podcast intro.

235
00:17:06,400 --> 00:17:11,560
Did you specifically say use bits of the same prompt to try and make sure they'd be the

236
00:17:11,560 --> 00:17:12,560
same?

237
00:17:12,560 --> 00:17:13,560
No, not at all.

238
00:17:13,560 --> 00:17:14,720
They were completely separate, actually.

239
00:17:14,720 --> 00:17:22,080
So the fact that it's gone down that path of itself is an interesting observation.

240
00:17:22,080 --> 00:17:24,720
That's quite fun.

241
00:17:24,720 --> 00:17:27,640
Maybe we'll have to try and work both those into the actual podcast over time.

242
00:17:27,640 --> 00:17:28,640
I don't know.

243
00:17:28,640 --> 00:17:29,640
That would be interesting.

244
00:17:29,640 --> 00:17:31,000
If you did a third one, then you might.

245
00:17:31,000 --> 00:17:32,000
What was your third little experiment?

246
00:17:32,000 --> 00:17:33,000
Yeah.

247
00:17:33,000 --> 00:17:37,440
So I thought I'd try something completely off-piste from just related to the podcast

248
00:17:37,440 --> 00:17:43,640
and I asked it to create a background score that we could use in a sales presentation

249
00:17:43,640 --> 00:18:00,680
if I was pitching in HubSpot to a client.

250
00:18:00,680 --> 00:18:09,040
Well, so that's pretty good as well.

251
00:18:09,040 --> 00:18:13,840
It has a different feel to the first two, so I definitely think you'd use the prompt

252
00:18:13,840 --> 00:18:15,480
in a different way.

253
00:18:15,480 --> 00:18:17,520
What were your feelings on that?

254
00:18:17,520 --> 00:18:18,520
Would you use that?

255
00:18:18,520 --> 00:18:23,000
I mean, I'm not sure I would actually use that just from a practical perspective.

256
00:18:23,000 --> 00:18:25,120
I'm not sure where I would play it, but I liked it.

257
00:18:25,120 --> 00:18:26,920
I did actually think that you...

258
00:18:26,920 --> 00:18:31,540
I could imagine a scenario where that was playing in a video or something and it wouldn't

259
00:18:31,540 --> 00:18:33,580
offend me.

260
00:18:33,580 --> 00:18:39,240
I wouldn't think it was out of place in exactly that setting.

261
00:18:39,240 --> 00:18:40,720
Yeah.

262
00:18:40,720 --> 00:18:41,720
I love those.

263
00:18:41,720 --> 00:18:45,040
Thanks for having a play with that, Martin, and sharing those with all of us.

264
00:18:45,040 --> 00:18:46,040
I had a play with the system.

265
00:18:46,040 --> 00:18:49,520
I didn't generate anything, but I listened to some examples on the side that other people

266
00:18:49,520 --> 00:18:56,680
have generated and I did feel like everything either felt a bit artificial, which I guess

267
00:18:56,680 --> 00:19:00,600
you would expect, but given that it was trained on audio, it should be able to replicate that

268
00:19:00,600 --> 00:19:06,040
audio, but everything felt a bit robotic or a bit electronica as a genre to me.

269
00:19:06,040 --> 00:19:10,720
Even the audio snippets I tried to look at that were supposed to be quite guitar type

270
00:19:10,720 --> 00:19:16,480
music, like a Nirvana style, but everything just felt a bit electronica.

271
00:19:16,480 --> 00:19:21,480
So I don't know if that's because of the training data set or whatever, but yeah, it strikes

272
00:19:21,480 --> 00:19:28,040
me that we're having a conversation about the musical equivalent of Dourley 2.

273
00:19:28,040 --> 00:19:31,800
It's like, oh, it's kind of cool, but by the way, it can't do hands.

274
00:19:31,800 --> 00:19:33,120
Sometimes faces looks like gould.

275
00:19:33,120 --> 00:19:34,280
It doesn't know what this is.

276
00:19:34,280 --> 00:19:37,000
So it's like, it does some cool stuff really well.

277
00:19:37,000 --> 00:19:41,560
It clearly has some problems, but the tools are learning to get better from here.

278
00:19:41,560 --> 00:19:42,560
Yeah.

279
00:19:42,560 --> 00:19:48,000
And you've got to just look at the rapid development of the stable diffusion model from when we

280
00:19:48,000 --> 00:19:51,520
first saw it, the big model that they recently released.

281
00:19:51,520 --> 00:19:52,520
Absolutely.

282
00:19:52,520 --> 00:19:53,520
Right.

283
00:19:53,520 --> 00:19:56,280
Let's move on to our next story, Martin.

284
00:19:56,280 --> 00:20:03,600
This is something of an AI fiasco coming to us from Microsoft.

285
00:20:03,600 --> 00:20:11,960
Microsoft's MSN news platform recently stirred controversy by publishing an ill-conceived

286
00:20:11,960 --> 00:20:19,640
and presumably AI generated article that tarnished the reputation of recently deceased NBA player

287
00:20:19,640 --> 00:20:21,680
Brandon Hunter.

288
00:20:21,680 --> 00:20:24,600
Labeled as useless in the article's headline.

289
00:20:24,600 --> 00:20:25,600
I shouldn't laugh.

290
00:20:25,600 --> 00:20:27,560
It's terrible.

291
00:20:27,560 --> 00:20:30,480
It labeled him as useless in the headline.

292
00:20:30,480 --> 00:20:33,960
And then the rest of the article was just incoherent text.

293
00:20:33,960 --> 00:20:41,180
And obviously there was a big backlash from the public on this and lots of people, rightly

294
00:20:41,180 --> 00:20:48,600
so started questioning the use of AI in journalism and the ethics of doing this, particularly

295
00:20:48,600 --> 00:20:51,160
the delicate matters like obituaries.

296
00:20:51,160 --> 00:20:54,080
The idea that this went out is just mind blowing to me.

297
00:20:54,080 --> 00:20:55,520
And this isn't the first time either.

298
00:20:55,520 --> 00:20:59,920
This is not an isolated incident with MSN.

299
00:20:59,920 --> 00:21:09,400
They previously published a questionable article with bizarre travel advice about Ottawa.

300
00:21:09,400 --> 00:21:14,400
And this really looks like a trend that's taken off after MSN laid off a lot of its

301
00:21:14,400 --> 00:21:24,720
human journalism team in 2020 and replaced them with an AI slash human hybrid publishing

302
00:21:24,720 --> 00:21:31,520
and content production system, which doesn't appear to be going swimmingly.

303
00:21:31,520 --> 00:21:37,160
So yeah, with questionable articles, low engagement, and what can only be described as a loss of

304
00:21:37,160 --> 00:21:41,960
editorial integrity, this shift towards automation appears to be backfiring.

305
00:21:41,960 --> 00:21:48,840
So yeah, just an interesting observation and something there about the risks of, well,

306
00:21:48,840 --> 00:21:51,040
the lack of human in the loop.

307
00:21:51,040 --> 00:21:54,440
We've talked about it a lot, right?

308
00:21:54,440 --> 00:21:58,000
While you're out inbound, there was a story last week I mentioned on the podcast about

309
00:21:58,000 --> 00:22:07,640
Gizmodo laying off its Gizmodo en Espanol, their Spanish language website, and that sometimes

310
00:22:07,640 --> 00:22:11,720
the translated articles would just switch to English halfway through.

311
00:22:11,720 --> 00:22:18,600
And I think in this pursuit of scalability through what is perceived to be smart automation,

312
00:22:18,600 --> 00:22:22,400
you remove that human in the loop and you're asking for trouble.

313
00:22:22,400 --> 00:22:29,200
We are nowhere near the point where AI generated content, apps, even at scale, can be just

314
00:22:29,200 --> 00:22:30,600
trusted to just go out.

315
00:22:30,600 --> 00:22:32,720
So you're going to try and scale it, Brian.

316
00:22:32,720 --> 00:22:35,680
You can scale initial production and initial draft.

317
00:22:35,680 --> 00:22:37,920
You've got to have editors checking everything.

318
00:22:37,920 --> 00:22:40,880
Otherwise, you're going to have mishaps like this.

319
00:22:40,880 --> 00:22:44,560
Of course, that's critical for brands, sorry for publishers, but it's also going to be

320
00:22:44,560 --> 00:22:46,640
critical for brands, right?

321
00:22:46,640 --> 00:22:51,440
The notion that somebody hasn't just cast their eye over this before it's gone out.

322
00:22:51,440 --> 00:22:56,960
In the most cursory sense, just a single read through of the article just to say is not

323
00:22:56,960 --> 00:23:02,880
even fact checking, but just calling somebody useless in the headline of an obituary.

324
00:23:02,880 --> 00:23:08,560
I mean, come on, there's some real basics that are overlooked there.

325
00:23:08,560 --> 00:23:12,760
I think that's one of the great dangers that we're going to face over the next six to 12

326
00:23:12,760 --> 00:23:17,360
months as more and more companies look at how they're going to implement AI.

327
00:23:17,360 --> 00:23:23,640
There's going to be a lot of it to boost efficiency and productivity, let's be honest, and having

328
00:23:23,640 --> 00:23:29,680
overconfidence in their belief to automate things in a smart way and requires less human

329
00:23:29,680 --> 00:23:32,960
oversight than is really needed.

330
00:23:32,960 --> 00:23:36,280
I think that's going to be true of every piece of content you could possibly create with

331
00:23:36,280 --> 00:23:40,400
AI support, but you should really also be building.

332
00:23:40,400 --> 00:23:44,120
If you're going to do it, and please don't do it, but if you are going to do it, at least

333
00:23:44,120 --> 00:23:46,780
have some sort of risk profile, right?

334
00:23:46,780 --> 00:23:51,720
Have someone check your obituaries, have someone check content that's more likely to give you

335
00:23:51,720 --> 00:23:58,560
issues around sensitivities, and honestly, for goodness sake, human in the loop, get

336
00:23:58,560 --> 00:24:05,720
your human editors involved because we know that AI's hallucinate, they talk rubbish,

337
00:24:05,720 --> 00:24:12,560
and they're going to produce a best case useless content, worst case offensive content, and

338
00:24:12,560 --> 00:24:14,720
you've got to be releasing that stuff.

339
00:24:14,720 --> 00:24:18,760
Right, next story is from Amazon.

340
00:24:18,760 --> 00:24:22,880
Amazon continues to step up its AI game, but this time it's not for shoppers, it's for

341
00:24:22,880 --> 00:24:23,880
sellers.

342
00:24:23,880 --> 00:24:28,600
So the online giant has launched a set of generative AI tools, and the aim of these

343
00:24:28,600 --> 00:24:32,320
tools is to simplify the process of listing products.

344
00:24:32,320 --> 00:24:38,740
So if you are dropshipping or working in e-commerce and you're using Amazon to sell products,

345
00:24:38,740 --> 00:24:40,360
this is for you.

346
00:24:40,360 --> 00:24:44,680
These new capabilities, essentially, they'll take a brief description from you as the seller,

347
00:24:44,680 --> 00:24:50,400
and then use that to generate complete product titles, bullet points about products, descriptions,

348
00:24:50,400 --> 00:24:52,560
et cetera, et cetera.

349
00:24:52,560 --> 00:24:57,480
With a foundation in large language models, Amazon's generative AI promises to produce

350
00:24:57,480 --> 00:25:02,680
high quality content that not only makes listing easier for sellers, but hopefully also enhances

351
00:25:02,680 --> 00:25:09,080
the shopping experience for customers by ensuring that all products have richer, more comprehensive

352
00:25:09,080 --> 00:25:11,920
product information.

353
00:25:11,920 --> 00:25:16,440
If you are in the e-commerce sector, then we would predict that Amazon's latest innovation

354
00:25:16,440 --> 00:25:18,960
could be a real time saving for you.

355
00:25:18,960 --> 00:25:25,520
But significantly given our last story, also how important it is to make sure that you

356
00:25:25,520 --> 00:25:31,080
are checking the automated descriptions and information that you're getting to make sure

357
00:25:31,080 --> 00:25:33,840
that the information is accurate.

358
00:25:33,840 --> 00:25:35,840
There's also going to be question marks.

359
00:25:35,840 --> 00:25:43,560
We think about how to maintain brand style and brand tone if you're handling over content

360
00:25:43,560 --> 00:25:49,200
generation to generative AI, which we would predict is going to produce a bunch of product

361
00:25:49,200 --> 00:25:52,800
descriptions that all begin to basically sound the same.

362
00:25:52,800 --> 00:25:57,080
If effective marketing is about differentiation, then the last thing that you want is for your

363
00:25:57,080 --> 00:26:01,240
content to sound the same, even when it's things like product descriptions.

364
00:26:01,240 --> 00:26:05,800
So yes, I think it's going to end up saving people a lot of time, but again, it's got to

365
00:26:05,800 --> 00:26:09,720
be used with honor.

366
00:26:09,720 --> 00:26:14,520
What do you think this means for the specialist copywriting services out there?

367
00:26:14,520 --> 00:26:18,200
Do you think they're facing a real existential threat?

368
00:26:18,200 --> 00:26:20,280
It's a good question.

369
00:26:20,280 --> 00:26:27,680
I think if your main services revolve around e-commerce, product listings at scale, then

370
00:26:27,680 --> 00:26:33,880
A, if you're not experimenting with this technology already, then you probably should be because

371
00:26:33,880 --> 00:26:37,960
I'm sure that you are best positioned to figure out where its strengths and weaknesses are

372
00:26:37,960 --> 00:26:43,280
going to be and perhaps how to scale your operations and be more efficient.

373
00:26:43,280 --> 00:26:44,920
But human in the loop is critical.

374
00:26:44,920 --> 00:26:51,160
I think if many companies think they can replace you with this AI, they're going to drift into

375
00:26:51,160 --> 00:26:54,480
the muddy gray middle of everything that sounds the same.

376
00:26:54,480 --> 00:27:00,080
And there are going to be certain products potentially where that's appropriate because

377
00:27:00,080 --> 00:27:03,040
1% of people read the product description anyway, maybe.

378
00:27:03,040 --> 00:27:08,360
But I think for those products where you really want to make sure that your content helps

379
00:27:08,360 --> 00:27:13,360
convince customers on Amazon to buy your product and not one of the competitive products, I

380
00:27:13,360 --> 00:27:18,360
think you're going to struggle to be a good, talented e-commerce copywriter.

381
00:27:18,360 --> 00:27:21,260
That being said, I haven't seen many of the outputs.

382
00:27:21,260 --> 00:27:28,120
Maybe they're using a lot of the writing frameworks and creativity in their ability to pitch products

383
00:27:28,120 --> 00:27:32,880
and write compelling product descriptions that humans would use.

384
00:27:32,880 --> 00:27:37,320
But somehow based on the outputs I've seen from other large language models, I would

385
00:27:37,320 --> 00:27:40,480
doubt that they're going to be better than the true experts.

386
00:27:40,480 --> 00:27:45,000
They'll be better than, if I'm honest, poor copywriters and average copywriters, but they

387
00:27:45,000 --> 00:27:46,360
won't be better than the experts.

388
00:27:46,360 --> 00:27:47,360
I wouldn't say.

389
00:27:47,360 --> 00:27:49,360
What are your thoughts?

390
00:27:49,360 --> 00:27:57,560
Likewise, if you are a copywriter and you're not playing with these tools already, then

391
00:27:57,560 --> 00:28:00,360
the whole thing is an existential threat to you.

392
00:28:00,360 --> 00:28:05,040
You've got to be looking at these tools to enhance what you're doing.

393
00:28:05,040 --> 00:28:09,600
I'm honestly interested from a consumer side as well, the impact of this, because there's

394
00:28:09,600 --> 00:28:13,440
loads of times that I've been on Amazon and looked at a product, a generic product, something

395
00:28:13,440 --> 00:28:18,360
that is some generic label product and the description is terrible.

396
00:28:18,360 --> 00:28:21,880
And it could just be that it's just badly worded English.

397
00:28:21,880 --> 00:28:26,360
It's been written by someone where it isn't their first language and you get this description

398
00:28:26,360 --> 00:28:29,160
and you're like, I don't really know what this means.

399
00:28:29,160 --> 00:28:34,360
If we can get from a consumer perspective, just those kinds of descriptions tidied up

400
00:28:34,360 --> 00:28:35,360
a little bit.

401
00:28:35,360 --> 00:28:41,760
I don't need a crafted copywriter to come in and write this great thing that tells me

402
00:28:41,760 --> 00:28:48,040
how good this de-bobbler is that I'm buying for a few quid on Amazon.

403
00:28:48,040 --> 00:28:50,800
I just needed to tell me some basics.

404
00:28:50,800 --> 00:28:52,720
Then great, I'm all for it.

405
00:28:52,720 --> 00:28:58,280
I'm delighted to see Amazon integrating this technology into their core products offering.

406
00:28:58,280 --> 00:29:02,680
Yeah, and I think there's an element of risk and reward.

407
00:29:02,680 --> 00:29:10,320
If I'm a dropshipper or I'm working in e-commerce and I'm selling 10,000 products on Amazon

408
00:29:10,320 --> 00:29:16,920
and I'm running a lean ship, I haven't got the bandwidth to hand write and hand check

409
00:29:16,920 --> 00:29:20,520
all of the descriptions for all the products, especially when we get into translations as

410
00:29:20,520 --> 00:29:21,520
well.

411
00:29:21,520 --> 00:29:25,680
I think there's an element of if sometimes the product descriptions are inaccurate or

412
00:29:25,680 --> 00:29:31,720
not very good, but they're on average better than nothing, I might take that risk because

413
00:29:31,720 --> 00:29:36,840
I might feel like, well, for 9,500 of the products, it worked really well and it's helping

414
00:29:36,840 --> 00:29:38,160
to sell more of them.

415
00:29:38,160 --> 00:29:46,000
For 500, maybe it got something wrong or the translation wasn't quite right or whatever.

416
00:29:46,000 --> 00:29:47,000
I do think it's a continuum.

417
00:29:47,000 --> 00:29:50,160
I know we've just talked about how you got to check everything and I think there's a

418
00:29:50,160 --> 00:29:54,200
number of different business cases where that is absolutely true, but it's going to come

419
00:29:54,200 --> 00:29:58,960
down to risk and reward, how much time and effort you're going to invest in certain things

420
00:29:58,960 --> 00:30:03,080
versus the improvement in results that you're going to get, and then as a business, just

421
00:30:03,080 --> 00:30:06,040
making the decision that's appropriate for you.

422
00:30:06,040 --> 00:30:11,240
Next story is from our friends at GoCharlie who we've had on the podcast in the past.

423
00:30:11,240 --> 00:30:17,100
So for those that don't know, they are a multimodal generative AI tool for marketers and they

424
00:30:17,100 --> 00:30:21,240
have just bagged $2 million in seed funding.

425
00:30:21,240 --> 00:30:25,360
The investment aims to accelerate the company's mission of becoming the trusted partner for

426
00:30:25,360 --> 00:30:29,520
enterprise customers in the AI enabled marketing space.

427
00:30:29,520 --> 00:30:35,640
Recent guest and GoCharlie's CRO, Brennan Woodratt, told Axios that GoCharlie is making

428
00:30:35,640 --> 00:30:40,480
strides in creating a large language model purpose built for marketers.

429
00:30:40,480 --> 00:30:44,920
The new tool aims to deliver personalized content for brands across multiple platforms,

430
00:30:44,920 --> 00:30:47,720
ensuring a consistent brand voice.

431
00:30:47,720 --> 00:30:52,360
Investors like Goodwater Capital and SRI International are throwing their weight behind GoCharlie,

432
00:30:52,360 --> 00:30:57,920
particularly valuing the startup's technical priorities and IP creation.

433
00:30:57,920 --> 00:31:01,880
In the pipeline, GoCharlie is eyeing a larger funding round next year and they're also

434
00:31:01,880 --> 00:31:04,920
on the lookout for potential M&A targets.

435
00:31:04,920 --> 00:31:09,320
Woodruff expects that by then they'll have some large enterprise deals sealed, making

436
00:31:09,320 --> 00:31:11,600
their company eye on.

437
00:31:11,600 --> 00:31:14,760
And they've really made some significant changes to the tool, haven't they, Myron?

438
00:31:14,760 --> 00:31:16,980
And I know you're a fan of GoCharlie.

439
00:31:16,980 --> 00:31:19,520
So what have you seen and what are your thoughts?

440
00:31:19,520 --> 00:31:24,800
Well following our conversation with Brennan a few weeks ago, they've actually launched

441
00:31:24,800 --> 00:31:30,480
their AI agent, which he teased on the episode.

442
00:31:30,480 --> 00:31:31,480
And it's pretty cool.

443
00:31:31,480 --> 00:31:34,680
I must admit, I've really enjoyed having a play with it.

444
00:31:34,680 --> 00:31:39,720
So it can do things like you can stick some URLs in it and ask it to write some tweets

445
00:31:39,720 --> 00:31:43,800
or turn it into a blog post or something like that, you know, write a thought leadership

446
00:31:43,800 --> 00:31:47,680
piece based on three articles that you found out elsewhere.

447
00:31:47,680 --> 00:31:54,320
But where it really made me go, oh, actually, I really like that, is on Silly Little Things.

448
00:31:54,320 --> 00:32:01,200
You can say to it, create a meme about the difficulties and the ironies of promoting

449
00:32:01,200 --> 00:32:06,560
a podcast about AI and marketing.

450
00:32:06,560 --> 00:32:13,080
And it will come back having used a meme image template and it's written a meme using the

451
00:32:13,080 --> 00:32:17,120
right template, making something quite funny about the thing that you've just said.

452
00:32:17,120 --> 00:32:18,120
It can create GIFs.

453
00:32:18,120 --> 00:32:21,620
So you can tell it to go off and create little GIFs.

454
00:32:21,620 --> 00:32:24,880
It will do campaign planning for you.

455
00:32:24,880 --> 00:32:29,680
Now don't get me wrong, this is not the finished product, right?

456
00:32:29,680 --> 00:32:34,640
There's more work to come on this, absolutely.

457
00:32:34,640 --> 00:32:42,800
But seeing it one week after launch and having a play with it, it could do some things that

458
00:32:42,800 --> 00:32:46,880
were more advanced than I was expecting, to be completely honest.

459
00:32:46,880 --> 00:32:47,880
That's quite cool.

460
00:32:47,880 --> 00:32:52,520
What's the quality of the GIFs like?

461
00:32:52,520 --> 00:32:57,200
So I've only tried it for a couple and it tends to kind of make an image.

462
00:32:57,200 --> 00:33:05,560
They're like slides, so it's almost like it's just flicking between slides on a...

463
00:33:05,560 --> 00:33:06,560
So static images.

464
00:33:06,560 --> 00:33:10,240
Static images, yeah, yeah.

465
00:33:10,240 --> 00:33:14,980
But the kind of thing that you might use for display advertising or something like that.

466
00:33:14,980 --> 00:33:19,960
But the memes, it fully understands the different types of memes that are out there.

467
00:33:19,960 --> 00:33:25,640
I asked it to create one and it used the Spider-Man pointing at each other meme and it used it

468
00:33:25,640 --> 00:33:27,680
in the right context.

469
00:33:27,680 --> 00:33:32,520
I've seen some other examples where it's used, well, other meme templates and it's used them

470
00:33:32,520 --> 00:33:33,520
well.

471
00:33:33,520 --> 00:33:36,160
So yeah, that was pretty cool.

472
00:33:36,160 --> 00:33:38,920
That sounds like quite fun to play with, actually.

473
00:33:38,920 --> 00:33:43,800
That's interesting because I think the other thing with these tools is you might get something

474
00:33:43,800 --> 00:33:46,600
that you can consider production ready.

475
00:33:46,600 --> 00:33:51,480
Like I'm going to provide this on my social profile right now, but it might also just

476
00:33:51,480 --> 00:33:55,600
give you new ideas that you're like, oh yeah, that's a funny way of using the Spider-Man

477
00:33:55,600 --> 00:33:56,600
meme.

478
00:33:56,600 --> 00:33:58,120
But I wouldn't phrase the text quite like that.

479
00:33:58,120 --> 00:34:01,480
And then you as a human, I assume, can go in and then tweet the text to get what you

480
00:34:01,480 --> 00:34:04,240
actually wanted in the first place.

481
00:34:04,240 --> 00:34:09,640
Yeah, so the actual interface for it, I think you'd be better off saying you'd actually

482
00:34:09,640 --> 00:34:15,000
tell it, put this text with the Spider-Man meme and it would go off and do that for you.

483
00:34:15,000 --> 00:34:16,000
Right.

484
00:34:16,000 --> 00:34:19,640
Because there's no way of actually editing the physical meme itself.

485
00:34:19,640 --> 00:34:23,160
But because it's entirely driven through this single prompt interface.

486
00:34:23,160 --> 00:34:24,160
Right.

487
00:34:24,160 --> 00:34:32,280
So that's both useful and maybe if we can edit the memes ourselves, Brennan and Ting,

488
00:34:32,280 --> 00:34:37,480
that might also be quite a fast and helpful and useful way to do that in the future.

489
00:34:37,480 --> 00:34:42,640
I think the way that it's doing it at the moment is the agent is using the API for a

490
00:34:42,640 --> 00:34:44,360
meme generator website.

491
00:34:44,360 --> 00:34:45,360
Right.

492
00:34:45,360 --> 00:34:48,480
I think, I don't know, that's my assumption.

493
00:34:48,480 --> 00:34:50,920
We'd have to get a product like that.

494
00:34:50,920 --> 00:34:54,760
As long as it understands your prompt correctly then, one assumes you're going to get the

495
00:34:54,760 --> 00:34:55,760
type of output you want.

496
00:34:55,760 --> 00:35:00,160
And it's almost like you're asking a, we always call AI an alien intern, don't we?

497
00:35:00,160 --> 00:35:01,760
It's like you're asking an alien intern.

498
00:35:01,760 --> 00:35:03,760
No, I don't want that text.

499
00:35:03,760 --> 00:35:05,560
Go and swap it for this.

500
00:35:05,560 --> 00:35:06,560
Yeah.

501
00:35:06,560 --> 00:35:07,560
Cool.

502
00:35:07,560 --> 00:35:09,560
Interesting stuff.

503
00:35:09,560 --> 00:35:13,760
So if you haven't played with Go Charlie, and it is a pretty cool tool, it's even cooler

504
00:35:13,760 --> 00:35:14,760
than it was before.

505
00:35:14,760 --> 00:35:16,640
So go and have a play.

506
00:35:16,640 --> 00:35:17,640
Right.

507
00:35:17,640 --> 00:35:18,640
Next story.

508
00:35:18,640 --> 00:35:26,440
Yeah, speaking of generative AI, one of the other players who was making waves a few weeks

509
00:35:26,440 --> 00:35:32,120
ago has made an even bigger splash and that is Adobe's Firefly.

510
00:35:32,120 --> 00:35:34,480
They've launched it properly.

511
00:35:34,480 --> 00:35:37,360
It's finally out of beta.

512
00:35:37,360 --> 00:35:42,440
So the tool is now widely available across the entire suite of products, across all of

513
00:35:42,440 --> 00:35:47,200
Creative Cloud, Express and Experience Cloud.

514
00:35:47,200 --> 00:35:53,760
It brings features like Photoshop's generative fill, Illustrator's vector recoloring, and

515
00:35:53,760 --> 00:36:02,760
Express's text to image effects to the wider user base of Creative Cloud products.

516
00:36:02,760 --> 00:36:08,640
So yeah, and now it comes with a commercial license, which is something that people were

517
00:36:08,640 --> 00:36:10,920
hoping to see as well.

518
00:36:10,920 --> 00:36:16,800
Firefly for Enterprise caters for businesses directly, offering a commercially safe environment

519
00:36:16,800 --> 00:36:22,760
by using Adobe stock and public domain content for training.

520
00:36:22,760 --> 00:36:29,160
And obviously that's been a big concern for a lot of users of any generative AI, particularly

521
00:36:29,160 --> 00:36:31,020
in the image space.

522
00:36:31,020 --> 00:36:35,880
Its generative AI capabilities offer a way for businesses to create brand specific content

523
00:36:35,880 --> 00:36:38,160
rapidly and safety.

524
00:36:38,160 --> 00:36:43,440
That even, and this is a really neat tool, provided this content credential, which is

525
00:36:43,440 --> 00:36:52,760
a digital label that showcases essential metadata, which is a big step in supporting the responsible

526
00:36:52,760 --> 00:36:54,800
use of AI generated content.

527
00:36:54,800 --> 00:37:02,480
Effectively, it's a kind of metadata watermark to say how AI was used in the production of

528
00:37:02,480 --> 00:37:04,920
any particular asset.

529
00:37:04,920 --> 00:37:06,400
Yeah.

530
00:37:06,400 --> 00:37:11,600
I mean, we've talked a lot about this tool and I think it's nice that you can use it

531
00:37:11,600 --> 00:37:16,640
now without having to like go download Photoshop, the beta version.

532
00:37:16,640 --> 00:37:20,040
Like that was going to be a bit painful for some people to like, I can't be bothered,

533
00:37:20,040 --> 00:37:24,100
but now this is going to be right there in the tools that you use every day.

534
00:37:24,100 --> 00:37:29,800
They are going to be increasing the price of the software.

535
00:37:29,800 --> 00:37:35,000
So if you're like a Adobe Creative Cloud subscriber, you're going to pay a little bit more for

536
00:37:35,000 --> 00:37:36,000
this.

537
00:37:36,000 --> 00:37:37,000
I don't think it's huge.

538
00:37:37,000 --> 00:37:42,560
You can get it out by like two or two to $5 a month on average, but they'll be increasing

539
00:37:42,560 --> 00:37:46,680
the cost and you can use this on the web.

540
00:37:46,680 --> 00:37:50,720
I think there's going to be like an Adobe Firefly website that you can actually go access

541
00:37:50,720 --> 00:37:57,280
some of the tool without having to be a subscriber, but you're going to pay for credits in essence

542
00:37:57,280 --> 00:38:02,440
a little bit like you do for a lot of the other image generation tools.

543
00:38:02,440 --> 00:38:07,160
I think it's interesting, A, this is cool, but we've talked a lot about Firefly.

544
00:38:07,160 --> 00:38:08,860
Maybe let's talk a bit about pricing, right?

545
00:38:08,860 --> 00:38:11,460
Because they've introduced this now without beta.

546
00:38:11,460 --> 00:38:15,060
If you want to continue to use it, your subscription is going to go up.

547
00:38:15,060 --> 00:38:19,800
One assumes your subscription is going up when you use it on the wall.

548
00:38:19,800 --> 00:38:23,000
That's a potential flash point for customers.

549
00:38:23,000 --> 00:38:28,480
And we talked last week about Zoom introducing AI features, but without changing the subscription

550
00:38:28,480 --> 00:38:33,520
price, and we also said today that we wouldn't pay the additional fee to access Google Do

551
00:38:33,520 --> 00:38:37,960
it, but I continued to test it honestly if it was part of my ongoing subscription.

552
00:38:37,960 --> 00:38:39,840
So what do you think is going to happen?

553
00:38:39,840 --> 00:38:42,200
How's this pricing wall going to play out?

554
00:38:42,200 --> 00:38:46,720
Do we think everybody's going to try and increase their subscription fees to cover the AI parts

555
00:38:46,720 --> 00:38:47,720
of their tools?

556
00:38:47,720 --> 00:38:50,600
Is it going to be a competitive advantage to not do that?

557
00:38:50,600 --> 00:38:53,240
What do you think is going to happen?

558
00:38:53,240 --> 00:39:00,080
The compute power suggests that the economics require people to be paying.

559
00:39:00,080 --> 00:39:05,720
There's a certain point, maybe you give away a few users a month.

560
00:39:05,720 --> 00:39:12,040
Stable Audio, for instance, are giving 20 generations of audio away to the free users

561
00:39:12,040 --> 00:39:14,520
every month.

562
00:39:14,520 --> 00:39:21,600
So I think it makes sense for companies to keep this a free tier option where you get

563
00:39:21,600 --> 00:39:26,880
to sample it, because at some point the cost of compute is going to come down and the capabilities

564
00:39:26,880 --> 00:39:31,760
of these products is going to go up to the point where it just makes sense to absolutely

565
00:39:31,760 --> 00:39:33,000
increase your subscription.

566
00:39:33,000 --> 00:39:37,180
You'll see the value at a certain point.

567
00:39:37,180 --> 00:39:41,480
With Do it AI, the time is not in half.

568
00:39:41,480 --> 00:39:48,260
With the Spirefly tools coming out across the Creative Cloud, I don't live in this suite.

569
00:39:48,260 --> 00:39:54,560
This is not my world at all, but from what I've seen of it and the commentary I've seen,

570
00:39:54,560 --> 00:39:58,640
this actually feels kind of production ready in a lot of instances.

571
00:39:58,640 --> 00:40:03,680
So for that low price point, you know, we're not talking an extra $30 a month like we are

572
00:40:03,680 --> 00:40:04,680
with Duet.

573
00:40:04,680 --> 00:40:14,320
If it's in there, $5, $10 a month, and it's going to enable me to be considerably more

574
00:40:14,320 --> 00:40:17,840
efficient and effective, then I'll pay that.

575
00:40:17,840 --> 00:40:24,520
Is that classic, oh, it's a cup of coffee or it's a pint of beer, whatever your standard

576
00:40:24,520 --> 00:40:25,520
value metric is?

577
00:40:25,520 --> 00:40:28,680
Yeah, I think that's the critical piece.

578
00:40:28,680 --> 00:40:34,280
Shouldn't keep battering Duet, but we're going to keep battering anyway, because it just

579
00:40:34,280 --> 00:40:35,920
doesn't help enough.

580
00:40:35,920 --> 00:40:40,880
It's kind of like an interesting oddity almost, like in terms of its power of use, and it

581
00:40:40,880 --> 00:40:46,480
just makes me want to go back to ChatGPT4 or Claude, honestly, whenever I use it.

582
00:40:46,480 --> 00:40:48,840
So that's why by definition, you wouldn't pay for it.

583
00:40:48,840 --> 00:40:55,700
But I think the dream is we'll pay more for AI tools that genuinely save us quite a bit

584
00:40:55,700 --> 00:41:00,720
of time or that allow us to do creative things that we couldn't do before.

585
00:41:00,720 --> 00:41:02,440
Then there's a value prop, right?

586
00:41:02,440 --> 00:41:06,720
The great belief around selling marketing automation systems is to be able to make a

587
00:41:06,720 --> 00:41:12,920
case for, would you spend $100 extra a month if I saved you 20 hours?

588
00:41:12,920 --> 00:41:17,960
Well, of course you would, and so would your employer, because your salary for those 20

589
00:41:17,960 --> 00:41:22,280
hours is a hell of a lot more than the cost of this marketing automation system.

590
00:41:22,280 --> 00:41:27,800
And I think that's where these tools need to find their space to justify how much you

591
00:41:27,800 --> 00:41:28,800
pay for them.

592
00:41:28,800 --> 00:41:30,520
I mentioned Loom last week.

593
00:41:30,520 --> 00:41:36,120
The AI elements of that are interesting and kind of useful.

594
00:41:36,120 --> 00:41:40,120
Like I'd like not having to write the title of the video on average, Loom got that right.

595
00:41:40,120 --> 00:41:45,200
But often the description was like error strewn and the chapter markers are not necessarily

596
00:41:45,200 --> 00:41:46,760
the chapter markers I would have used.

597
00:41:46,760 --> 00:41:51,080
And so it doesn't save me much time and actually introduces a source of error.

598
00:41:51,080 --> 00:41:55,720
Hence not going to pay for that as an additional, but I'll use it if it's there in terms of

599
00:41:55,720 --> 00:41:56,920
what I'm already paying.

600
00:41:56,920 --> 00:42:00,440
So I think the key here then is utility, isn't it?

601
00:42:00,440 --> 00:42:03,220
If you've got genuine utility, we'll pay for it.

602
00:42:03,220 --> 00:42:08,640
If it's kind of cool, but half-faked, I don't think we're going to.

603
00:42:08,640 --> 00:42:11,480
My next story, it's Coca-Cola.

604
00:42:11,480 --> 00:42:15,920
So Coca-Cola is tempting us with an intriguing proposition.

605
00:42:15,920 --> 00:42:20,840
I don't know if Mr. Saul this story this week, they've released a limited edition drink called

606
00:42:20,840 --> 00:42:27,800
Y3000 that's been developed for a unique blend of AI and human insights.

607
00:42:27,800 --> 00:42:33,520
Coca-Cola has not been upfront about the actual flavor profile for Y3000, as you would probably

608
00:42:33,520 --> 00:42:34,520
imagine.

609
00:42:34,520 --> 00:42:38,760
There's a rumor that there's only ever like six people in the world alive who know the

610
00:42:38,760 --> 00:42:39,760
recipe for Coca-Cola.

611
00:42:39,760 --> 00:42:41,960
Is that, I don't know if that's true or not.

612
00:42:41,960 --> 00:42:46,080
And they're never allowed to travel in the same vehicle at the same time.

613
00:42:46,080 --> 00:42:47,080
Yeah.

614
00:42:47,080 --> 00:42:55,020
So a subset of those, they know the recipe for Y3000 as one assumes so do OpenAI at this

615
00:42:55,020 --> 00:43:00,400
point and are using it to further train their model perhaps, who knows.

616
00:43:00,400 --> 00:43:05,560
So this is a pretty interesting one because Coca-Cola are like tapping into advanced technology

617
00:43:05,560 --> 00:43:08,760
to help further develop their product.

618
00:43:08,760 --> 00:43:13,800
Martin and I often talk about on this show, we really lean into promotions, marketing

619
00:43:13,800 --> 00:43:14,800
and communications.

620
00:43:14,800 --> 00:43:20,200
And if we actually think more broadly in terms of the four P's or the seven P's or the 26

621
00:43:20,200 --> 00:43:23,200
P's of marketing, products are an important part of that.

622
00:43:23,200 --> 00:43:28,120
And this is a great example of AI being used to enhance a product.

623
00:43:28,120 --> 00:43:33,680
At the same time, they used AI to help create the artwork for the can and all this other

624
00:43:33,680 --> 00:43:34,680
stuff.

625
00:43:34,680 --> 00:43:39,480
So really, really interesting use of AI in a number of different areas.

626
00:43:39,480 --> 00:43:46,120
And for marketers, it's kind of like an interesting lesson in the power of using technology as

627
00:43:46,120 --> 00:43:52,120
part of developing your products and then telling the story around those products.

628
00:43:52,120 --> 00:43:59,080
Coca-Cola, of course, are using this to create an even wider story around taste the future,

629
00:43:59,080 --> 00:44:05,260
why 3000 is this is what Coca-Cola could taste like in the year 3000 because AI is developing

630
00:44:05,260 --> 00:44:07,800
all the products in the world and all that good stuff.

631
00:44:07,800 --> 00:44:13,200
I guess the question here, Martin, is this kind of like a marketing gimmick or do we

632
00:44:13,200 --> 00:44:20,460
think we're going to see AI being used more in combined product development and promotion?

633
00:44:20,460 --> 00:44:24,320
In the short term, yes, marketing gimmick and I am all for it.

634
00:44:24,320 --> 00:44:29,520
I am leaning into this marketing gimmick and we are sat here promoting it on the show.

635
00:44:29,520 --> 00:44:31,200
They've made a bit of a story on it.

636
00:44:31,200 --> 00:44:32,280
Yeah, I love it.

637
00:44:32,280 --> 00:44:33,680
And I like to see these stories.

638
00:44:33,680 --> 00:44:41,720
I want to see brands doing cool stuff with their product development using AI.

639
00:44:41,720 --> 00:44:43,080
In the longer term, yeah, absolutely.

640
00:44:43,080 --> 00:44:51,320
I think it will be because all of us are going to be using AI tools all of the time.

641
00:44:51,320 --> 00:44:56,720
So it's going to be part of the product manager's toolkit.

642
00:44:56,720 --> 00:45:03,240
I don't think it's going to make a splash in terms of a story like this one, Will, but

643
00:45:03,240 --> 00:45:07,800
we're going to see AI as part of the product development process because it's good at ideation.

644
00:45:07,800 --> 00:45:13,760
It's good at helping come up with ways around.

645
00:45:13,760 --> 00:45:18,960
It is quite creative, much as people don't think it is creative, it is quite creative.

646
00:45:18,960 --> 00:45:21,960
So it can be used in brainstorming sessions.

647
00:45:21,960 --> 00:45:26,640
So yes, AI for product development is absolutely something we're going to have more of in the

648
00:45:26,640 --> 00:45:27,640
future.

649
00:45:27,640 --> 00:45:34,160
Yeah, if I reflect on this and I'm a product manager, I'm probably like, if you're a product

650
00:45:34,160 --> 00:45:37,560
manager and you're listening to this, I think it's a very interesting thought experiment

651
00:45:37,560 --> 00:45:47,560
to think about how can I leverage AI and the hype train around AI to bake AI in a cool

652
00:45:47,560 --> 00:45:53,120
and interesting way into my product and then tell an interesting story about it.

653
00:45:53,120 --> 00:45:58,280
At the same time, how can I do that in a way that genuinely is interesting and doesn't

654
00:45:58,280 --> 00:45:59,280
seem shoehorned?

655
00:45:59,280 --> 00:46:04,320
Because I think you can also get this massively wrong and it's another case of, oh, guess

656
00:46:04,320 --> 00:46:06,440
what, another AI first product.

657
00:46:06,440 --> 00:46:10,960
My shoe allows me to jump further and faster and I'm better at basketball because it's

658
00:46:10,960 --> 00:46:11,960
got AI in it.

659
00:46:11,960 --> 00:46:16,960
Oh, guess what, this chewing gum lasts 5% longer when I stick it on the bedpost overnight

660
00:46:16,960 --> 00:46:19,160
because it's got AI in it.

661
00:46:19,160 --> 00:46:21,400
Like at some point, that's just going to become too much.

662
00:46:21,400 --> 00:46:27,260
But what Coca-Cola have done here is they've made it all feel relevant.

663
00:46:27,260 --> 00:46:32,080
Like it was not just a PR stunt, but that actually it was quite a cool and interesting

664
00:46:32,080 --> 00:46:33,920
thing to try with their product.

665
00:46:33,920 --> 00:46:37,280
And then they thought, oh, we can tell a brand story around this and we can actually make

666
00:46:37,280 --> 00:46:42,600
AI and generative AI part of the marketing promotion strategy, help them create some

667
00:46:42,600 --> 00:46:45,040
of the content and all those types of things.

668
00:46:45,040 --> 00:46:48,880
So maybe a good thought experiment for the product managers out there.

669
00:46:48,880 --> 00:46:54,320
And if you're more of a marketing communications pro, book in a meeting with one of the product

670
00:46:54,320 --> 00:46:59,600
managers, get ChatGPT open and just start brainstorming between the three of you how

671
00:46:59,600 --> 00:47:07,480
you might leverage this boom in AI to enhance your products and tell great stories about.

672
00:47:07,480 --> 00:47:11,160
Right, next story, Martin.

673
00:47:11,160 --> 00:47:18,760
EY, formerly Ernst & Young has announced a massive, is massive, $1.4 billion investment

674
00:47:18,760 --> 00:47:20,560
into AI technologies.

675
00:47:20,560 --> 00:47:27,680
And they're taking aim at the burgeoning market for AI embedded consulting services.

676
00:47:27,680 --> 00:47:34,760
So this initiative places EY in direct competition with other consulting giants like KPMG and

677
00:47:34,760 --> 00:47:42,320
Accenture, all of whom are investing heavily in AI as the next frontier for business growth.

678
00:47:42,320 --> 00:47:49,720
The newly unveiled EY.AI platform, I said that with no complication, definitely didn't

679
00:47:49,720 --> 00:47:51,640
trip over it, absolutely nailed it.

680
00:47:51,640 --> 00:47:52,640
Chef's kiss.

681
00:47:52,640 --> 00:47:53,640
He knew.

682
00:47:53,640 --> 00:47:58,160
EY.AI, you can't do it twice in a row.

683
00:47:58,160 --> 00:48:00,520
That's just ridiculous.

684
00:48:00,520 --> 00:48:07,120
The new platform comes with AI infused versions of EY's existing product suite, including

685
00:48:07,120 --> 00:48:13,040
EY Fabric, which is a data management tool already being used by their 60,000 clients.

686
00:48:13,040 --> 00:48:18,440
So their intent is clear and it's really to integrate AI into every facet of their business

687
00:48:18,440 --> 00:48:20,880
operations.

688
00:48:20,880 --> 00:48:30,640
Apparently they're also looking to release their own large language model called EY.AIEYQ.

689
00:48:30,640 --> 00:48:34,600
Get the brand team on that.

690
00:48:34,600 --> 00:48:37,080
But there's a lot of work to do there on the front.

691
00:48:37,080 --> 00:48:40,440
You've done spectacularly, Martin, to keep getting that right.

692
00:48:40,440 --> 00:48:42,720
It's 5.30 PM on a Friday.

693
00:48:42,720 --> 00:48:46,440
We should arguably be in the pub and you're trying to get your head around that.

694
00:48:46,440 --> 00:48:47,840
EY, you've got to help us out.

695
00:48:47,840 --> 00:48:49,800
We need a snappy name for that.

696
00:48:49,800 --> 00:48:53,560
We'll ask TrackGPT for some snappy names and we'll email them over.

697
00:48:53,560 --> 00:48:54,560
Yeah.

698
00:48:54,560 --> 00:49:01,080
So this sounds like it's going to be used for an internal tool for EY staff, but who

699
00:49:01,080 --> 00:49:03,320
knows, maybe this will be rolled out to their customers.

700
00:49:03,320 --> 00:49:05,240
No news on that at the moment.

701
00:49:05,240 --> 00:49:11,680
They've secured an early access partnership with OpenAI's capabilities through Microsoft

702
00:49:11,680 --> 00:49:18,080
Azure, including GPT-3 and GPT-4, and they formed some partnerships with tech giants

703
00:49:18,080 --> 00:49:21,400
like Dell, SAP, and Thomson Reuters.

704
00:49:21,400 --> 00:49:31,040
I think what we're seeing here is just another step by all of the big consultancies to jump

705
00:49:31,040 --> 00:49:33,280
onto the AI hype train.

706
00:49:33,280 --> 00:49:36,080
Deloitte have done it, McKinsey are doing it.

707
00:49:36,080 --> 00:49:40,520
We've seen their KPMG and Accenture are doing it.

708
00:49:40,520 --> 00:49:46,560
It's becoming integral and big companies are clearly going to these consultancies saying,

709
00:49:46,560 --> 00:49:48,720
hey, we need help in this space.

710
00:49:48,720 --> 00:49:53,320
So they've got to be making a lot of noise saying, we know what we're talking about.

711
00:49:53,320 --> 00:49:55,040
Look at these investments that we're making.

712
00:49:55,040 --> 00:49:56,040
We're on it.

713
00:49:56,040 --> 00:49:57,040
You can trust us.

714
00:49:57,040 --> 00:49:58,040
We're ahead of the game.

715
00:49:58,040 --> 00:49:59,040
I agree.

716
00:49:59,040 --> 00:50:04,240
I think there's a question here, right, which is, is AI embedded consultancy service, is

717
00:50:04,240 --> 00:50:06,440
that the norm pretty much from here on out?

718
00:50:06,440 --> 00:50:08,320
I mean, what do you think about that, Ryan?

719
00:50:08,320 --> 00:50:12,240
Well, I think it is because AI is going to be baked into everything.

720
00:50:12,240 --> 00:50:13,800
So you've got to speak that language.

721
00:50:13,800 --> 00:50:19,580
You've got to know and understand what AI is in the same way that these big companies

722
00:50:19,580 --> 00:50:25,880
for the past decade or so, the buzzword has been digital transformation.

723
00:50:25,880 --> 00:50:28,640
That's what they've been doing.

724
00:50:28,640 --> 00:50:33,040
Deloitte Digital has had loads of thought leadership pieces on this.

725
00:50:33,040 --> 00:50:37,920
You go on the McKinsey website and it's talking about how digital supports growth.

726
00:50:37,920 --> 00:50:40,960
It's going to shift from digital to AI.

727
00:50:40,960 --> 00:50:42,960
That's what we're about to see.

728
00:50:42,960 --> 00:50:43,960
Absolutely.

729
00:50:43,960 --> 00:50:49,760
In fact, what an opportunity for us to talk about our AI transformation services offered

730
00:50:49,760 --> 00:50:55,160
by the team at Heart's Officially Intelligent Marketing to help you bring AI to your organization.

731
00:50:55,160 --> 00:50:57,000
It weren't called AI transformation services.

732
00:50:57,000 --> 00:51:01,880
We were just offering training and consultancy, but I like that, Mahin, and I think we should

733
00:51:01,880 --> 00:51:02,880
TM it.

734
00:51:02,880 --> 00:51:06,400
It's probably been TN'd already and bought it on the website anyway.

735
00:51:06,400 --> 00:51:07,400
I agree with you.

736
00:51:07,400 --> 00:51:11,200
There's a number of aspects to this.

737
00:51:11,200 --> 00:51:17,720
I can imagine a future honestly where you can pay to access a McKinsey bot that is your

738
00:51:17,720 --> 00:51:26,320
first call for questions about your business and you can get a form of guidance through

739
00:51:26,320 --> 00:51:28,120
the bot and you subscribe to it.

740
00:51:28,120 --> 00:51:36,120
But as we know, digital transformation, implementations of CRMs, implementing AI in your business,

741
00:51:36,120 --> 00:51:39,800
it's not just about having experts help you know what to do.

742
00:51:39,800 --> 00:51:44,320
It's the change management and the training that comes with actually embedding that in

743
00:51:44,320 --> 00:51:52,920
your organization, which I think the humans at McKinsey and at EY and people like ourselves

744
00:51:52,920 --> 00:51:56,720
are going to be doing for quite a while because bots are not going to be able to do that because

745
00:51:56,720 --> 00:52:01,880
that's going to be about working with people and bringing people along on the journey,

746
00:52:01,880 --> 00:52:06,560
inspiring them about what's possible, enabling them through training, etc.

747
00:52:06,560 --> 00:52:10,720
Right, last story this week, we are talking MailChimp.

748
00:52:10,720 --> 00:52:17,840
So MailChimp, which was acquired by Intuit not that long ago for crazily large money,

749
00:52:17,840 --> 00:52:22,880
is bringing AI to its suite of tools.

750
00:52:22,880 --> 00:52:29,680
So it's basically baking in Intuit Assist, which is a AI chatbot style tool that Intuit

751
00:52:29,680 --> 00:52:33,920
has obviously been developing, but now it's going to be made available if you are a subscriber

752
00:52:33,920 --> 00:52:36,440
to and user of MailChimp.

753
00:52:36,440 --> 00:52:37,840
Why is this interesting?

754
00:52:37,840 --> 00:52:41,720
Because it's bringing some of that power to small and mid-sized businesses, right?

755
00:52:41,720 --> 00:52:46,040
MailChimp has always been a fantastic tool for smaller companies to use if they didn't

756
00:52:46,040 --> 00:52:51,540
have the budget to invest in more expensive tools like, let's say, Salesforce and the

757
00:52:51,540 --> 00:52:55,040
higher tier options of a product like HubSpot.

758
00:52:55,040 --> 00:52:56,200
So that's kind of the goal here.

759
00:52:56,200 --> 00:53:02,600
They want to be able to bring generative AI to help those SMBs be more efficient.

760
00:53:02,600 --> 00:53:07,680
And the tool promises to enhance marketing journeys, help customers better understand

761
00:53:07,680 --> 00:53:14,520
customer desires, and provide a host of features from connected e-commerce integration to inline

762
00:53:14,520 --> 00:53:17,160
email editing.

763
00:53:17,160 --> 00:53:21,040
Intuit Assist is geared to help businesses generate content, tailor their marketing strategies

764
00:53:21,040 --> 00:53:25,320
and quickly learn from their campaigns, and offers capabilities like real-time content

765
00:53:25,320 --> 00:53:29,960
generation for new product campaigns and even multilingual copy translation.

766
00:53:29,960 --> 00:53:35,360
This tool is integrated seamlessly into the MailChimp ecosystem, so it should be pretty

767
00:53:35,360 --> 00:53:40,280
easy for small businesses to use without having to break the bank.

768
00:53:40,280 --> 00:53:47,260
So this is really cool because it's bringing really significant generative AI power to

769
00:53:47,260 --> 00:53:51,760
smaller businesses and making it much easier for them to leverage.

770
00:53:51,760 --> 00:53:52,760
What do you think, Mark?

771
00:53:52,760 --> 00:53:58,320
Do you think that MailChimp's Intuit Assist is going to democratize access to AI in the

772
00:53:58,320 --> 00:54:04,760
SMB space, or is it not just about having access to technologies, training and understanding

773
00:54:04,760 --> 00:54:09,840
what the tools and stuff can do, still going to be a bit of a barrier, do you think?

774
00:54:09,840 --> 00:54:16,600
This story really caught my eye because I'm a QuickBooks user, and Intuit Assist has also

775
00:54:16,600 --> 00:54:19,600
been integrated into QuickBooks.

776
00:54:19,600 --> 00:54:24,520
And apparently that is allowing QuickBooks users to automate tasks like drafting invoices

777
00:54:24,520 --> 00:54:32,280
and drafting invoice reminders and managing reports and all sorts of things like personalized

778
00:54:32,280 --> 00:54:38,560
recommendations and helping on a bunch of tasks that as a small business owner, you

779
00:54:38,560 --> 00:54:44,420
just get swamped under or you might push under the carpet and forget about.

780
00:54:44,420 --> 00:54:47,960
So as a QuickBooks user, I was quite excited to see that it's being rolled out to that

781
00:54:47,960 --> 00:54:48,960
as well.

782
00:54:48,960 --> 00:54:55,720
I think it is going to be a big benefit to smaller mid-sized business owners.

783
00:54:55,720 --> 00:55:00,880
In the same way that HubSpot is baking in the generative AI tech and just making it

784
00:55:00,880 --> 00:55:06,600
really easy to use in the tools that you're already using every day, it sounds like that's

785
00:55:06,600 --> 00:55:08,120
what's going to happen with MailChimp.

786
00:55:08,120 --> 00:55:15,360
So I'm expecting, and the devil is in the detail, the detail here being the user experience,

787
00:55:15,360 --> 00:55:17,400
how do they build it into the workflow?

788
00:55:17,400 --> 00:55:22,720
If I'm creating an email marketing campaign, does the generative AI do the whole thing

789
00:55:22,720 --> 00:55:29,280
for me in like a wizard based editor or are they adding in a tool that will write me a

790
00:55:29,280 --> 00:55:34,220
smart subject line where I'm typically writing the subject line?

791
00:55:34,220 --> 00:55:37,040
How does it work?

792
00:55:37,040 --> 00:55:43,720
I think it's going to be a big benefit to people looking to, well, anyone that doesn't

793
00:55:43,720 --> 00:55:51,120
know how to use AI and is interested in it, having this capability baked into that platform

794
00:55:51,120 --> 00:55:54,440
is going to obviously increase AI adoption.

795
00:55:54,440 --> 00:56:03,080
Yeah, it strikes me that the race to embed AI in all of the tools we already use is on.

796
00:56:03,080 --> 00:56:06,220
I think that will fizzle out quite quickly.

797
00:56:06,220 --> 00:56:13,760
It could become table stakes reasonably quickly that you expect some sort of support agent

798
00:56:13,760 --> 00:56:18,080
to be able to do things for you based on natural language prompts to help you create content,

799
00:56:18,080 --> 00:56:23,880
to answer questions about your data and help you do data analysis, to smart automate certain

800
00:56:23,880 --> 00:56:25,920
tasks.

801
00:56:25,920 --> 00:56:28,040
And then user experience becomes key, right?

802
00:56:28,040 --> 00:56:30,400
Who makes that the easiest to do?

803
00:56:30,400 --> 00:56:36,000
Assuming that the underlying technology, the large language models that power this ability

804
00:56:36,000 --> 00:56:41,240
to understand natural language and take action based on it that these bots will have is basically

805
00:56:41,240 --> 00:56:42,800
the same, right?

806
00:56:42,800 --> 00:56:47,800
If that's always the same, then it's about user experience and how easy they are to use

807
00:56:47,800 --> 00:56:52,320
and having them in the place where you already do stuff you already do, like Mailchimp.

808
00:56:52,320 --> 00:56:56,440
If you're a Mailchimp user or Microsoft Word or HubSpot or any other platform that you've

809
00:56:56,440 --> 00:56:57,440
got.

810
00:56:57,440 --> 00:57:05,400
So it might be interesting for us to look back at the end of 2024 when every piece of

811
00:57:05,400 --> 00:57:10,080
software's little chat icon that everybody has in the bottom right that basically will

812
00:57:10,080 --> 00:57:15,580
help you answer simple questions or sends a message to a human in customer support is

813
00:57:15,580 --> 00:57:18,400
in essence not that at all anymore.

814
00:57:18,400 --> 00:57:24,760
It's a LLM powered conversational agent that doesn't involve any humans that helps you

815
00:57:24,760 --> 00:57:29,760
do a heck of a lot more in the software that you're using than you were able to do before

816
00:57:29,760 --> 00:57:35,360
all powered in an interface that's kind of similar to the kind of chat, web chat interfaces

817
00:57:35,360 --> 00:57:38,680
that we already use today, but just powered in a different way.

818
00:57:38,680 --> 00:57:47,880
Yeah, and before the end of 2024, I'd say by the end of 2023, saying that your software

819
00:57:47,880 --> 00:57:54,520
product has AI capabilities is going to be so passe, it's like saying that you've got

820
00:57:54,520 --> 00:57:55,520
a website.

821
00:57:55,520 --> 00:57:56,520
Absolutely.

822
00:57:56,520 --> 00:58:03,080
So there you have it folks, that's what's been going on in the world of AI this week.

823
00:58:03,080 --> 00:58:06,840
As you know, we deliberately focus on all the stories that we think are going to be

824
00:58:06,840 --> 00:58:08,400
the most important for marketers.

825
00:58:08,400 --> 00:58:14,640
So there's been a bunch of stories this week like all the key industry bots have been meeting,

826
00:58:14,640 --> 00:58:19,440
having important meetings in Washington about the ethics of AI and all that good stuff.

827
00:58:19,440 --> 00:58:22,720
It's not that that stuff's not important or we recognize it's important, but we do like

828
00:58:22,720 --> 00:58:28,560
to focus as much as we can on the practical aspects of AI that are going to be impacting

829
00:58:28,560 --> 00:58:32,840
the workflows of marketers and small business owners across the world.

830
00:58:32,840 --> 00:58:35,680
If you like this podcast, then please subscribe.

831
00:58:35,680 --> 00:58:40,020
If you really like it, tell your friends about it, especially if you think they find it interesting

832
00:58:40,020 --> 00:58:42,400
or they would benefit.

833
00:58:42,400 --> 00:58:43,680
Thanks as always for your time, Martin.

834
00:58:43,680 --> 00:58:46,120
I will look forward to catching up with you next week.

835
00:58:46,120 --> 00:59:03,720
Speak to you next week, Paul.