1
00:00:00,000 --> 00:00:07,560
Welcome to Artificially Intelligent Marketing, a weekly podcast where we stay on top of the
2
00:00:07,560 --> 00:00:13,300
latest trends, tips and tools in the world of marketing AI, helping you get the best
3
00:00:13,300 --> 00:00:16,120
results from your marketing efforts.
4
00:00:16,120 --> 00:00:24,640
Now let's join our hosts, Paul Avery and Martin Broadhurst.
5
00:00:24,640 --> 00:00:32,120
Welcome to Artificially Intelligent Marketing, a podcast for marketers who need to know about
6
00:00:32,120 --> 00:00:33,120
AI.
7
00:00:33,120 --> 00:00:36,600
I'm joined with my co-host here today, Martin Broadhurst.
8
00:00:36,600 --> 00:00:38,720
Hi Martin, how are you doing?
9
00:00:38,720 --> 00:00:40,160
I'm good, thank you.
10
00:00:40,160 --> 00:00:45,920
Recovering from the misery that was watching my football club, Derby County, get trashed
11
00:00:45,920 --> 00:00:49,200
by Barnsley 4-1 at the weekend.
12
00:00:49,200 --> 00:00:52,080
So the week has progressively got better, I must admit.
13
00:00:52,080 --> 00:00:55,280
There we go, we're going to have a podcast where we're going to talk a little bit about
14
00:00:55,280 --> 00:01:00,880
AI marketing and quite a lot about just how difficult and painful it is for Martin to
15
00:01:00,880 --> 00:01:02,720
be a Derby County fan.
16
00:01:02,720 --> 00:01:09,520
Myself, I was in San Diego this week, sunny San Diego, not as sunny as usual, pretty darn
17
00:01:09,520 --> 00:01:14,960
cold to be honest, at a drug discovery conference, which is actually rather awesome.
18
00:01:14,960 --> 00:01:19,240
I just stepped off a plane a couple of hours ago and walked into my house 15 minutes ago
19
00:01:19,240 --> 00:01:20,680
to do this podcast with Martin.
20
00:01:20,680 --> 00:01:26,280
So if I make even less sense than I normally do, Martin, that is going to be my excuse
21
00:01:26,280 --> 00:01:27,680
at least for this week.
22
00:01:27,680 --> 00:01:32,280
And then when we do next week's podcast, I'll obviously have to come up with a new one.
23
00:01:32,280 --> 00:01:36,840
Right, Martin, why don't we give people just a feel for who we are seeing as this is the
24
00:01:36,840 --> 00:01:42,880
first episode and then tell them what they can expect if they subscribe to this hopefully
25
00:01:42,880 --> 00:01:45,840
useful but somewhat full of nonsense podcast.
26
00:01:45,840 --> 00:01:50,920
Yeah, well, so I introduce myself first, Derby County fan, as you already know.
27
00:01:50,920 --> 00:01:52,720
But yeah, my name's Martin Broadhurst.
28
00:01:52,720 --> 00:02:00,520
I run a small consultancy advising companies on sales and marketing technology, so CRM
29
00:02:00,520 --> 00:02:01,520
marketing automation.
30
00:02:01,520 --> 00:02:06,040
And in recent years, I've been doing a lot more on the artificial intelligence front
31
00:02:06,040 --> 00:02:07,040
as well.
32
00:02:07,040 --> 00:02:12,780
Regularly speaking at conferences about marketing AI and recently published a paper in the Applied
33
00:02:12,780 --> 00:02:21,200
Marketing Analytics Journal about how businesses of any size can deploy AI in their marketing.
34
00:02:21,200 --> 00:02:27,280
So that's a kind of flavor actually, what you can expect, the theme of that paper is
35
00:02:27,280 --> 00:02:33,080
how can businesses deploy AI in their business from a marketing and sales perspective?
36
00:02:33,080 --> 00:02:37,000
That's very much what we're going to be talking about, looking at the latest technologies.
37
00:02:37,000 --> 00:02:42,640
You'll be hearing all the familiar names like ChattGPT, OpenAI and the big players, but
38
00:02:42,640 --> 00:02:48,840
also looking at the startups, some of the emerging technologies as well.
39
00:02:48,840 --> 00:02:49,840
Good stuff.
40
00:02:49,840 --> 00:02:52,680
Yeah, I'm really excited to get into some of that with you today, Martin.
41
00:02:52,680 --> 00:02:57,080
I know we've got some great stories this week to look at and a really interesting tool for
42
00:02:57,080 --> 00:02:58,640
tool of the week.
43
00:02:58,640 --> 00:03:00,280
Hi, everyone.
44
00:03:00,280 --> 00:03:01,400
My name is Paul Avery.
45
00:03:01,400 --> 00:03:06,280
I am CEO of a life science marketing agency called Biostrata.
46
00:03:06,280 --> 00:03:09,520
But the only real things you need to know about me in the context is this is I am a
47
00:03:09,520 --> 00:03:10,520
nerd.
48
00:03:10,520 --> 00:03:11,640
I am a science nerd.
49
00:03:11,640 --> 00:03:13,680
I am a computer nerd.
50
00:03:13,680 --> 00:03:15,760
I am an AI nerd.
51
00:03:15,760 --> 00:03:17,800
And I just love diving into all this stuff.
52
00:03:17,800 --> 00:03:19,400
I do it for fun.
53
00:03:19,400 --> 00:03:23,040
Martin and I are constantly chatting over WhatsApp and on the phone.
54
00:03:23,040 --> 00:03:26,560
And we thought, you know what, all this stuff we're reading about, all this stuff we're
55
00:03:26,560 --> 00:03:30,720
learning, I wonder if we can actually just help save some other people time and keep
56
00:03:30,720 --> 00:03:34,120
them up to date on what's happening in the world of marketing AI by doing all the heavy
57
00:03:34,120 --> 00:03:39,080
lifting for them of sifting through the stories to figuring out what is the most important
58
00:03:39,080 --> 00:03:43,520
thing for us all to be knowing and hearing about this week, but also playing with those
59
00:03:43,520 --> 00:03:47,960
different tools to get a feel for what they can do, what their strengths are, what the
60
00:03:47,960 --> 00:03:54,120
best applications are so that you as a listener can stay on top of what tools and tips and
61
00:03:54,120 --> 00:03:58,280
tricks are out there for you to get the best out of using AI for marketing, doing that
62
00:03:58,280 --> 00:04:00,280
work so that you don't have to.
63
00:04:00,280 --> 00:04:01,840
Cool, cool.
64
00:04:01,840 --> 00:04:06,840
Let's chuck ourselves straight into story number one then, Martin.
65
00:04:06,840 --> 00:04:13,600
Why don't you tell us about this fairly big deal with the release of ChatGPT and Whisper
66
00:04:13,600 --> 00:04:14,600
APIs?
67
00:04:14,600 --> 00:04:21,800
Yeah, so anyone listening to this podcast will likely know that ChatGPT was launched
68
00:04:21,800 --> 00:04:30,560
in November, caused a bit of a stir and it wasn't in and of itself new tech in the purest
69
00:04:30,560 --> 00:04:31,560
sense.
70
00:04:31,560 --> 00:04:36,280
It was built upon GPT-3 that had been around for a couple of years and lots of developers
71
00:04:36,280 --> 00:04:39,400
had been integrating into their tool stack.
72
00:04:39,400 --> 00:04:46,280
But because of the way that the interface was just like using iMessage or WhatsApp,
73
00:04:46,280 --> 00:04:49,840
it just boomed and was a massive success.
74
00:04:49,840 --> 00:04:55,480
They launched ChatGPT Pro, or plus I can never remember which one it is, a couple of weeks
75
00:04:55,480 --> 00:04:56,480
ago.
76
00:04:56,480 --> 00:05:02,560
But now developers can actually start using it in their apps, which is massive.
77
00:05:02,560 --> 00:05:10,880
Yeah, really big deal, they've announced that ChatGPT API is available, as is Whisper, which
78
00:05:10,880 --> 00:05:13,600
is a completely different model.
79
00:05:13,600 --> 00:05:17,720
And it's a transcription tool, really.
80
00:05:17,720 --> 00:05:21,680
So it will take audio and turn it into text and there's different things that you can
81
00:05:21,680 --> 00:05:23,280
do with that.
82
00:05:23,280 --> 00:05:30,640
But that's been quite a challenging AI to deploy.
83
00:05:30,640 --> 00:05:34,440
Lots of developers have said it was quite tricky.
84
00:05:34,440 --> 00:05:37,320
And now they've said, it's fine, we'll deal with it.
85
00:05:37,320 --> 00:05:41,080
We'll make it available via API.
86
00:05:41,080 --> 00:05:42,520
So basically crack on.
87
00:05:42,520 --> 00:05:44,880
So yeah, very exciting stuff.
88
00:05:44,880 --> 00:05:50,780
So this is pretty interesting because there's already been an explosion of GPT-driven tools
89
00:05:50,780 --> 00:05:51,780
for writing.
90
00:05:51,780 --> 00:05:58,440
I've got Writer, Jasper, and a huge host of all of those types of tools.
91
00:05:58,440 --> 00:06:03,480
We already have Otter and a bunch of other transcription tools, transcription being baked
92
00:06:03,480 --> 00:06:08,200
into many different even web conferencing software like Microsoft Teams.
93
00:06:08,200 --> 00:06:12,800
What I think was really interesting here is that reduction in cost.
94
00:06:12,800 --> 00:06:19,240
So any of these apps that use OpenAI's infrastructure through the API, they have to pay.
95
00:06:19,240 --> 00:06:21,080
It's fairly nominal.
96
00:06:21,080 --> 00:06:27,960
Cost per token as it's defined, which in essence is kind of like per word or per word generated
97
00:06:27,960 --> 00:06:30,160
or per word transcribed.
98
00:06:30,160 --> 00:06:32,840
But they've reduced the cost by 10 times.
99
00:06:32,840 --> 00:06:38,520
So that'll be interesting to see a further explosion of even more tools with the costs
100
00:06:38,520 --> 00:06:41,360
of actually leveraging OpenAI's platform dropping.
101
00:06:41,360 --> 00:06:46,960
I wonder if that will put a pressure on some of the business models of some of the early
102
00:06:46,960 --> 00:06:52,840
writing and transcription tools because presumably these new tools that we see emerge from here
103
00:06:52,840 --> 00:06:57,680
can do it even cheaper than was being done before.
104
00:06:57,680 --> 00:07:07,600
Yeah, I think switching over the API to the ChatGPT one is fairly straightforward.
105
00:07:07,600 --> 00:07:13,040
Most of the writing tools that I've been using made that switch immediately themselves.
106
00:07:13,040 --> 00:07:22,480
Certainly within hours, I was using the ChatGPT API via one of the platforms that I use, Haptip
107
00:07:22,480 --> 00:07:30,680
to Riku, R-I-K-U, great little tool if you're interested in playing around with AI models.
108
00:07:30,680 --> 00:07:31,920
They got it deployed very quickly.
109
00:07:31,920 --> 00:07:38,600
But yeah, that 10x reduction in cost is massive, an order of magnitude cost reduction.
110
00:07:38,600 --> 00:07:41,520
And that is not to be sniffed at.
111
00:07:41,520 --> 00:07:47,360
And quite exciting for, you know, if that's the direction that we're heading.
112
00:07:47,360 --> 00:07:51,760
What I do find interesting is how the alternative models are going to compete.
113
00:07:51,760 --> 00:07:56,400
So OpenAI is the big player in this space.
114
00:07:56,400 --> 00:08:03,280
The equivalent large language models to GPT-3, the predecessor to ChatGPT, there's lots of
115
00:08:03,280 --> 00:08:04,280
those available.
116
00:08:04,280 --> 00:08:08,200
So AI-21 has a model called Jurassic.
117
00:08:08,200 --> 00:08:10,840
Cohair has a large model as well.
118
00:08:10,840 --> 00:08:13,000
And then there's numerous others.
119
00:08:13,000 --> 00:08:17,940
Their cost per token usage was very similar.
120
00:08:17,940 --> 00:08:22,920
They were all in and around the same ballpark.
121
00:08:22,920 --> 00:08:29,120
And OpenAI have really thrown a hand grenade into the mix there.
122
00:08:29,120 --> 00:08:30,480
Yeah, absolutely.
123
00:08:30,480 --> 00:08:36,540
So fingers crossed, we'll see even more tools emerge at very attractive price points and
124
00:08:36,540 --> 00:08:41,920
potentially even more and more specialized and niche applications now that people can
125
00:08:41,920 --> 00:08:48,200
get hold of that API at a very low price and start creating apps for very specific use
126
00:08:48,200 --> 00:08:49,200
cases.
127
00:08:49,200 --> 00:08:52,160
Yeah, all being well.
128
00:08:52,160 --> 00:08:58,360
And I think the availability of Whisper as well is not to be overlooked.
129
00:08:58,360 --> 00:09:01,360
That is really going to transform app development.
130
00:09:01,360 --> 00:09:07,180
And I think when people start chaining together some of these tools, you know, Whisper working
131
00:09:07,180 --> 00:09:13,880
alongside ChatGPT, that's going to be a really interesting development as well.
132
00:09:13,880 --> 00:09:15,240
Marketers watch this space.
133
00:09:15,240 --> 00:09:16,240
Absolutely.
134
00:09:16,240 --> 00:09:22,960
You can imagine if you've got a content production workflow where you might need to interview
135
00:09:22,960 --> 00:09:28,120
a subject matter expert, you might ask ChatGPT to give you some inspiration for good interview
136
00:09:28,120 --> 00:09:32,760
questions on a particular topic, record the interview and have something like Whisper
137
00:09:32,760 --> 00:09:40,940
auto transcribe it and then have ChatGPT auto summarize or create a blog post off the basis
138
00:09:40,940 --> 00:09:42,400
of that transcription.
139
00:09:42,400 --> 00:09:50,240
So yes, very heavily influenced content production process by chaining those tools, as you said,
140
00:09:50,240 --> 00:09:51,240
Martin.
141
00:09:51,240 --> 00:09:55,320
From there, let's move a bit into talking about Cohere, one of these other large language
142
00:09:55,320 --> 00:10:01,920
models, because you spotted this week that they have released a beta of their new summarize
143
00:10:01,920 --> 00:10:04,040
tool and you've been having a play.
144
00:10:04,040 --> 00:10:07,640
So tell us a bit more about this tool and what your experiences have been with it this
145
00:10:07,640 --> 00:10:08,640
week.
146
00:10:08,640 --> 00:10:16,280
Yeah, so Cohere, an alternative provider of large language models and AI language technology,
147
00:10:16,280 --> 00:10:22,240
in fact, recently announced that they were going for a valuation of around $6 billion.
148
00:10:22,240 --> 00:10:27,280
So fundraising for that, you know, feel free to chip in Paul, if you've got a few coffers
149
00:10:27,280 --> 00:10:28,280
down the back of the sofa.
150
00:10:28,280 --> 00:10:32,960
I could buy 0.0000000000001% of that.
151
00:10:32,960 --> 00:10:40,220
Well, you know, I'm sure they'll be willing to listen.
152
00:10:40,220 --> 00:10:46,080
But they've just announced that one of their models, they've trained especially as a summarization
153
00:10:46,080 --> 00:10:47,080
engine.
154
00:10:47,080 --> 00:10:52,120
And this is quite interesting because ChatGPT, if you go into ChatGPT and just copy and paste
155
00:10:52,120 --> 00:10:56,360
an article from the web and stick it in, don't say anything, just literally throw in the
156
00:10:56,360 --> 00:11:01,440
article and press enter, it will give you a kind of summary of that article.
157
00:11:01,440 --> 00:11:05,440
Just it will say, well, this is what that's about.
158
00:11:05,440 --> 00:11:08,280
The thing is, it's limited on character count.
159
00:11:08,280 --> 00:11:12,640
So it's still limited on the number of tokens as we've just been talking about, and it's
160
00:11:12,640 --> 00:11:14,760
got quite a small limit.
161
00:11:14,760 --> 00:11:20,640
But Cohere announced that they were basically creating a summarization tool that could have
162
00:11:20,640 --> 00:11:25,960
up to 50,000 characters, which is an enormous amount of text.
163
00:11:25,960 --> 00:11:28,800
In fact, I have been playing around with it this week.
164
00:11:28,800 --> 00:11:35,900
And I took a journal article about, it was a study looking at, in fact, it doesn't matter
165
00:11:35,900 --> 00:11:43,200
what it was, but it was about marketing AI and the current state of marketing and artificial
166
00:11:43,200 --> 00:11:44,200
intelligence.
167
00:11:44,200 --> 00:11:52,200
15 page PDF, I copied all of the text out of that, threw it into this summarization
168
00:11:52,200 --> 00:11:53,200
tool.
169
00:11:53,200 --> 00:11:57,760
By the time I deleted all the references and all of that, it was coming in at about 40,000
170
00:11:57,760 --> 00:11:58,760
characters.
171
00:11:58,760 --> 00:12:03,120
It was a really long meaty bit of text.
172
00:12:03,120 --> 00:12:04,840
And I asked it to summarize it.
173
00:12:04,840 --> 00:12:07,680
And yeah, sure, it summarized it.
174
00:12:07,680 --> 00:12:11,280
I would describe the summary as being comically small.
175
00:12:11,280 --> 00:12:13,600
It was four sentences.
176
00:12:13,600 --> 00:12:17,200
I mean, it really, really boiled it down.
177
00:12:17,200 --> 00:12:19,480
It actually has a setting on it.
178
00:12:19,480 --> 00:12:23,960
It's this summarization tool where you can go small, medium and large.
179
00:12:23,960 --> 00:12:27,760
And I thought, that's quite useful if you've got a really long piece of text, you might
180
00:12:27,760 --> 00:12:34,520
want to summarize 40,000 characters down to, I don't know, like a thousand characters or
181
00:12:34,520 --> 00:12:35,520
something.
182
00:12:35,520 --> 00:12:36,520
But no, four sentences.
183
00:12:36,520 --> 00:12:42,760
Don't get me wrong, it was accurate, but I would have liked a little more.
184
00:12:42,760 --> 00:12:43,760
Yeah.
185
00:12:43,760 --> 00:12:45,240
But it was accurate, you feel?
186
00:12:45,240 --> 00:12:47,960
Yeah, it was accurate.
187
00:12:47,960 --> 00:12:50,600
It did a good job.
188
00:12:50,600 --> 00:12:57,760
So I can't knock it too much, but it did just feel like it had summarized hyper aggressively.
189
00:12:57,760 --> 00:13:02,280
Yeah, boiled it right down to the truest essence.
190
00:13:02,280 --> 00:13:07,480
And I just Googled 50,000 characters is somewhere between 7,000 and 12,000 words.
191
00:13:07,480 --> 00:13:10,800
So we're talking about pretty chunky pieces here.
192
00:13:10,800 --> 00:13:15,520
And it's interesting that your experience was that it was accurate because some of the
193
00:13:15,520 --> 00:13:21,200
summarized tools that I've been playing with and even some of the meeting summarized tools
194
00:13:21,200 --> 00:13:25,720
like Tactic that has now AI driven meeting summarization in.
195
00:13:25,720 --> 00:13:32,420
So far, if I'm honest, when I look at the summary, I feel like there are key elements
196
00:13:32,420 --> 00:13:38,080
of that call or that paper or whatever it is that I'm summarizing that these tools are
197
00:13:38,080 --> 00:13:39,080
missing.
198
00:13:39,080 --> 00:13:46,160
And I think when I reflect upon, even if I have a meeting with three or four people and
199
00:13:46,160 --> 00:13:50,000
I come out of that and say, what do you think was the most important thing?
200
00:13:50,000 --> 00:13:54,960
Usually like the key bits we probably all agree on, but people see different things.
201
00:13:54,960 --> 00:13:58,440
Even if you and I read a book, Martin, and I say, what was the most interesting thing
202
00:13:58,440 --> 00:13:59,440
for you?
203
00:13:59,440 --> 00:14:00,540
It's going to be different, right?
204
00:14:00,540 --> 00:14:04,120
Because the context that you bring to reading that book, the previous knowledge that you
205
00:14:04,120 --> 00:14:08,840
have, the other books you've read, all that other stuff is going to influence what you
206
00:14:08,840 --> 00:14:09,840
find interesting.
207
00:14:09,840 --> 00:14:15,440
So I think it's going to be really interesting space to see how this plays out, how these
208
00:14:15,440 --> 00:14:22,200
tools improve and also whether or not they just cause a bit of a homogenization in our
209
00:14:22,200 --> 00:14:27,240
interpretation of information that may or may not be a good thing, right?
210
00:14:27,240 --> 00:14:31,480
Sometimes it's good for you and I to take different things away from a piece of text,
211
00:14:31,480 --> 00:14:36,080
let's say, because that the interesting magic is probably somewhere in the middle.
212
00:14:36,080 --> 00:14:37,460
Like why did you find that interesting?
213
00:14:37,460 --> 00:14:38,800
Why did I find that interesting?
214
00:14:38,800 --> 00:14:40,920
What conversation can we have about that?
215
00:14:40,920 --> 00:14:43,680
We rely too heavily on some of these summarization tools.
216
00:14:43,680 --> 00:14:46,200
We might lose a bit of that, I worry.
217
00:14:46,200 --> 00:14:47,800
Yeah, hugely.
218
00:14:47,800 --> 00:14:51,880
And I actually find, so I've been using a summarization tool, which is no longer available
219
00:14:51,880 --> 00:14:52,880
actually.
220
00:14:52,880 --> 00:15:00,320
It was a great tool called Summary, or they've changed their pricing model anyway.
221
00:15:00,320 --> 00:15:04,520
Their tool was brilliant for, let's say I was on Harvard Business Review and there's
222
00:15:04,520 --> 00:15:06,360
a long form article.
223
00:15:06,360 --> 00:15:10,760
I could read the kind of cliff notes, just give me the summary.
224
00:15:10,760 --> 00:15:14,600
And then if I wanted to get the full piece, if I thought, oh, that's quite interesting.
225
00:15:14,600 --> 00:15:18,160
I want the detail and I would jump in and read the full thing.
226
00:15:18,160 --> 00:15:21,880
So I do think there's a place for summarization, but you're absolutely right.
227
00:15:21,880 --> 00:15:30,960
We have to be careful about how we approach and not shortcut our way to going, well, I've
228
00:15:30,960 --> 00:15:39,120
read the four sentence summary of that 8,000 word document and I think I know it all now.
229
00:15:39,120 --> 00:15:40,120
Absolutely agree.
230
00:15:40,120 --> 00:15:41,120
Absolutely agree.
231
00:15:41,120 --> 00:15:42,120
Right.
232
00:15:42,120 --> 00:15:46,960
Let's move on to our next story, which is about Hugging Face and AWS partnering to bring
233
00:15:46,960 --> 00:15:49,000
open source models to developers.
234
00:15:49,000 --> 00:15:53,280
Again, this is another one you picked up, Martin, while I was out signing myself in
235
00:15:53,280 --> 00:15:54,280
San Diego.
236
00:15:54,280 --> 00:15:56,160
What's the story here?
237
00:15:56,160 --> 00:15:59,080
What do marketers need to know?
238
00:15:59,080 --> 00:16:07,160
So this is, for those of you that don't know who Hugging Face are, so Hugging Face is a
239
00:16:07,160 --> 00:16:13,280
community of artificial intelligence and machine learning researchers and enthusiasts.
240
00:16:13,280 --> 00:16:22,240
And it's kind of the, if you're a developer, the GitHub for AI ML projects.
241
00:16:22,240 --> 00:16:27,440
And it's where models will be posted and shared and you can have a bit of a play with them
242
00:16:27,440 --> 00:16:31,560
and there'll be summary cards explaining what they do.
243
00:16:31,560 --> 00:16:36,280
And it covers all areas of artificial intelligence and machine learning.
244
00:16:36,280 --> 00:16:42,240
This announcement is a partnership between Amazon Web Services, the cloud services from
245
00:16:42,240 --> 00:16:48,440
Amazon and Hugging Face to say that they're going to make some of these models available
246
00:16:48,440 --> 00:16:50,580
through AWS.
247
00:16:50,580 --> 00:17:00,880
So in the way that Microsoft's cloud services, Azure has open AI access now, so you can plug
248
00:17:00,880 --> 00:17:06,600
into the open AI tools, AWS is bringing in some of these alternative models.
249
00:17:06,600 --> 00:17:11,720
And actually AWS, I think, already has a large language model through AI 21.
250
00:17:11,720 --> 00:17:16,880
I believe that they are hosted or they're accessible via API on there as well.
251
00:17:16,880 --> 00:17:22,240
So if you're an app developer or a web developer or you're planning a new digital project or
252
00:17:22,240 --> 00:17:26,920
product, should I say, you can plug into these new models.
253
00:17:26,920 --> 00:17:32,080
And Hugging Face have said that their open source models, I don't think they've announced
254
00:17:32,080 --> 00:17:38,520
the extent to which they're going to be putting them on there, like how many and how frequently
255
00:17:38,520 --> 00:17:43,280
or what have you, but open source models will be made available on AWS.
256
00:17:43,280 --> 00:17:47,880
Although it should be said that they've also said it's not an exclusivity.
257
00:17:47,880 --> 00:17:51,640
They don't want to tie themselves in with one partner.
258
00:17:51,640 --> 00:17:54,920
Good stuff.
259
00:17:54,920 --> 00:17:56,920
What does this mean for marketers specifically?
260
00:17:56,920 --> 00:18:04,440
Well, it certainly means they're going to be less vulnerable to single source dependency.
261
00:18:04,440 --> 00:18:10,840
When we saw open AI go down last week and all of a sudden chat GPT and GPT-3 went down,
262
00:18:10,840 --> 00:18:12,040
the whole world panicked.
263
00:18:12,040 --> 00:18:16,080
I think this makes us less reliant on a single point.
264
00:18:16,080 --> 00:18:18,840
There'll be more competition in the market.
265
00:18:18,840 --> 00:18:20,720
But actually it just means that we've got more tools.
266
00:18:20,720 --> 00:18:28,620
If you're doing some sort of app development for your product, you can bring in new technologies.
267
00:18:28,620 --> 00:18:34,600
You can extend your product with these bits of tech more easily.
268
00:18:34,600 --> 00:18:39,760
You're not going to have to self-host and self-configure these AI ML models.
269
00:18:39,760 --> 00:18:44,600
They'll just be available on the cloud platforms that are already the most popular.
270
00:18:44,600 --> 00:18:45,600
Cool.
271
00:18:45,600 --> 00:18:46,600
Cool.
272
00:18:46,600 --> 00:18:50,400
There was another story you spotted with Hugging Face this week as well, collaborating with
273
00:18:50,400 --> 00:18:55,520
Stability AI and Canva around a new AI research nonprofit.
274
00:18:55,520 --> 00:18:59,240
What's happening there that marketers need to know about?
275
00:18:59,240 --> 00:19:06,080
Stability AI, Hugging Face and Canva are creating a not-for-profit AI research lab.
276
00:19:06,080 --> 00:19:10,400
Canva you'll be familiar with from the design world.
277
00:19:10,400 --> 00:19:17,040
Stability AI, they create the stable diffusion text-to-image generation model.
278
00:19:17,040 --> 00:19:25,480
They've created this new foundation, which is going to be looking at ethics in AI, really.
279
00:19:25,480 --> 00:19:31,560
That's what they're focusing on, ethics and alignment of AI.
280
00:19:31,560 --> 00:19:38,320
How we humans align with AI with the intent, there's a whole thing around alignment.
281
00:19:38,320 --> 00:19:43,960
It's somewhat controversial though because it's got publics, sorry, private sector investment
282
00:19:43,960 --> 00:19:45,920
and private sector backers.
283
00:19:45,920 --> 00:19:52,400
There are people saying that, yes, this is a great foundation, a great cause, but is
284
00:19:52,400 --> 00:19:56,920
it going to be truly impartial?
285
00:19:56,920 --> 00:19:57,920
We'll wait and see.
286
00:19:57,920 --> 00:20:03,440
What marketers need to be aware of though is that with any AI models, we have to be
287
00:20:03,440 --> 00:20:07,320
aware of weights and biases in the models.
288
00:20:07,320 --> 00:20:16,040
We don't want AI just putting out bad information or being unethical or bringing all of those
289
00:20:16,040 --> 00:20:19,800
terrible examples that we've seen in the past where it's misogynistic, racist, take your
290
00:20:19,800 --> 00:20:20,800
pick.
291
00:20:20,800 --> 00:20:24,360
They're doing a lot of work around this space.
292
00:20:24,360 --> 00:20:29,640
They might be one of the organizations leading the way in terms of the ethical frameworks
293
00:20:29,640 --> 00:20:38,600
and advising marketers and businesses more broadly how best to deploy AI in a ethical
294
00:20:38,600 --> 00:20:39,600
and compliant way.
295
00:20:39,600 --> 00:20:41,600
Yeah, I think that's really interesting.
296
00:20:41,600 --> 00:20:47,080
I've been listening a fair bit this week to Paul Reutzer over at the Marketing AI Institute
297
00:20:47,080 --> 00:20:52,680
talking about the ethics and guidelines of how we might use AI.
298
00:20:52,680 --> 00:20:57,960
I think one of the key things they've been chatting about that really resonated with
299
00:20:57,960 --> 00:21:02,440
me is that it does look a little bit like government are going to move too slowly on
300
00:21:02,440 --> 00:21:08,000
this to really be able to put anything in place to really apply regulations to how we
301
00:21:08,000 --> 00:21:12,080
use AI and certainly in marketing and communications.
302
00:21:12,080 --> 00:21:18,880
And that probably companies are going to have to take the lead in being clear about what
303
00:21:18,880 --> 00:21:23,400
they use AI for, especially in their marketing and communications, what was produced using
304
00:21:23,400 --> 00:21:26,520
say chat GPT, for example.
305
00:21:26,520 --> 00:21:32,240
So seeing that there's lots of people who are thinking about this, almost taking on
306
00:21:32,240 --> 00:21:36,520
the role potentially of regulators and government would traditionally have because I guess the
307
00:21:36,520 --> 00:21:40,880
assumption is they're going to move too slow because these technologies are just racing
308
00:21:40,880 --> 00:21:42,880
out the gate and developing so quickly.
309
00:21:42,880 --> 00:21:45,120
Yeah, I think that's a fair shout.
310
00:21:45,120 --> 00:21:50,000
I think the EU in the defense are putting together some AI legislation.
311
00:21:50,000 --> 00:21:55,240
I think the White House in the US has also got policy wonks looking at AI.
312
00:21:55,240 --> 00:21:59,440
But yes, the pace, when you just look at the pace of change in the industry, how quick
313
00:21:59,440 --> 00:22:03,840
products are coming out, I don't know how legislators could hope to catch up.
314
00:22:03,840 --> 00:22:10,960
But you mentioned the AI Marketing or the Marketing AI Institute, they've got the Creative
315
00:22:10,960 --> 00:22:16,920
Commons licensed manifesto that anyone can go and use for their own organization to put
316
00:22:16,920 --> 00:22:23,360
in place some kind of structure around how they plan to use AI and make that publicly
317
00:22:23,360 --> 00:22:24,360
available.
318
00:22:24,360 --> 00:22:26,600
So that's a resource that's well worth going and checking out.
319
00:22:26,600 --> 00:22:32,000
Yeah, I can imagine a time where we have our privacy statements on all of our websites
320
00:22:32,000 --> 00:22:37,200
saying how we use cookies and how we use customer and visitor data and all that stuff.
321
00:22:37,200 --> 00:22:42,400
We're going to be needing to add our policies on AI to those documents, or perhaps having
322
00:22:42,400 --> 00:22:48,480
completely separate web pages or documents on our sites explaining how we're using AI
323
00:22:48,480 --> 00:22:53,200
tools in our marketing and communications with prospects and customers and visitors.
324
00:22:53,200 --> 00:22:57,360
Yeah, certainly, I would welcome that kind of transparency.
325
00:22:57,360 --> 00:23:00,920
Yeah, I think it will become key.
326
00:23:00,920 --> 00:23:02,320
We can talk about this a bit more in a moment.
327
00:23:02,320 --> 00:23:06,360
It's probably a bit of a segue from some of the news we've been seeing to some of the
328
00:23:06,360 --> 00:23:14,200
things that have been on our minds as it relates to AI and marketing is what happens when we
329
00:23:14,200 --> 00:23:21,840
see a huge amount of content hit the market generated by ChatGPT and other tools.
330
00:23:21,840 --> 00:23:27,160
We get an even bigger content deluge than we've had over the last 10 years with the
331
00:23:27,160 --> 00:23:30,940
emergence and maturation of content marketing and inbound marketing.
332
00:23:30,940 --> 00:23:36,920
Content now just became even faster and cheaper and easier to just pump out content, whether
333
00:23:36,920 --> 00:23:39,080
it's actually useful to anyone or not.
334
00:23:39,080 --> 00:23:44,840
But if it's easy to do, we'll see lots of people doing it.
335
00:23:44,840 --> 00:23:51,720
Actually I think we'll see that emergence of people needing to sift through all that
336
00:23:51,720 --> 00:23:58,560
crap, trying to figure out what's actually worth listening to and guaranteed human content
337
00:23:58,560 --> 00:24:02,000
becoming a bit of a premium and a bit of a marker for trust that you're going to get
338
00:24:02,000 --> 00:24:05,760
some interesting novel ideas.
339
00:24:05,760 --> 00:24:09,600
So that as a marketer thinking about, well, how can I get ahead of that?
340
00:24:09,600 --> 00:24:11,600
And what types of content can you produce?
341
00:24:11,600 --> 00:24:15,280
Stuff like the podcast that we're doing now, right?
342
00:24:15,280 --> 00:24:19,000
I would love to see the robotic version of Martin.
343
00:24:19,000 --> 00:24:22,480
I suspect he's nowhere near as interesting as funny as you are.
344
00:24:22,480 --> 00:24:26,600
But we're certainly not there yet in terms of being, this isn't a deep fake, although
345
00:24:26,600 --> 00:24:29,240
that would be interesting as well.
346
00:24:29,240 --> 00:24:34,240
So things like podcasts and other ways to just bring that authenticity to content as
347
00:24:34,240 --> 00:24:41,440
we see an explosion of this stuff only helped by having a little line of text at the bottom
348
00:24:41,440 --> 00:24:46,040
of a blog post that says, we use Jack GPT to help put together this blog post, but all
349
00:24:46,040 --> 00:24:47,040
the ideas are ours.
350
00:24:47,040 --> 00:24:48,040
Right?
351
00:24:48,040 --> 00:24:50,840
That would be really, really helpful for people trying to figure out what to listen to.
352
00:24:50,840 --> 00:24:53,080
What are your thoughts on all of this, Mian?
353
00:24:53,080 --> 00:24:59,360
Yeah, I echo that sentiment is just a deluge of content and it's so easy to create articles
354
00:24:59,360 --> 00:25:00,360
now.
355
00:25:00,360 --> 00:25:06,840
The first time you show someone the simple prompt of write an article about such and
356
00:25:06,840 --> 00:25:13,080
such, their eyes light up, particularly small business owners that I've shown this to in
357
00:25:13,080 --> 00:25:14,600
recent weeks and months.
358
00:25:14,600 --> 00:25:19,240
And they get very excited and you do have to go, yeah, just because you can doesn't
359
00:25:19,240 --> 00:25:20,840
mean that you should.
360
00:25:20,840 --> 00:25:23,880
This is an important thing to remember.
361
00:25:23,880 --> 00:25:31,360
So yeah, human in the loop and actually human being the majority of the loop actually, I
362
00:25:31,360 --> 00:25:33,600
think is going to be the big part.
363
00:25:33,600 --> 00:25:34,920
These are assistants at the moment.
364
00:25:34,920 --> 00:25:44,800
I saw a nice headline this morning actually saying, we should think about less of AI being
365
00:25:44,800 --> 00:25:49,800
another person in the sense of like replacing an employee for instance, but think of it
366
00:25:49,800 --> 00:25:52,280
more as being your own second brain.
367
00:25:52,280 --> 00:25:55,320
And I do quite like that.
368
00:25:55,320 --> 00:25:57,320
It's an alternate way of thinking.
369
00:25:57,320 --> 00:26:01,280
It helps me write content better.
370
00:26:01,280 --> 00:26:07,520
It helps me produce articles quicker or gives me a different angle when brainstorming.
371
00:26:07,520 --> 00:26:16,080
But also it can do tasks like, I don't know, audio editing, using something like Descript
372
00:26:16,080 --> 00:26:20,520
that I otherwise don't have the skills to do, so it kind of adds these skills in.
373
00:26:20,520 --> 00:26:29,400
So yeah, I think that's an interesting thing for us to consider is how transparent we want
374
00:26:29,400 --> 00:26:31,400
to be with that.
375
00:26:31,400 --> 00:26:39,240
How do we be responsible marketers using AI and how, what's the word I'm looking for,
376
00:26:39,240 --> 00:26:40,240
front and center.
377
00:26:40,240 --> 00:26:46,280
I do like the idea of the disclaimers though on articles and whatnot.
378
00:26:46,280 --> 00:26:47,360
Yeah.
379
00:26:47,360 --> 00:26:53,300
I think enabling augmentation is definitely how I'm viewing the AI tools, certainly within
380
00:26:53,300 --> 00:26:56,040
our agency and within my own work.
381
00:26:56,040 --> 00:27:00,080
And I think to your point, what we kind of all need is an uncle Ben on our shoulder,
382
00:27:00,080 --> 00:27:01,080
right?
383
00:27:01,080 --> 00:27:02,880
With great power comes great responsibility.
384
00:27:02,880 --> 00:27:07,280
Marketers are very well known for taking cool new emerging tech and then massively overusing
385
00:27:07,280 --> 00:27:09,360
it and ruining it for everybody else.
386
00:27:09,360 --> 00:27:13,560
So come on marketers, let's do our best to try and control ourselves this time and not
387
00:27:13,560 --> 00:27:15,080
let that happen.
388
00:27:15,080 --> 00:27:16,160
Right.
389
00:27:16,160 --> 00:27:20,800
Let's talk about prompt engineering, Martin, because I've been thinking about this a lot.
390
00:27:20,800 --> 00:27:23,400
I know we've been talking a fair bit about prompt engineering.
391
00:27:23,400 --> 00:27:28,440
So for those that are not familiar, when you're using a number of AI tools, whether it's Dually2
392
00:27:28,440 --> 00:27:33,640
for image generation, ChatGPT to help you come up with interview questions or blog posts,
393
00:27:33,640 --> 00:27:37,360
in essence, you have to prompt the system to give you the thing that you want.
394
00:27:37,360 --> 00:27:40,320
And you do that by writing in the text box what you want.
395
00:27:40,320 --> 00:27:47,880
So you might say, create a blog post for me about the five exciting applications of confocal
396
00:27:47,880 --> 00:27:50,280
microscopy and then it will do its best to write it.
397
00:27:50,280 --> 00:27:54,200
And you can say things like write it for a PhD level audience or other different things
398
00:27:54,200 --> 00:27:59,720
that you can add to your prompt to further shape what you want the tool to give you and
399
00:27:59,720 --> 00:28:03,200
the types of quality output that you'll get.
400
00:28:03,200 --> 00:28:08,120
And we've been talking a bit about this as it relates to why marketers are actually perfectly
401
00:28:08,120 --> 00:28:15,020
positioned to get the most out of these tools, because creating a great prompt is like creating
402
00:28:15,020 --> 00:28:16,560
a great marketing brief.
403
00:28:16,560 --> 00:28:19,560
It's about thinking about who you're trying to reach.
404
00:28:19,560 --> 00:28:21,600
What's the key messages that you want to convey?
405
00:28:21,600 --> 00:28:24,080
What sort of style do you want to write in?
406
00:28:24,080 --> 00:28:30,560
What sort of key items and data points do you want to focus on?
407
00:28:30,560 --> 00:28:34,160
So what are your thoughts, Martin, marketers as prompt experts?
408
00:28:34,160 --> 00:28:38,320
Should we be halfway along already really in terms of being good briefers?
409
00:28:38,320 --> 00:28:39,320
Absolutely.
410
00:28:39,320 --> 00:28:43,680
A good brief is going to have all of those elements you've just spoken about.
411
00:28:43,680 --> 00:28:45,680
It's going to say, what do you want to achieve?
412
00:28:45,680 --> 00:28:51,240
In fact, actually, just thinking about the prompting, it's interesting using chat GPT
413
00:28:51,240 --> 00:28:57,760
versus GPT-3 in the way that you would prompt them.
414
00:28:57,760 --> 00:29:05,000
I've found that with chat GPT, you're telling the chat that they are a role.
415
00:29:05,000 --> 00:29:06,000
You're giving them a role.
416
00:29:06,000 --> 00:29:11,280
You are a such and such, and then it will play that role in your chat.
417
00:29:11,280 --> 00:29:17,160
Whereas with GPT-3, you can write your prompts from a first-person perspective.
418
00:29:17,160 --> 00:29:21,440
Some of the best prompts that I've personally written have had all of those elements that
419
00:29:21,440 --> 00:29:22,800
you've spoken about there.
420
00:29:22,800 --> 00:29:30,960
So I've maybe started my prompt saying, I am an award-winning technology journalist
421
00:29:30,960 --> 00:29:34,620
writing for techradar.com.
422
00:29:34,620 --> 00:29:39,960
My editor came to me and asked that I write an article about this, this, this, and this.
423
00:29:39,960 --> 00:29:43,880
All of that is the detail that you've just spoken about there.
424
00:29:43,880 --> 00:29:49,880
So it's about the product, the key messages that I've got to hit, the audience that it's
425
00:29:49,880 --> 00:29:56,080
been written for, the level of detail that might be the strengths and weaknesses of a
426
00:29:56,080 --> 00:30:00,800
particular strategy or approach or what have you.
427
00:30:00,800 --> 00:30:05,480
And then I will end my prompt with, this was my final written draft.
428
00:30:05,480 --> 00:30:09,280
And then it will write the article for me.
429
00:30:09,280 --> 00:30:10,280
Right.
430
00:30:10,280 --> 00:30:11,280
Yeah.
431
00:30:11,280 --> 00:30:14,080
I mean, such is the explosion of this.
432
00:30:14,080 --> 00:30:17,600
I don't know if you've seen this, Mark, but now there are marketplaces for prompts.
433
00:30:17,600 --> 00:30:22,960
Have you seen these like prompt base where you can buy and sell prompts in order to get
434
00:30:22,960 --> 00:30:24,520
the very most out of these tools?
435
00:30:24,520 --> 00:30:26,920
Have you had to play with any of those?
436
00:30:26,920 --> 00:30:27,920
I've not.
437
00:30:27,920 --> 00:30:35,680
I saw them burst onto the scene, particularly after the text to image platforms came out.
438
00:30:35,680 --> 00:30:37,800
That was where there was a big boom.
439
00:30:37,800 --> 00:30:42,760
And then I think, you know, the resurgence, well, I say resurgence, the emergence of,
440
00:30:42,760 --> 00:30:45,520
of chat GPT just took it to a whole new level.
441
00:30:45,520 --> 00:30:47,400
Yeah, it's fine.
442
00:30:47,400 --> 00:30:54,440
I'm so I'm on the prompt base now and there's a business model generator for $4.99 and SEO
443
00:30:54,440 --> 00:30:57,440
blog article generated for $2.99.
444
00:30:57,440 --> 00:31:05,120
The ingenuity of people to jump on these trends and figure out ways, you know, to make a buck
445
00:31:05,120 --> 00:31:09,200
off of the back of things that they've spent, you know, time getting good at, because certainly
446
00:31:09,200 --> 00:31:14,040
writing good prompts is not, it's you have to practice, you have to learn how to get
447
00:31:14,040 --> 00:31:15,240
the most out of the tools.
448
00:31:15,240 --> 00:31:17,240
So absolutely fascinating to see that.
449
00:31:17,240 --> 00:31:20,880
There is a dark side to this mine and I am going to have a rant.
450
00:31:20,880 --> 00:31:26,160
I see you smile because you know already how mad I am about this, which is my LinkedIn
451
00:31:26,160 --> 00:31:30,200
feed and I've probably done this to myself by clicking on too many of them and telling
452
00:31:30,200 --> 00:31:37,760
the algorithm I love them is so full of sliding carousels of the different types of prompts
453
00:31:37,760 --> 00:31:44,000
that you can give chat GPT that I've now seen eight or nine variations of the same prompt
454
00:31:44,000 --> 00:31:48,840
where people jump on this bandwagon to try and get lots of likes and shares by coming
455
00:31:48,840 --> 00:31:49,840
up with these carousels.
456
00:31:49,840 --> 00:31:53,880
So that annoys me a bit because it's like if it's done, it's done, right?
457
00:31:53,880 --> 00:31:55,520
Come up with something new.
458
00:31:55,520 --> 00:31:59,680
The second thing that triggers a bit of a rant from me is now I'm starting to see some
459
00:31:59,680 --> 00:32:04,440
of these prompts really move into areas that I think are dangerous.
460
00:32:04,440 --> 00:32:08,340
If people are going to use these prompts without really knowing how the underlying technology
461
00:32:08,340 --> 00:32:12,680
works and what you can rely on something like chat GPT for, they're actually going to get
462
00:32:12,680 --> 00:32:14,880
themselves in a bit of a sticky situation.
463
00:32:14,880 --> 00:32:20,280
The first one that comes to mind that's been bugging me this week is asking chat GPT to
464
00:32:20,280 --> 00:32:23,240
create bio personas for you.
465
00:32:23,240 --> 00:32:25,240
Chat GPT doesn't know who your customers are.
466
00:32:25,240 --> 00:32:27,160
He's never spoken to any of your customers.
467
00:32:27,160 --> 00:32:33,560
It is a prediction engine that's in effect read lots and lots of materials like books
468
00:32:33,560 --> 00:32:36,080
and web pages, et cetera.
469
00:32:36,080 --> 00:32:39,440
Its knowledge only even goes up to early 2021.
470
00:32:39,440 --> 00:32:40,720
It doesn't know your customers, right?
471
00:32:40,720 --> 00:32:45,640
It can come up with some sensible sounding things, but if you're going to base your entire
472
00:32:45,640 --> 00:32:51,360
messaging strategy on a prediction engine rather than actually speaking to your customers,
473
00:32:51,360 --> 00:32:52,360
you've got a problem, right?
474
00:32:52,360 --> 00:32:56,840
And I think this nicely highlights how we've talked about it a bit earlier.
475
00:32:56,840 --> 00:33:01,600
Great power comes great responsibility and really just understanding what these tools
476
00:33:01,600 --> 00:33:03,140
are good at and what they're not.
477
00:33:03,140 --> 00:33:09,000
What you can do is ask chat GPT to give you some questions to go ask your customers to
478
00:33:09,000 --> 00:33:10,840
help gather the intel you need.
479
00:33:10,840 --> 00:33:15,480
It's great for that type of brainstorming and idea generation for how you might go about
480
00:33:15,480 --> 00:33:18,560
doing something.
481
00:33:18,560 --> 00:33:23,360
Another area where I've seen this is an SEO where people are using it to help them do
482
00:33:23,360 --> 00:33:25,480
keyword analysis.
483
00:33:25,480 --> 00:33:27,360
It doesn't know the search volumes.
484
00:33:27,360 --> 00:33:32,600
All of its data that it's trained on is really quite old now in some cases.
485
00:33:32,600 --> 00:33:38,800
It can give you ideas for possible keywords that you can go feed into a proper tool like
486
00:33:38,800 --> 00:33:42,560
a SEM rush or Spifer or something like that.
487
00:33:42,560 --> 00:33:49,000
But it's not going to be the be all and end all in terms of helping you do your keyword
488
00:33:49,000 --> 00:33:52,280
research or any of these other things.
489
00:33:52,280 --> 00:33:54,600
I wanted to get that out there because it's been on my mind.
490
00:33:54,600 --> 00:33:58,920
I wanted to warn people to use these tools carefully, use them for things they can actually
491
00:33:58,920 --> 00:33:59,920
help with.
492
00:33:59,920 --> 00:34:04,280
Have you seen anything in this area, Martin, that made you worry or made you particularly
493
00:34:04,280 --> 00:34:06,040
excited around prompts?
494
00:34:06,040 --> 00:34:11,920
Well, I have a Twitter feed which is very similar to the LinkedIn carousel feed with
495
00:34:11,920 --> 00:34:19,840
just numerous tweets going, download my ebook full of prompts, get this Google Sheet with
496
00:34:19,840 --> 00:34:26,520
in fact, I saw one shared today, 700 chat GPT prompts, access it for free here.
497
00:34:26,520 --> 00:34:32,280
Yeah, I'll be honest, I'm over it.
498
00:34:32,280 --> 00:34:34,160
Just done with it.
499
00:34:34,160 --> 00:34:38,920
But fair play to people for jumping on the bandwagon and trying to make hay when the
500
00:34:38,920 --> 00:34:39,920
sun is shining.
501
00:34:39,920 --> 00:34:45,760
If you've been listening to the leaders in this space that quite literally, the Sam Altman
502
00:34:45,760 --> 00:34:51,560
of this world, he's a CEO of OpenAI if people aren't aware.
503
00:34:51,560 --> 00:34:58,360
He said recently that ideally prompt engineering isn't going to be a thing.
504
00:34:58,360 --> 00:35:04,240
You're not going to need to engineer a prompt because in the future, the AI, the language
505
00:35:04,240 --> 00:35:11,000
models will just be so good at understanding your intent that you don't need to engineer
506
00:35:11,000 --> 00:35:12,440
or craft these prompts.
507
00:35:12,440 --> 00:35:18,440
So fair play if you want to make the 6,000th version of the same carousel and get a few
508
00:35:18,440 --> 00:35:22,000
likes, tweets, and reposts and go for it.
509
00:35:22,000 --> 00:35:25,360
But I wouldn't be looking at it as a long-term strategy.
510
00:35:25,360 --> 00:35:26,360
Yeah, agreed.
511
00:35:26,360 --> 00:35:28,360
That sounds great, Martin.
512
00:35:28,360 --> 00:35:33,520
I certainly look forward to the day when the tools are good enough to understand our intent
513
00:35:33,520 --> 00:35:39,880
enough that we don't need to worry about that too much.
514
00:35:39,880 --> 00:35:45,920
Last part of today's podcast then, we're going to look at our tool of the week.
515
00:35:45,920 --> 00:35:49,880
This week, you're going to take us through Market Muse, Martin, something you've been
516
00:35:49,880 --> 00:35:50,880
using a lot of.
517
00:35:50,880 --> 00:35:52,280
I'm particularly excited to get into this.
518
00:35:52,280 --> 00:35:57,120
I have not used it much and it's not a generative AI tool.
519
00:35:57,120 --> 00:35:58,120
Hurrah!
520
00:35:58,120 --> 00:36:01,680
AI can help with huge amounts of things, help marketers be more effective.
521
00:36:01,680 --> 00:36:05,920
We've heard so much about tools that can create images and texts and video.
522
00:36:05,920 --> 00:36:08,600
Let's dive a bit more into Market Muse, Martin.
523
00:36:08,600 --> 00:36:14,000
Tell us a bit more about what it is and what it does and who it's for.
524
00:36:14,000 --> 00:36:20,340
Market Muse is for any marketing team that produces content and has content at the heart
525
00:36:20,340 --> 00:36:22,960
of their strategy.
526
00:36:22,960 --> 00:36:31,120
It is a NLP natural language processing powered content optimization tool and it is a beast.
527
00:36:31,120 --> 00:36:32,120
I love it.
528
00:36:32,120 --> 00:36:36,240
There are two main features of it really.
529
00:36:36,240 --> 00:36:47,600
They have the single page optimization element and you can do research and then optimization
530
00:36:47,600 --> 00:36:55,160
or you've got, if you go for the premium tier plan, they've got a tool called inventory
531
00:36:55,160 --> 00:36:59,440
and this is where you get full site-wide optimization.
532
00:36:59,440 --> 00:37:01,080
How does it all work?
533
00:37:01,080 --> 00:37:05,720
You start off by putting in your keywords and what they do, you put in your keyword
534
00:37:05,720 --> 00:37:12,120
or your key phrase and it will scrape the SERPs, the search engine results pages, get
535
00:37:12,120 --> 00:37:19,440
all of the top results for say like the top 30 pages and it will read those pages.
536
00:37:19,440 --> 00:37:25,800
Then using natural language processes, it creates a kind of topic model for that keyword.
537
00:37:25,800 --> 00:37:32,360
Let's say you want to rank for Premier League football.
538
00:37:32,360 --> 00:37:35,960
If you write an article about Premier League football, you're going to have to talk about
539
00:37:35,960 --> 00:37:38,680
things like the clubs.
540
00:37:38,680 --> 00:37:41,600
You're going to have to talk about Manchester United, Arsenal, Man City.
541
00:37:41,600 --> 00:37:43,560
Not Derby County though, aren't they?
542
00:37:43,560 --> 00:37:47,800
You would not be focusing much on Derby County unless you mentioned that the worst team in
543
00:37:47,800 --> 00:37:55,320
Premier League history was Derby County with 11 points in 2006, 2007 or 2007, 2008.
544
00:37:55,320 --> 00:37:56,320
I forget.
545
00:37:56,320 --> 00:37:57,320
It was all a traumatic experience.
546
00:37:57,320 --> 00:37:58,320
We digress.
547
00:37:58,320 --> 00:37:59,320
We digress.
548
00:37:59,320 --> 00:38:07,360
But yes, you will be talking about these associated topics, these related topics.
549
00:38:07,360 --> 00:38:13,040
The idea being that if you want to rank, if you're producing a blog, we all write blogs
550
00:38:13,040 --> 00:38:20,040
and write articles with the hopes of getting found in search.
551
00:38:20,040 --> 00:38:24,480
Ideally you're doing maybe a topic cluster, you're using pillar content and all of that
552
00:38:24,480 --> 00:38:27,280
kind of good stuff for an SEO model.
553
00:38:27,280 --> 00:38:30,760
But you still want all of the individual articles to be the best that they can be.
554
00:38:30,760 --> 00:38:33,400
You're spending time, effort, energy writing them.
555
00:38:33,400 --> 00:38:39,520
With the research tool within MarketMuse, it basically gives you all of the topics and
556
00:38:39,520 --> 00:38:41,480
themes that you need to be addressing.
557
00:38:41,480 --> 00:38:44,640
It has incredible capabilities.
558
00:38:44,640 --> 00:38:49,440
It will just uncover things that when you take a step back from writing an article,
559
00:38:49,440 --> 00:38:52,200
you go, it's really obvious I should have spoken about that.
560
00:38:52,200 --> 00:38:56,640
I'm writing an article about the Premier League and I didn't mention relegation.
561
00:38:56,640 --> 00:39:01,560
Yeah, of course I need to mention relegation and the relegation zone if people are going
562
00:39:01,560 --> 00:39:02,560
to...
563
00:39:02,560 --> 00:39:03,760
Is that kind of thing.
564
00:39:03,760 --> 00:39:05,960
So it is very good.
565
00:39:05,960 --> 00:39:12,560
I used it for one article a few, it was about 18 months ago.
566
00:39:12,560 --> 00:39:17,520
I wanted to rank for a keyword related to Sostak.
567
00:39:17,520 --> 00:39:21,040
This is what really got me excited about the tool, I should say.
568
00:39:21,040 --> 00:39:24,240
Sostak, digital marketing planning model.
569
00:39:24,240 --> 00:39:26,880
I teach it to my marketing students.
570
00:39:26,880 --> 00:39:29,040
I thought, I've got something to say about that.
571
00:39:29,040 --> 00:39:30,040
I can share some bits.
572
00:39:30,040 --> 00:39:36,640
I've written a few blogs, all of them 1000 to 1500 words long.
573
00:39:36,640 --> 00:39:38,400
None of them made an impact.
574
00:39:38,400 --> 00:39:43,080
Just I might as well have been leaning out of the window and just shouting Sostak and
575
00:39:43,080 --> 00:39:47,560
seeing if anything happened for all the good they did me.
576
00:39:47,560 --> 00:39:54,560
I put an article into MarketMuse and it made clear to me that I needed to go into more
577
00:39:54,560 --> 00:39:55,760
depth in some areas.
578
00:39:55,760 --> 00:39:58,520
I ended up writing an article considerably longer.
579
00:39:58,520 --> 00:40:07,000
It was about 80% longer than the previous article I'd written, so just over 2000 words.
580
00:40:07,000 --> 00:40:10,440
It started to get traction.
581
00:40:10,440 --> 00:40:15,920
Over time it built and built and built because it was a very authoritative piece.
582
00:40:15,920 --> 00:40:24,920
Now that blog accounts for anywhere between 700 to 1200 hits on any given month, just
583
00:40:24,920 --> 00:40:26,720
that one article alone.
584
00:40:26,720 --> 00:40:29,000
I would never have achieved that without MarketMuse.
585
00:40:29,000 --> 00:40:31,400
That's just how you optimize one post.
586
00:40:31,400 --> 00:40:36,280
If you think about that then scaled across your whole website where it's looking at whole
587
00:40:36,280 --> 00:40:40,120
topic gaps in your whole content strategy.
588
00:40:40,120 --> 00:40:44,120
It's looking at your competitors and saying, well, they're here and you've got an opportunity
589
00:40:44,120 --> 00:40:47,680
to get here just by tweaking a few pages.
590
00:40:47,680 --> 00:40:50,200
It's incredibly powerful.
591
00:40:50,200 --> 00:40:56,520
What I like about it compared to the generative AI that we've been talking about today is
592
00:40:56,520 --> 00:41:03,480
that whereas generative AI can be really useful in creating content or giving you ideas, it
593
00:41:03,480 --> 00:41:06,280
doesn't really care about you or your strategy.
594
00:41:06,280 --> 00:41:08,160
It doesn't know your business.
595
00:41:08,160 --> 00:41:10,240
It doesn't know what you're trying to achieve.
596
00:41:10,240 --> 00:41:12,160
It doesn't have any interest in you.
597
00:41:12,160 --> 00:41:15,760
It just responds to a prompt.
598
00:41:15,760 --> 00:41:19,200
MarketMuse is a really strategic tool.
599
00:41:19,200 --> 00:41:25,040
It's strategic and tactical rolled up in one because it helps you set that content strategy
600
00:41:25,040 --> 00:41:30,480
and then when you're executing, it helps optimize all of those individual outputs so that you're
601
00:41:30,480 --> 00:41:37,600
not wasting time writing articles that are a bit of a damp squib really.
602
00:41:37,600 --> 00:41:38,600
Wow.
603
00:41:38,600 --> 00:41:40,400
Let me play that back then.
604
00:41:40,400 --> 00:41:42,040
I'm a marketer.
605
00:41:42,040 --> 00:41:46,560
I'm creating content on my website, blogs or online articles as you mentioned.
606
00:41:46,560 --> 00:41:48,240
I've got some keywords I want to rank for.
607
00:41:48,240 --> 00:41:50,880
I've produced some content in the past and I got crickets.
608
00:41:50,880 --> 00:41:54,480
Didn't rank, didn't get any traffic from Google or other search engines.
609
00:41:54,480 --> 00:42:00,560
I could throw that topic into MarketMuse and MarketMuse would identify almost the subtopics
610
00:42:00,560 --> 00:42:06,680
that I need to expand in my online article or blog post on that topic in order to be
611
00:42:06,680 --> 00:42:11,880
able to be seen as authoritative by Google on the topic and then start ranking.
612
00:42:11,880 --> 00:42:16,960
You've done that in exactly that way for a term that's not easy to rank for where you
613
00:42:16,960 --> 00:42:22,120
did the work the traditional way, the logical way, but you weren't getting anything.
614
00:42:22,120 --> 00:42:27,720
You applied MarketMuse's AI driven intelligence and insights and then you started ranking
615
00:42:27,720 --> 00:42:30,200
and getting all this traffic.
616
00:42:30,200 --> 00:42:31,200
Exactly that.
617
00:42:31,200 --> 00:42:35,800
The process, the actual user interface for doing it is a dream.
618
00:42:35,800 --> 00:42:40,200
With the optimized tool, for instance, you put in the keyword that you want to rank for
619
00:42:40,200 --> 00:42:43,580
and then you can drop in the URL of a blog post that you've already written.
620
00:42:43,580 --> 00:42:46,280
Maybe you've done something that's just not landed.
621
00:42:46,280 --> 00:42:48,880
You then hit fetch and run.
622
00:42:48,880 --> 00:42:50,980
It then scrapes that article.
623
00:42:50,980 --> 00:42:53,360
It runs its research in the backend.
624
00:42:53,360 --> 00:42:59,080
It pulls through your article into a very standard kind of Word document rich text editing
625
00:42:59,080 --> 00:43:04,880
interface and it has at the sidebar all of the research it's pulled up, all of the topics
626
00:43:04,880 --> 00:43:10,880
with suggested distribution and basically how important different keywords are, the
627
00:43:10,880 --> 00:43:12,520
weighting of them.
628
00:43:12,520 --> 00:43:16,240
It highlights them in your text and it says, yeah, you're using this term, but you're not
629
00:43:16,240 --> 00:43:17,440
using this term.
630
00:43:17,440 --> 00:43:24,120
You can then just start to add your H2s in, add a new section, write a new paragraph and
631
00:43:24,120 --> 00:43:26,720
just tick off all of those bits.
632
00:43:26,720 --> 00:43:29,440
Yeah, it's incredibly powerful.
633
00:43:29,440 --> 00:43:31,560
It's guiding you through all of that.
634
00:43:31,560 --> 00:43:35,660
You don't have to be an SEO pro to be able to do this by the sounds of things.
635
00:43:35,660 --> 00:43:36,660
Not at all.
636
00:43:36,660 --> 00:43:45,920
I venture to say that you could use this tool without knowing anything about SEO.
637
00:43:45,920 --> 00:43:51,640
You wouldn't have to know what a meta title is to be able to just use it.
638
00:43:51,640 --> 00:43:52,640
Right.
639
00:43:52,640 --> 00:43:53,640
Cool.
640
00:43:53,640 --> 00:43:58,680
Well, for those of you then that want to rank better, perhaps have even been producing content
641
00:43:58,680 --> 00:44:02,740
on a given topic and haven't got the results that you've been looking for, give Market
642
00:44:02,740 --> 00:44:03,740
Music a go.
643
00:44:03,740 --> 00:44:04,740
Brilliant.
644
00:44:04,740 --> 00:44:08,480
Well, I think that brings us to the end of today's podcast.
645
00:44:08,480 --> 00:44:10,820
Thanks very much for your time and your insights, Martin.
646
00:44:10,820 --> 00:44:14,000
Always an absolute pleasure to hang out with you.
647
00:44:14,000 --> 00:44:15,880
For everybody else, we hope you've enjoyed this.
648
00:44:15,880 --> 00:44:21,800
If you have, please do subscribe, share this with your marketing friends who also need
649
00:44:21,800 --> 00:44:28,840
to know about AI and tell them about the Artificially Intelligent Marketing podcast.
650
00:44:28,840 --> 00:44:31,680
Hopefully we will see you all again next week.
651
00:44:31,680 --> 00:44:34,280
If you want to get in touch with us, please do.
652
00:44:34,280 --> 00:44:36,280
You'll find us on the LinkedIns and the Twitters.
653
00:44:36,280 --> 00:44:39,580
We'd love to hear what you think of the podcast, any thoughts you have, any stories you'd like
654
00:44:39,580 --> 00:44:42,960
us to cover and any feedback you've got in general, really.
655
00:44:42,960 --> 00:44:43,960
Thank you, Martin.
656
00:44:43,960 --> 00:44:44,960
Bye bye.
657
00:44:44,960 --> 00:44:45,960
We'll see you next week.
658
00:44:45,960 --> 00:44:46,960
See you next week.
659
00:44:46,960 --> 00:44:47,960
Cheers.
660
00:44:47,960 --> 00:44:54,000
Thank you for listening to Artificially Intelligent Marketing.
661
00:44:54,000 --> 00:45:00,060
To stay on top of the latest trends, tips and tools in the world of marketing AI, be
662
00:45:00,060 --> 00:45:01,800
sure to subscribe.
663
00:45:01,800 --> 00:45:12,600
We look forward to seeing you again next week.