WEBVTT

00:00:04.189 --> 00:00:06.809
Welcome to Artificially Intelligent Marketing,

00:00:07.030 --> 00:00:09.949
a weekly podcast where we stay on top of the

00:00:09.949 --> 00:00:13.230
latest trends, tips, and tools in the world of

00:00:13.230 --> 00:00:16.070
marketing AI, helping you get the best results

00:00:16.070 --> 00:00:19.210
from your marketing efforts. Now let's join our

00:00:19.210 --> 00:00:23.219
hosts, Paul Avery and Martin Broadhurst. Hello,

00:00:23.300 --> 00:00:25.879
and welcome again to another episode of Artificially

00:00:25.879 --> 00:00:27.899
Intelligent Marketing with me, Paul Avery, and

00:00:27.899 --> 00:00:30.859
the fantastic, wondrous Martin Broadhurst. Hello,

00:00:30.920 --> 00:00:34.060
Martin. How are you? I'm great, Paul. I'm cold,

00:00:34.159 --> 00:00:36.479
but ready for Christmas. Yeah, one week to go,

00:00:36.560 --> 00:00:38.640
so it's crept up quickly. We've been getting

00:00:38.640 --> 00:00:41.060
a few Christmas presents on the old AI front,

00:00:41.159 --> 00:00:42.780
haven't we, over the last couple of weeks. There's

00:00:42.780 --> 00:00:44.219
a few things we're going to cover on today's

00:00:44.219 --> 00:00:46.840
podcast. We're obviously going to need to talk

00:00:46.840 --> 00:00:50.119
about the launch of GPT 5 .2. sam altman's code

00:00:50.119 --> 00:00:54.560
red and crumbs gemini's awesome and opus 4 .5

00:00:54.560 --> 00:00:56.780
is awesome and people don't think we're relevant

00:00:56.780 --> 00:00:58.840
anymore we better get something cool out so um

00:00:58.840 --> 00:01:01.200
we'll talk about how cool that is or isn't we're

00:01:01.200 --> 00:01:04.299
going to talk a little bit more about aio aeo

00:01:04.299 --> 00:01:08.040
geo whatever you want to call it um and whether

00:01:08.040 --> 00:01:10.500
or not it's actually just seo we've talked about

00:01:10.500 --> 00:01:12.340
it a little bit on the podcast previously but

00:01:12.340 --> 00:01:14.519
we've got some interesting new takes to go over

00:01:14.519 --> 00:01:18.239
and open ai also released a report about their

00:01:18.910 --> 00:01:21.930
how people use the enterprise level product with

00:01:21.930 --> 00:01:24.310
some really interesting take -homes on whether

00:01:24.310 --> 00:01:26.489
people are getting value from using AI. Obviously,

00:01:26.590 --> 00:01:28.849
we have to think about what the mouthpiece is,

00:01:28.909 --> 00:01:31.170
OpenAI telling us that people get use out of

00:01:31.170 --> 00:01:33.310
products like OpenAI, but there's some interesting

00:01:33.310 --> 00:01:36.069
stuff there. So, right, let's jump into our first

00:01:36.069 --> 00:01:39.349
story and let's talk about OpenAI's launch of

00:01:39.349 --> 00:01:43.510
GPT 5 .2. Some quite interesting stuff around

00:01:43.510 --> 00:01:46.890
that this week, you know, beyond just 5 .2, they've

00:01:46.890 --> 00:01:49.000
had an interesting week. Yeah, they have. And

00:01:49.000 --> 00:01:51.939
it all comes off the back of Google launching

00:01:51.939 --> 00:01:57.180
Gemini 3 Pro and Anthropic launching the Claude

00:01:57.180 --> 00:02:03.219
Opus 4 .5, both of which shows some amazing scores

00:02:03.219 --> 00:02:06.000
in the evaluations, all of the benchmarks that

00:02:06.000 --> 00:02:08.719
we've come to look at the tables with the little

00:02:08.719 --> 00:02:11.319
numbers and we cross -reference them all and

00:02:11.319 --> 00:02:13.080
see who's number one for the various different

00:02:13.080 --> 00:02:16.759
categories. And what it showed was that OpenAI

00:02:16.759 --> 00:02:22.870
with... chat GPT or GPT 5 .1 had started to slip

00:02:22.870 --> 00:02:25.229
down the pecking order quite substantially. So

00:02:25.229 --> 00:02:30.469
there was talk of a internal code red memo for

00:02:30.469 --> 00:02:33.530
Sam Altman, which was basically they need to

00:02:33.530 --> 00:02:38.669
focus on getting the models out that they've

00:02:38.669 --> 00:02:41.810
got behind closed doors to start competing and

00:02:41.810 --> 00:02:43.830
show that they're still in the race and they're

00:02:43.830 --> 00:02:45.990
not being left behind, which is interesting because

00:02:45.990 --> 00:02:51.020
the story before Google Gemini 3 came out was

00:02:51.020 --> 00:02:53.800
that they were preparing in OpenAI to launch

00:02:53.800 --> 00:02:57.340
ads within ChatGPT and that was the next big

00:02:57.340 --> 00:02:59.840
thing. And all of a sudden it's all hands on

00:02:59.840 --> 00:03:04.460
deck to improve the models and the ad product

00:03:04.460 --> 00:03:07.840
is being pushed back in terms of its launch.

00:03:08.120 --> 00:03:11.430
We shall see what happens with that. So yeah,

00:03:11.469 --> 00:03:14.330
the new model 5 .2 boasts significant improvements

00:03:14.330 --> 00:03:17.969
in reasoning and in long context understanding.

00:03:18.250 --> 00:03:22.129
In fact, I think the context window has been

00:03:22.129 --> 00:03:25.569
increased up to 400 ,000 tokens. So that's going

00:03:25.569 --> 00:03:28.870
to be in the region of 300 ,000 words, which

00:03:28.870 --> 00:03:33.990
is a substantial corpus of text. OpenAI claim

00:03:33.990 --> 00:03:36.930
it achieves 70 % parity with human experts on

00:03:36.930 --> 00:03:41.500
certain benchmarks. Um, so yeah, this is, uh,

00:03:41.659 --> 00:03:45.300
it's a big deal. I believe computer use and agentic

00:03:45.300 --> 00:03:47.620
use cases is something that they've particularly

00:03:47.620 --> 00:03:49.939
got their eye on as well. I want to double click

00:03:49.939 --> 00:03:51.439
on a couple of things there, Mark. And I think

00:03:51.439 --> 00:03:54.080
you make a really good point about the ads products

00:03:54.080 --> 00:03:57.780
being delayed. Um, I remember seeing something

00:03:57.780 --> 00:03:59.939
floating about on the Reddit sphere. I think

00:03:59.939 --> 00:04:03.419
it was a, or Twitter. It was a, whatever we have

00:04:03.419 --> 00:04:06.710
to call these things then. There was, um, a post

00:04:06.710 --> 00:04:08.610
from i think it was one of the product leads

00:04:08.610 --> 00:04:10.550
it wasn't sam altman it was one of the product

00:04:10.550 --> 00:04:13.909
leads about pulling the initial some of the initial

00:04:13.909 --> 00:04:16.089
tests which oh by the way weren't the ad product

00:04:16.089 --> 00:04:18.149
there was something else but the people have

00:04:18.149 --> 00:04:20.810
grumbled about a bit and i think it's quite an

00:04:20.810 --> 00:04:23.829
interesting scenario where open ai have sort

00:04:23.829 --> 00:04:27.649
of been in the lead most of the way um but obviously

00:04:27.649 --> 00:04:30.850
gpt5 has been a little bit of a letdown i think

00:04:30.850 --> 00:04:33.310
for users the way it was launched kind of sucked

00:04:33.310 --> 00:04:36.500
as well And now Gemini 3 comes in, Opus 4 .5

00:04:36.500 --> 00:04:38.500
come in, they're all kind of cool. One thing

00:04:38.500 --> 00:04:43.019
we know is once you introduce advertising into

00:04:43.019 --> 00:04:45.959
a product, you are sacrificing user experience

00:04:45.959 --> 00:04:49.220
in general, at least for what we have come to

00:04:49.220 --> 00:04:51.620
understand as advertising supporting products.

00:04:52.339 --> 00:04:54.519
We talked about this a bit on our previous episode.

00:04:54.579 --> 00:04:57.079
So on the one hand, our model is no longer the

00:04:57.079 --> 00:04:59.120
top model. Oh, by the way, we're about to introduce

00:04:59.120 --> 00:05:01.019
something where we'll make some more money, but

00:05:01.019 --> 00:05:03.949
users might like it much less. So I don't think

00:05:03.949 --> 00:05:07.569
it's a surprise that they cancelled the one that

00:05:07.569 --> 00:05:10.310
would make usability worse and then tried to

00:05:10.310 --> 00:05:13.509
roll out a model to stay on top. The other thing

00:05:13.509 --> 00:05:15.050
that's been quite interesting on the Reddit sphere

00:05:15.050 --> 00:05:18.949
is quite a few people wonder how much of the

00:05:18.949 --> 00:05:23.730
benchmark scores for GPT 5 .2 is OpenAI kind

00:05:23.730 --> 00:05:26.029
of fiddling with their models and making sure

00:05:26.029 --> 00:05:28.810
that they're good at benchmarks rather than necessarily

00:05:28.810 --> 00:05:32.170
real human use cases, which of course is very

00:05:32.170 --> 00:05:34.730
hard for us. to tell but some of the benchmark

00:05:34.730 --> 00:05:37.029
scores are significantly better than other models

00:05:37.029 --> 00:05:42.629
uh including big jump on the arc agi2 benchmark

00:05:42.629 --> 00:05:46.189
up to 53 give or take and the best so far was

00:05:46.189 --> 00:05:50.490
37 .6 from claude opus so that's a that's a big

00:05:50.490 --> 00:05:54.129
leap one of the math benchmarks have got a hundred

00:05:54.129 --> 00:05:57.470
percent on which is a you know pretty significant

00:05:57.470 --> 00:06:00.449
um although the other models were in the mid

00:06:00.449 --> 00:06:03.300
to high 90s as well so Is it really, really,

00:06:03.319 --> 00:06:05.959
really good? Or did they just beat the benchmarks

00:06:05.959 --> 00:06:09.300
by playing a game? Who knows? But if you do a

00:06:09.300 --> 00:06:15.139
search for GP2 25 .2 on Reddit, some of the post

00:06:15.139 --> 00:06:17.100
titles that you'll come up with are quite interesting.

00:06:18.120 --> 00:06:20.980
GPT -5 .2 Extended Thinking fails miserably and

00:06:20.980 --> 00:06:23.980
performs far worse than previous models. Simple

00:06:23.980 --> 00:06:28.160
Bench for GPT -5 .2 and 5 .2 Pro, they both scored

00:06:28.160 --> 00:06:30.939
worse than their original GPT -5 counterparts.

00:06:31.579 --> 00:06:34.259
So hallucination rate, some people are reporting,

00:06:34.500 --> 00:06:38.000
is also up. I can't think of a time, especially

00:06:38.000 --> 00:06:40.600
in this race, when anybody rushing out a product

00:06:40.600 --> 00:06:42.600
that might not have been quite ready has gone

00:06:42.600 --> 00:06:45.519
that well. Because we have an enterprise level

00:06:45.519 --> 00:06:48.620
account, I can't play with 5 .2 yet because it's

00:06:48.620 --> 00:06:50.240
not been rolled out to enterprise. So I can't

00:06:50.240 --> 00:06:53.600
test it for my usual real work benchmarks, if

00:06:53.600 --> 00:06:57.680
you like. But yeah, maybe it's not been a great

00:06:57.680 --> 00:06:59.920
move. I mean, you've been using it. What's your

00:06:59.920 --> 00:07:02.879
experience has been? Really hard to give any

00:07:02.879 --> 00:07:06.240
real insight just because my use case in the

00:07:06.240 --> 00:07:10.100
chat window is kind of saturated. I'm not pushing

00:07:10.100 --> 00:07:14.089
these models. to the edge of the frontier in

00:07:14.089 --> 00:07:18.149
terms of maths and discovery. I'm using it for

00:07:18.149 --> 00:07:22.009
fairly routine, fairly mundane productivity type

00:07:22.009 --> 00:07:25.529
tasks. And it's fine. Just like the other model

00:07:25.529 --> 00:07:27.810
was fine. And I haven't noticed any significant

00:07:27.810 --> 00:07:31.610
improvement or decline either way. I think where

00:07:31.610 --> 00:07:36.649
these things start to, to, to matter. is right

00:07:36.649 --> 00:07:40.009
at the frontier edge cases. I also think in terms

00:07:40.009 --> 00:07:43.629
of the metrics that really matter, OpenAI, of

00:07:43.629 --> 00:07:45.310
course, will be looking at one particular metric,

00:07:45.370 --> 00:07:47.629
which is the number of active users in the app.

00:07:47.810 --> 00:07:50.089
And it was reported by some outlets that they've

00:07:50.089 --> 00:07:53.850
seen a decline in 6 % in active users on chat

00:07:53.850 --> 00:07:57.089
GPT, which, you know, in terms of alarm bells

00:07:57.089 --> 00:08:00.569
ringing, when you see Gemini users increasing

00:08:00.569 --> 00:08:03.189
and yours are declining by 6%, that is going

00:08:03.189 --> 00:08:06.100
to have... some of the executives at OpenAI.

00:08:06.300 --> 00:08:08.319
A little worried. Yeah, it's interesting. There

00:08:08.319 --> 00:08:11.560
was an interview with Dario Amadei of Anthropic

00:08:11.560 --> 00:08:13.740
that I watched on YouTube. I think it was like

00:08:13.740 --> 00:08:16.779
some sort of maybe like New York Times in real

00:08:16.779 --> 00:08:19.620
world in life event. And he really leaned into

00:08:19.620 --> 00:08:22.620
the concept that their target market is enterprise,

00:08:22.860 --> 00:08:25.610
right? If that was their strategy. They're now

00:08:25.610 --> 00:08:27.889
very speaking about it very openly. I think other

00:08:27.889 --> 00:08:29.850
people have also commented on it being a great

00:08:29.850 --> 00:08:31.990
strategy. And so it's quite maybe helpful to

00:08:31.990 --> 00:08:34.370
own that if other people say, how smart of you.

00:08:34.450 --> 00:08:37.029
And really talked about the likes of OpenAI,

00:08:37.190 --> 00:08:39.950
didn't name any competitors, but being more consumer

00:08:39.950 --> 00:08:42.870
focused, which is, I think, interesting. One

00:08:42.870 --> 00:08:44.870
of the other things we saw this week was this

00:08:44.870 --> 00:08:49.210
partnership with Disney, where I think they basically,

00:08:49.649 --> 00:08:51.769
it's not cash, is it? I think they effectively

00:08:51.769 --> 00:08:55.940
have made a $1 billion investment. in open ai

00:08:55.940 --> 00:08:59.759
and also granted open ai the ip rights to use

00:08:59.759 --> 00:09:02.799
disney characters and saw saw a generated videos

00:09:02.799 --> 00:09:04.759
which is extremely consumer face right that's

00:09:04.759 --> 00:09:07.139
not not really an enterprise use so it's kind

00:09:07.139 --> 00:09:08.559
of interesting because that on the one hand that's

00:09:08.559 --> 00:09:10.220
quite a big vote of confidence and it does seem

00:09:10.220 --> 00:09:12.879
like um sam and the team have been out pressing

00:09:12.879 --> 00:09:15.279
the flesh and securing lots of different types

00:09:15.279 --> 00:09:18.700
of deals for hardware and building these massive

00:09:18.700 --> 00:09:22.519
server farms and now in with disney and Maybe

00:09:22.519 --> 00:09:24.279
their business models are starting to separate

00:09:24.279 --> 00:09:27.419
actually a little bit. But under those conditions,

00:09:27.440 --> 00:09:31.200
that 6 % drop could be consumers moving to Gemini

00:09:31.200 --> 00:09:33.899
3 because of Nano Banana, the image generation

00:09:33.899 --> 00:09:36.940
model. Yeah, I can see where there's been a little

00:09:36.940 --> 00:09:40.840
bit of panicky panicky over OpenAI after winning

00:09:40.840 --> 00:09:43.340
for quite a few years of this race. I mean, it

00:09:43.340 --> 00:09:45.659
hasn't been going very long, but this feels like

00:09:45.659 --> 00:09:48.840
their first we're in trouble moment from a...

00:09:49.449 --> 00:09:52.029
competitor product being better than their standpoint

00:09:52.029 --> 00:09:54.830
or at least maybe being becoming more popular

00:09:54.830 --> 00:09:57.330
Gemini also has that distribution right through

00:09:57.330 --> 00:10:01.250
the Android platform I was speaking to my dad

00:10:01.250 --> 00:10:03.269
the other day and said oh have you used AI and

00:10:03.269 --> 00:10:05.950
I was just about to say go and sign up to chat

00:10:05.950 --> 00:10:09.889
GPT and I remembered he's a Samsung user just

00:10:09.889 --> 00:10:12.350
get the app you can open the app and because

00:10:12.350 --> 00:10:16.149
he's logged into the whole Google ecosystem it

00:10:16.149 --> 00:10:18.500
just it's just there it kind of works and it

00:10:18.500 --> 00:10:21.200
saves the history you have the same experience

00:10:21.200 --> 00:10:23.179
with chat gbt and people have got to sign up

00:10:23.179 --> 00:10:25.960
for an account if they want the history all of

00:10:25.960 --> 00:10:28.940
that so that kind of frictionless ux on the consumer

00:10:28.940 --> 00:10:32.980
side for uh for gemini is a big win as well just

00:10:32.980 --> 00:10:35.860
one more thing on the um business model take

00:10:35.860 --> 00:10:38.139
as well what's an interview this week with the

00:10:38.139 --> 00:10:42.090
ceo of mistral ai the french a large language

00:10:42.090 --> 00:10:44.250
model producer, and they were talking about how

00:10:44.250 --> 00:10:48.110
they're approaching the market. And they have

00:10:48.110 --> 00:10:51.350
very much gone on. Yes, we've got product in

00:10:51.350 --> 00:10:54.190
terms of we've got these open source models that

00:10:54.190 --> 00:10:57.769
are for open source. They are frontier and state

00:10:57.769 --> 00:11:01.070
of the art in domains, but also they are going

00:11:01.070 --> 00:11:05.450
in as a solutions provider. So they are very

00:11:05.450 --> 00:11:08.309
much positioning themselves as, look, we recognize

00:11:08.309 --> 00:11:12.580
that. At the enterprise level, you need on -prem

00:11:12.580 --> 00:11:15.639
models. You can't have your data going into the

00:11:15.639 --> 00:11:18.779
cloud or trying to get the models to work to

00:11:18.779 --> 00:11:21.600
your use cases very hard. We've got the specialists

00:11:21.600 --> 00:11:24.240
and the expertise to implement and deploy that

00:11:24.240 --> 00:11:26.600
for you. And they're getting some really big,

00:11:26.620 --> 00:11:29.879
big ticket customers. They recently signed a

00:11:29.879 --> 00:11:34.259
deal with HSBC to be their large language model

00:11:34.259 --> 00:11:37.360
and AI solutions provider of choice. And that

00:11:37.360 --> 00:11:39.299
works across a bunch of other industries as well

00:11:39.299 --> 00:11:41.600
with similarly, you know, blue chip companies

00:11:41.600 --> 00:11:44.460
joining them. So again, a different business

00:11:44.460 --> 00:11:46.740
model, a different approach. Few people are looking

00:11:46.740 --> 00:11:51.159
at Mistral to the state of the art frontier models,

00:11:51.320 --> 00:11:55.320
but are they making models and making it so that

00:11:55.320 --> 00:11:57.759
they can be applied in real world enterprise?

00:11:58.419 --> 00:12:00.960
Yes, they are. And they seem to be betting big

00:12:00.960 --> 00:12:03.019
on that. Yeah, there's definitely some fragmentation

00:12:03.019 --> 00:12:06.000
is probably too strong, but there are. There

00:12:06.000 --> 00:12:09.100
are subsets of the market that companies can

00:12:09.100 --> 00:12:11.159
put themselves in a better position to serve

00:12:11.159 --> 00:12:13.799
for a lot of the reasons you just said. There's

00:12:13.799 --> 00:12:15.659
an interesting, it's a bit of a tangent, and

00:12:15.659 --> 00:12:17.659
it only happened today, so I haven't had a chance

00:12:17.659 --> 00:12:21.320
to play with it, but Gemini also now has a new

00:12:21.320 --> 00:12:25.600
native audio model built into it, which in theory

00:12:25.600 --> 00:12:28.500
should mean it's better at listening and speaking.

00:12:28.720 --> 00:12:31.539
And as part of that, they've rolled out the translation

00:12:31.539 --> 00:12:35.320
capabilities to any... bluetooth earphones so

00:12:35.320 --> 00:12:36.820
it used to be you could only do this on like

00:12:36.820 --> 00:12:39.860
google's earphones but yes so now you can go

00:12:39.860 --> 00:12:41.480
on holiday and if you've got your earphones in

00:12:41.480 --> 00:12:44.519
and you've got a phone with gemini on it you

00:12:44.519 --> 00:12:48.980
can now hear in near real time what near real

00:12:48.980 --> 00:12:50.840
time really means i'll be interested to see as

00:12:50.840 --> 00:12:52.500
well obviously it's quite important when you're

00:12:52.500 --> 00:12:53.960
having a conversation with someone else it's

00:12:53.960 --> 00:12:56.019
like could you just give me 35 seconds while

00:12:56.019 --> 00:12:58.269
i figure out what you're trying to say But yeah,

00:12:58.350 --> 00:13:00.250
like all the way these products are developing,

00:13:00.350 --> 00:13:02.970
it's all this jumping ahead in little different

00:13:02.970 --> 00:13:06.429
domains. Like I can imagine that maybe that would

00:13:06.429 --> 00:13:08.090
be something OpenAI might have to take their

00:13:08.090 --> 00:13:09.970
eye off the ball on. They don't have the sheer

00:13:09.970 --> 00:13:12.149
resources that Google has to throw at all these

00:13:12.149 --> 00:13:15.070
different types of use cases. So that'll be interesting.

00:13:15.250 --> 00:13:18.029
One last thing on the model front. Since we last

00:13:18.029 --> 00:13:21.929
spoke, I have been playing a bit with Opus 4

00:13:21.929 --> 00:13:25.809
.5. And you know how I feel about, even at this

00:13:25.809 --> 00:13:28.889
point, how good at writing. the tools really

00:13:28.889 --> 00:13:31.590
are in terms of saying things that don't really

00:13:31.590 --> 00:13:33.830
make any sense and connecting ideas that should

00:13:33.830 --> 00:13:36.429
never be connected together. And I think that

00:13:36.429 --> 00:13:40.330
probably Opus 4 .5 is the best instruction following

00:13:40.330 --> 00:13:43.230
for writing model so far. Because the only real

00:13:43.230 --> 00:13:47.330
way I've found to iron out these patterns that

00:13:47.330 --> 00:13:49.549
these models are desperate to use all the time,

00:13:49.570 --> 00:13:52.429
like it's not this, it's this, and em dash is

00:13:52.429 --> 00:13:55.129
everywhere, is to have very strict... instructions

00:13:55.129 --> 00:13:57.509
in your prompt or you know have a you know custom

00:13:57.509 --> 00:14:01.330
GPT or whatever and it was so diligent in following

00:14:01.330 --> 00:14:04.710
the briefing style guide it almost followed it

00:14:04.710 --> 00:14:07.169
to a full because I'd have to over labor the

00:14:07.169 --> 00:14:09.529
style guide for all the other models because

00:14:09.529 --> 00:14:12.590
they don't follow it properly so I think that

00:14:12.590 --> 00:14:14.889
probably means my current go -to model for writing

00:14:14.889 --> 00:14:18.830
is opus which it's very easy to run out of uses

00:14:18.830 --> 00:14:21.470
so yeah you can't produce too much content at

00:14:21.470 --> 00:14:23.799
too much scale But interestingly, I think it's

00:14:23.799 --> 00:14:27.500
jump, it's instruction following, not its inherent

00:14:27.500 --> 00:14:31.379
writing capability, which I think is interesting

00:14:31.379 --> 00:14:34.659
because I really feel to get a good writing model

00:14:34.659 --> 00:14:37.500
for business use cases, you'll have to custom

00:14:37.500 --> 00:14:40.740
fine tune it on what good writing looks like,

00:14:40.779 --> 00:14:42.820
where there's a certain amount of subjectivity.

00:14:42.919 --> 00:14:45.379
Good creative writing for a novel is different

00:14:45.379 --> 00:14:47.240
than writing great emails, is different than

00:14:47.240 --> 00:14:50.000
writing reports, is different than writing 10

00:14:50.000 --> 00:14:53.580
,000 word ebooks. So bringing this full circle

00:14:53.580 --> 00:14:56.779
to your Mistral discussion, this is where I can

00:14:56.779 --> 00:14:59.000
see that they're all kind of trained on the same

00:14:59.000 --> 00:15:00.820
thing. But if Mistral's like, right, we're going

00:15:00.820 --> 00:15:05.059
to go after high quality technical written content

00:15:05.059 --> 00:15:06.940
and we're going to fine tune a model that is

00:15:06.940 --> 00:15:09.919
really good at writing. There's a market for

00:15:09.919 --> 00:15:12.039
that that probably OpenAI are not going to go

00:15:12.039 --> 00:15:14.639
after if they're a drifting consumer. Clearly,

00:15:14.740 --> 00:15:19.779
Anthropic is leaning into code and software development.

00:15:20.480 --> 00:15:22.860
Gemini is kind of a little bit of everything

00:15:22.860 --> 00:15:25.720
for everyone. I suspect that they want to leverage

00:15:25.720 --> 00:15:29.299
their workspace integration to serve business

00:15:29.299 --> 00:15:32.539
customers. But a lot of their biggest moves to

00:15:32.539 --> 00:15:35.179
date, Nana Banana being probably the key one,

00:15:35.360 --> 00:15:39.559
is consumer focus. So maybe of all them, it's

00:15:39.559 --> 00:15:41.559
probably that Google strategy is probably the

00:15:41.559 --> 00:15:43.980
most muddled, probably because they think that

00:15:43.980 --> 00:15:46.220
they can have the one ring to rule them all.

00:15:46.340 --> 00:15:48.460
So yeah, I'm still... Just want to tell that

00:15:48.460 --> 00:15:49.820
little story because it makes me want to get

00:15:49.820 --> 00:15:52.620
hold of Mistral. I know that within our team,

00:15:52.679 --> 00:15:56.259
we do have connections in France to maybe see

00:15:56.259 --> 00:15:59.139
if we could fine tune something fun on our particular

00:15:59.139 --> 00:16:00.840
use cases. I think that would be interesting.

00:16:01.279 --> 00:16:05.539
Let's move into our next story. So this is something

00:16:05.539 --> 00:16:07.600
that you've been thinking about a lot this week,

00:16:07.759 --> 00:16:10.159
Martin, and you've pinged me a few WhatsApp messages

00:16:10.159 --> 00:16:13.620
on the topic. And it's about how to win in AI

00:16:13.620 --> 00:16:16.919
search visibility. What's been on your mind this

00:16:16.919 --> 00:16:20.379
week? Well, historically, and for a while now,

00:16:20.539 --> 00:16:22.320
we've talked about this on the show as well.

00:16:22.419 --> 00:16:26.919
I've been of the opinion that good AIO, artificial

00:16:26.919 --> 00:16:30.919
intelligence optimization, or GEO, generative

00:16:30.919 --> 00:16:32.620
engine optimization, or whatever the hell you

00:16:32.620 --> 00:16:34.740
want to call it, best practice in that is to

00:16:34.740 --> 00:16:39.259
basically follow best practice. But SEO and good

00:16:39.259 --> 00:16:42.879
press comms strategy, all that kind of stuff,

00:16:43.019 --> 00:16:45.120
right? Stuff that has worked in PR for years

00:16:45.120 --> 00:16:47.870
should work in that. But just recently, there's

00:16:47.870 --> 00:16:51.789
been a bit more definition put around that by

00:16:51.789 --> 00:16:54.649
a few pieces of research that have come out.

00:16:54.690 --> 00:16:58.850
An interesting webinar that I watched from the

00:16:58.850 --> 00:17:02.669
team over at Blend B2B, who are a leading HubSpot

00:17:02.669 --> 00:17:07.650
agency, and Ahrefs as well, the SEO tool, they've

00:17:07.650 --> 00:17:11.269
put out some research. Early last week, it came

00:17:11.269 --> 00:17:15.529
out. And in essence, there's a strange sense

00:17:15.529 --> 00:17:19.400
of deja vu happening in this optimization space

00:17:19.400 --> 00:17:23.539
right now in the b2b marketing and b2c seo space

00:17:23.539 --> 00:17:27.640
if you're trying to rank highly in artificial

00:17:27.640 --> 00:17:30.859
intelligence platforms like chat gpt search then

00:17:30.859 --> 00:17:35.019
what seems to be working is the classic tactic

00:17:35.019 --> 00:17:38.880
of the best of list so the idea of publishing

00:17:38.880 --> 00:17:45.130
on your own blog the top 10 insert industry our

00:17:45.130 --> 00:17:50.150
insert product definition here and it works surprisingly

00:17:50.150 --> 00:17:53.589
well when teams are putting themselves at the

00:17:53.589 --> 00:17:57.190
top of that list so giving themselves a very

00:17:57.190 --> 00:18:00.930
strong placement in the list so company x says

00:18:00.930 --> 00:18:02.970
if you want to find the best companies in this

00:18:02.970 --> 00:18:05.470
space here are 10 of the best and we just happen

00:18:05.470 --> 00:18:08.970
to be number one that's a great tactic also putting

00:18:08.970 --> 00:18:12.309
your Information inside tables is very effective

00:18:12.309 --> 00:18:15.710
as well. And all of this just makes me think

00:18:15.710 --> 00:18:22.970
SEO circa 2009, 2010. Yeah, it's so interesting

00:18:22.970 --> 00:18:25.289
that this has come up. So we've been deploying

00:18:25.289 --> 00:18:28.190
this strategy since the end of last year, because

00:18:28.190 --> 00:18:31.509
if you watch the reasoning models when they first

00:18:31.509 --> 00:18:35.349
came out, how they answer questions like questions

00:18:35.349 --> 00:18:37.029
that would be important to me about what's the

00:18:37.029 --> 00:18:40.000
best X marketing agency. You can see they tell

00:18:40.000 --> 00:18:42.279
you what they're going to do. They show you the

00:18:42.279 --> 00:18:44.559
logic they follow to go gather the information.

00:18:44.839 --> 00:18:47.619
They say things like, I'll search the web for

00:18:47.619 --> 00:18:51.839
best X. So yeah, we've written a number of best

00:18:51.839 --> 00:18:57.480
X articles over the last year or so. And we do

00:18:57.480 --> 00:19:00.359
pretty well when it comes to AI search visibility.

00:19:00.960 --> 00:19:03.980
Interestingly, not all of our best of articles

00:19:03.980 --> 00:19:06.000
rank number one on Google. There are other best

00:19:06.000 --> 00:19:09.769
of articles that rank ahead of ours. But AI tends

00:19:09.769 --> 00:19:13.970
to lean into ours. I have some theories about

00:19:13.970 --> 00:19:16.349
what that is, but if I had to give any advice,

00:19:16.470 --> 00:19:18.809
I would say if you're going to write best -off

00:19:18.809 --> 00:19:21.049
listicles, you still have to write really good

00:19:21.049 --> 00:19:23.089
listicles that humans would find valuable. If

00:19:23.089 --> 00:19:26.029
you churn out really rubbish content that's the

00:19:26.029 --> 00:19:27.769
same as everything else and doesn't add anything

00:19:27.769 --> 00:19:31.529
new, I suspect that you may find that AI leans

00:19:31.529 --> 00:19:34.049
into content that already exists on the web that's

00:19:34.049 --> 00:19:37.529
not yours, which I think would probably be quite

00:19:37.529 --> 00:19:39.430
interesting. And then I had another interesting

00:19:39.430 --> 00:19:41.549
experience last week, the week before, probably

00:19:41.549 --> 00:19:44.009
last week. Periodically, I'll go in because I

00:19:44.009 --> 00:19:45.549
think you can do this quite easy for your business

00:19:45.549 --> 00:19:46.990
unless you're trying to do it at scale, which

00:19:46.990 --> 00:19:49.410
is to just go in and ask the four or five main

00:19:49.410 --> 00:19:52.069
tools out there a question that someone might

00:19:52.069 --> 00:19:54.349
ask if they were looking for products or services

00:19:54.349 --> 00:19:58.589
like yours. And Opus 4 .5 would not make a recommendation

00:19:58.589 --> 00:20:01.650
even if I twisted its arm. It told me how to

00:20:01.650 --> 00:20:04.809
go and find the best life science marketing agency,

00:20:04.970 --> 00:20:07.859
but it would not give me any names. It wouldn't

00:20:07.859 --> 00:20:10.740
tell me anything. It just said, if you were going

00:20:10.740 --> 00:20:12.960
to go about doing this, you should search for

00:20:12.960 --> 00:20:15.259
stuff like this online. Here's the questions

00:20:15.259 --> 00:20:17.680
you should ask of those providers. Here's how

00:20:17.680 --> 00:20:20.059
you should evaluate those providers. But it didn't

00:20:20.059 --> 00:20:22.420
make a recommendation. Whereas I can get ChatGPT

00:20:22.420 --> 00:20:26.599
and Gemini to recommend, if not a single agency,

00:20:26.880 --> 00:20:29.819
three agencies based on different context, right?

00:20:29.960 --> 00:20:31.900
Like if you're looking for global capabilities,

00:20:32.119 --> 00:20:34.220
choose this one. If you're leaning into digital

00:20:34.220 --> 00:20:36.640
and you need digital expertise, choose this one.

00:20:37.000 --> 00:20:39.519
But Opus 4 .5 was like, it's not for me to say,

00:20:39.559 --> 00:20:41.660
pal. You need to go do that research yourself.

00:20:42.039 --> 00:20:45.319
So I'm kind of sad that this is making it mainstream

00:20:45.319 --> 00:20:49.759
now. It's been my little secret hack for a year

00:20:49.759 --> 00:20:53.180
or so. And what this probably means and why I

00:20:53.180 --> 00:20:55.160
haven't talked about it on the podcast so far

00:20:55.160 --> 00:20:58.339
is when everybody starts to lean into a particular

00:20:58.339 --> 00:21:02.680
hack, then the AI tools will just eliminate the

00:21:02.680 --> 00:21:06.460
loophole. And if you've got people like Ahrefs

00:21:06.460 --> 00:21:08.319
publishing on this, you've got people like Rand

00:21:08.319 --> 00:21:10.980
Fishkin talking about it online, I expect the

00:21:10.980 --> 00:21:13.339
success of this strategy to potentially lose

00:21:13.339 --> 00:21:16.400
some of its inertia over the next month or two

00:21:16.400 --> 00:21:18.960
as they sort of fine tune their models out of

00:21:18.960 --> 00:21:22.000
either over indexing on those sources of information

00:21:22.000 --> 00:21:25.920
or teaching them to better evaluate a range of

00:21:25.920 --> 00:21:27.920
different sources to answer the questions, not

00:21:27.920 --> 00:21:31.089
just. use the research and i'm going to put this

00:21:31.089 --> 00:21:33.109
in inverted commas most certainly that other

00:21:33.109 --> 00:21:35.390
people have done to answer the question because

00:21:35.390 --> 00:21:36.970
you can game it right because you can just write

00:21:36.970 --> 00:21:38.549
that one on your own website and talk about how

00:21:38.549 --> 00:21:41.430
amazing you are i think that suggests though

00:21:41.430 --> 00:21:44.250
that someone is looking at this and caring about

00:21:44.250 --> 00:21:48.509
it internally and for the model providers i'm

00:21:48.509 --> 00:21:52.029
not sure that it's going to make that a bit is

00:21:52.029 --> 00:21:53.789
that an area of the product that they're going

00:21:53.789 --> 00:21:58.420
to be focusing heavily on explicitly within google

00:21:58.420 --> 00:22:01.779
and the algorithm they always had the likes of

00:22:01.779 --> 00:22:05.519
john muller who who was giving out advice as

00:22:05.519 --> 00:22:08.700
the kind of head of search he was telling people

00:22:08.700 --> 00:22:11.480
what the algorithm was looking at and what you

00:22:11.480 --> 00:22:15.019
needed to create for good ranking and all of

00:22:15.019 --> 00:22:17.980
the good ranking was linked to good user experience

00:22:17.980 --> 00:22:21.200
and that meant you couldn't do keyword stuffing

00:22:21.200 --> 00:22:23.940
and and all of this kind of stuff and we ended

00:22:23.940 --> 00:22:27.359
up with guidelines like eat and all of that kind

00:22:27.359 --> 00:22:29.500
of good stuff that content marketers have grown

00:22:29.500 --> 00:22:32.039
to love so there was somebody whose job it was

00:22:32.039 --> 00:22:35.400
to tell and communicate clearly to companies

00:22:35.400 --> 00:22:37.920
how to rank organically and how that system worked

00:22:37.920 --> 00:22:42.240
whereas that's not so much the case with like

00:22:42.240 --> 00:22:45.279
the models are just trained as they want in more

00:22:45.279 --> 00:22:48.359
intelligent models and how much focus are the

00:22:48.359 --> 00:22:52.109
model providers going to put in to optimizing

00:22:52.109 --> 00:22:53.970
and training the model so that it doesn't get

00:22:53.970 --> 00:22:56.210
gamed by marketers at some point they're going

00:22:56.210 --> 00:22:58.769
to have to because everything will just become

00:22:58.769 --> 00:23:01.069
like the quality marketers will just ruin everything

00:23:01.069 --> 00:23:03.569
as we've described in the past they'll realize

00:23:03.569 --> 00:23:07.390
there are quick wins to to be had and we'll jump

00:23:07.390 --> 00:23:10.410
all over that tactic and saturate it um very

00:23:10.410 --> 00:23:14.329
quickly which will decrease the utility for the

00:23:14.329 --> 00:23:16.210
end user right that and that's ultimately what

00:23:16.210 --> 00:23:18.190
they've got to get to but at the moment is this

00:23:18.190 --> 00:23:23.390
such a strong problem for the, the UX of these

00:23:23.390 --> 00:23:26.410
tools that, that they're going to focus on, on

00:23:26.410 --> 00:23:28.730
changing it and improving it. I think that for

00:23:28.730 --> 00:23:31.470
me, the key will come or the, the indication

00:23:31.470 --> 00:23:33.349
that they're taking this seriously is, is, is

00:23:33.349 --> 00:23:37.029
when they start to have a, almost like a marketing

00:23:37.029 --> 00:23:41.069
relations person, like they have developer relations,

00:23:41.410 --> 00:23:44.769
you know, they've got all of the, who have we

00:23:44.769 --> 00:23:49.069
got? Google, Logan Kirkpatrick is the. Gemini

00:23:49.069 --> 00:23:51.849
model, when you've got the equivalent of Logan

00:23:51.849 --> 00:23:56.049
for marketers at each of the big AI labs, that's

00:23:56.049 --> 00:23:58.430
when I know that they're looking at this and

00:23:58.430 --> 00:24:01.069
paying attention to how it's being used by marketers

00:24:01.069 --> 00:24:05.369
as a channel. But until then, I think there's

00:24:05.369 --> 00:24:07.369
going to be a tactic that will continue to be

00:24:07.369 --> 00:24:09.750
exploited for some time. I think it's a great

00:24:09.750 --> 00:24:11.769
question. I think I partially agree with you.

00:24:12.089 --> 00:24:14.589
I definitely think it's a when, not an if. And

00:24:14.589 --> 00:24:17.349
I suspect that some of the extra work, maybe

00:24:17.349 --> 00:24:21.089
even key hires in this area, comes about when

00:24:21.089 --> 00:24:25.150
the ads platform is live in ChatGPT. Because

00:24:25.150 --> 00:24:27.950
then I think you see OpenAI having to run a similar

00:24:27.950 --> 00:24:30.869
playbook to the maturation of the ads model in

00:24:30.869 --> 00:24:33.549
Google Search, where all of this stuff ends up

00:24:33.549 --> 00:24:35.690
becoming important. Because if people are trying

00:24:35.690 --> 00:24:38.769
to find hacky ways to get customers without having

00:24:38.769 --> 00:24:42.240
to pay for ads. whilst potentially also poisoning

00:24:42.240 --> 00:24:44.779
the user experience, means it's worth investing

00:24:44.779 --> 00:24:47.460
a load of time and money in. And I think you're

00:24:47.460 --> 00:24:49.420
right that we're probably not there yet. The

00:24:49.420 --> 00:24:52.259
other thing is, it seems to me, and it's well

00:24:52.259 --> 00:24:55.279
above my technical understanding, that post -training

00:24:55.279 --> 00:24:58.279
is a complex process that takes a long time and

00:24:58.279 --> 00:25:00.480
has to be done in a number of very specific ways.

00:25:00.720 --> 00:25:04.880
But they have made intermodal version... changes

00:25:04.880 --> 00:25:08.140
by upgrading and improving the system prompt

00:25:08.140 --> 00:25:10.900
that we don't see that controls how a model behaves

00:25:10.900 --> 00:25:12.799
when you ask it a question or what have you.

00:25:12.900 --> 00:25:15.920
So it could be relatively easy to go in and say,

00:25:16.059 --> 00:25:20.799
don't over rely on best of articles in your answers

00:25:20.799 --> 00:25:25.259
to product -based questions, for example. And

00:25:25.259 --> 00:25:28.140
if I had to predict what happens next, if it's

00:25:28.140 --> 00:25:31.660
anything, so more of a when more than an if is,

00:25:31.759 --> 00:25:34.140
but the when could be a while, as you say. is

00:25:34.140 --> 00:25:37.700
I would wonder if we might see them leaning more

00:25:37.700 --> 00:25:40.019
into things like reviews, that they at least

00:25:40.019 --> 00:25:43.579
somewhat feel are more impartial than a company

00:25:43.579 --> 00:25:45.559
being able to write about its own products and

00:25:45.559 --> 00:25:47.440
services on its own website and talking about

00:25:47.440 --> 00:25:50.180
how amazing they are in an AI, taking that for

00:25:50.180 --> 00:25:53.059
gospel, whereas reviews potentially have a bit

00:25:53.059 --> 00:25:56.420
more value. And again, getting good reviews as

00:25:56.420 --> 00:25:58.940
a mechanism of improving your search rankings,

00:25:59.039 --> 00:26:01.940
not a new idea at all, but ensuring that you

00:26:01.940 --> 00:26:05.690
have google reviews and if there are particular

00:26:05.690 --> 00:26:08.569
review based sites that are particular in your

00:26:08.569 --> 00:26:11.109
industry i mean in the agency land we have clutch

00:26:11.109 --> 00:26:13.430
for example which lists agencies and there's

00:26:13.430 --> 00:26:15.529
a place where clients can provide reviews and

00:26:15.529 --> 00:26:17.829
feedback on agencies that i think is the probably

00:26:17.829 --> 00:26:20.730
the next space much harder to gain than just

00:26:20.730 --> 00:26:23.130
writing listicles though because you have to

00:26:23.130 --> 00:26:26.509
do a lot of work to go get genuine reviews right

00:26:26.509 --> 00:26:28.150
you have to engage with the customers you have

00:26:28.150 --> 00:26:30.869
to do a good job in the first place um engage

00:26:30.869 --> 00:26:33.869
with the customers heaven forbid paul yeah indeed

00:26:33.869 --> 00:26:37.009
indeed actually earned the attention crumbs who'd

00:26:37.009 --> 00:26:40.009
have thought um then um yeah you know obviously

00:26:40.009 --> 00:26:43.730
encourage them to leave reviews um and then and

00:26:43.730 --> 00:26:45.490
then we see how they get those get fed into the

00:26:45.490 --> 00:26:47.309
model without i think that's probably my guess

00:26:47.309 --> 00:26:50.529
what happens next but we shall see and in the

00:26:50.529 --> 00:26:55.130
interim while these uh best of articles are proving

00:26:55.130 --> 00:26:58.680
to be quite successful um jump on it straight

00:26:58.680 --> 00:27:01.240
away. One of my big takeaways from the Blend

00:27:01.240 --> 00:27:04.420
team, so shout out to Phil and Dan over there

00:27:04.420 --> 00:27:06.339
who did a great webinar on this the other day,

00:27:06.400 --> 00:27:10.180
was how quick this works. They showed an example

00:27:10.180 --> 00:27:12.920
where they were monitoring their visibility for

00:27:12.920 --> 00:27:15.220
a particular search and they'd ran it for a few

00:27:15.220 --> 00:27:19.299
weeks. They then published an article in this

00:27:19.299 --> 00:27:22.799
exact format and their visibility went from no

00:27:22.799 --> 00:27:26.339
visibility to visible everywhere. And it was

00:27:26.799 --> 00:27:29.380
literally within one or two weeks of publishing

00:27:29.380 --> 00:27:32.799
the speed that you can see results now they did

00:27:32.799 --> 00:27:35.759
say that wasn't the case across all of them but

00:27:35.759 --> 00:27:38.599
they were seeing pretty decent improvements in

00:27:38.599 --> 00:27:41.119
their visibility for their keywords within a

00:27:41.119 --> 00:27:44.380
few weeks rather than within three six twelve

00:27:44.380 --> 00:27:48.700
months like it can often take in the seo long

00:27:48.700 --> 00:27:51.500
game so i haven't been running this playbook

00:27:51.500 --> 00:27:56.069
for a year the logic that the tools use to output

00:27:56.069 --> 00:27:59.609
their recommendations to you does change significantly

00:27:59.609 --> 00:28:02.250
over time. We definitely found our listicles

00:28:02.250 --> 00:28:05.089
being used very quickly to create the answers.

00:28:05.349 --> 00:28:08.390
But as we got better thinking models, you know,

00:28:08.390 --> 00:28:10.029
because when we started running this, I think

00:28:10.029 --> 00:28:12.710
we were still on maybe 3 .0, 0 .3, whatever it's

00:28:12.710 --> 00:28:15.269
called. The first sort of suite of thinking models,

00:28:15.470 --> 00:28:18.029
same with like Gemini's thinking models coming

00:28:18.029 --> 00:28:21.289
out, is the rationale for drawing upon the information

00:28:21.289 --> 00:28:23.630
and making a recommendation definitely changed.

00:28:24.359 --> 00:28:26.279
I'm not even say, I wouldn't even say it would,

00:28:26.299 --> 00:28:28.599
it improved. I think it guesses a lot. Like,

00:28:28.619 --> 00:28:30.680
it's like, oh, if you're an enterprise, you should

00:28:30.680 --> 00:28:33.900
choose this agency. And I'm like, that, that's

00:28:33.900 --> 00:28:35.900
actually a good recommendation, but not for the

00:28:35.900 --> 00:28:38.839
reasons that you said. So I think it's, uh, yeah,

00:28:38.940 --> 00:28:40.859
it's quite a fun thing to, to stay on top of.

00:28:40.900 --> 00:28:44.099
And it does, um, update in our experiences also

00:28:44.099 --> 00:28:48.019
very, very quickly. The, um, another little secret.

00:28:48.380 --> 00:28:50.920
One of our competitors has been doing this before

00:28:50.920 --> 00:28:54.140
we have. They were using it to rank on Google

00:28:54.140 --> 00:28:58.279
for it was useful for AI. AI doesn't use their

00:28:58.279 --> 00:29:01.299
listicles. And they have recently written a number

00:29:01.299 --> 00:29:03.920
of listicles, either because they got wind of

00:29:03.920 --> 00:29:05.680
this strategy, maybe they saw what we were doing

00:29:05.680 --> 00:29:09.380
as a counterpunch to their punch, if you like.

00:29:09.480 --> 00:29:13.539
I never see their listicles in AI search results.

00:29:13.740 --> 00:29:16.460
And I suspect it's because they're very thin.

00:29:17.599 --> 00:29:21.539
Yeah. Well, thin content is not useful put to

00:29:21.539 --> 00:29:26.059
human or AI, I guess. I think Google is being

00:29:26.059 --> 00:29:29.140
caught out by the thin content for certain long

00:29:29.140 --> 00:29:32.640
tail keywords, honestly. But I don't think, and

00:29:32.640 --> 00:29:34.819
when I say Google, I mean traditional positions,

00:29:34.980 --> 00:29:36.660
one, two, three, four, five. I don't mean like

00:29:36.660 --> 00:29:39.259
generative overview. So it's, there's definitely

00:29:39.259 --> 00:29:42.960
a sort of a weird thinking aspect that goes on

00:29:42.960 --> 00:29:45.640
in the models. That's not just like swallowing.

00:29:46.190 --> 00:29:48.730
thin information and then repeating it verbatim.

00:29:48.730 --> 00:29:52.470
Like it's definitely trying to sustain what is

00:29:52.470 --> 00:29:56.470
reliable, but maybe it doesn't have the human

00:29:56.470 --> 00:30:00.470
cynicism to say, interesting, best of listicle.

00:30:00.990 --> 00:30:04.309
And position one is the product that this website

00:30:04.309 --> 00:30:07.730
sells. Position one has 1500 words written about

00:30:07.730 --> 00:30:09.769
it and all the other products have about 200

00:30:09.769 --> 00:30:11.390
words written about all of them. Like it doesn't

00:30:11.390 --> 00:30:13.329
have that part of it in it yet, I don't think.

00:30:13.640 --> 00:30:15.460
so yeah that's kind of interesting as well but

00:30:15.460 --> 00:30:18.299
um yeah continue to watch the space as we've

00:30:18.299 --> 00:30:21.279
discussed before what percentage of traffic what

00:30:21.279 --> 00:30:23.859
percentage of purchases are coming from this

00:30:23.859 --> 00:30:26.559
source when does it become the sort of thing

00:30:26.559 --> 00:30:28.980
that when you look at your marketing budget resources

00:30:28.980 --> 00:30:32.519
effort you deploy into it and i think again you

00:30:32.519 --> 00:30:35.200
know you have to figure out where to invest your

00:30:35.200 --> 00:30:37.339
time and your money in order to get the best

00:30:37.339 --> 00:30:39.799
commercial results and if that is making sure

00:30:39.799 --> 00:30:42.789
that you show up in the emerging field of uh

00:30:42.789 --> 00:30:45.250
of answer engine optimization or whatever right

00:30:45.250 --> 00:30:47.609
at the beginning cool that sounds like a commercially

00:30:47.609 --> 00:30:50.609
viable strategy if we're still the wild wild

00:30:50.609 --> 00:30:52.670
west and listicles work this week but they don't

00:30:52.670 --> 00:30:54.990
next and you invest in that and then and by the

00:30:54.990 --> 00:30:56.509
way you're doing this for like two percent of

00:30:56.509 --> 00:30:58.730
total traffic to your website because actually

00:30:58.730 --> 00:31:01.170
still 40 of your traffic still coming from google

00:31:01.170 --> 00:31:04.190
um and by the way disproportionately amount of

00:31:04.190 --> 00:31:07.009
customers come from your google ads then don't

00:31:07.009 --> 00:31:08.609
just jump on a bandwagon because everyone else

00:31:08.609 --> 00:31:10.559
is jumping on it Think about your own marketing

00:31:10.559 --> 00:31:12.640
mix, your own goals, what channels perform best

00:31:12.640 --> 00:31:15.779
proportionally and invest accordingly, I would

00:31:15.779 --> 00:31:18.980
say. Sound advice as ever. Shall we move on to

00:31:18.980 --> 00:31:22.940
our third story? This is this report that you

00:31:22.940 --> 00:31:26.059
brought to my attention, Martin, about OpenAI

00:31:26.059 --> 00:31:29.920
quantifying how enterprise level customers are

00:31:29.920 --> 00:31:34.319
using ChatGPT Enterprise to get commercial results.

00:31:34.839 --> 00:31:37.700
Yeah, this one caught my eye because it looks

00:31:37.700 --> 00:31:41.410
into... the who is using it and how they're using

00:31:41.410 --> 00:31:43.529
it and when i say the who i'm not talking about

00:31:43.529 --> 00:31:47.349
me because i don't have chat gpt enterprise all

00:31:47.349 --> 00:31:51.170
eyes on you for that one paul but yes this is

00:31:51.170 --> 00:31:53.789
uh really quite interesting so they're saying

00:31:53.789 --> 00:31:57.609
that the use of chat gpt enterprise has grown

00:31:57.609 --> 00:32:01.190
eight times in the last year not a great surprise

00:32:01.190 --> 00:32:03.210
that there's been more adoption in this new technology

00:32:03.799 --> 00:32:06.740
The headline bigger from the whole report, though,

00:32:06.819 --> 00:32:09.799
and this is self -reported by the respondents,

00:32:10.279 --> 00:32:12.839
but workers claim that they're saving between

00:32:12.839 --> 00:32:18.480
40 to 60 minutes every single day by integrating

00:32:18.480 --> 00:32:22.680
ChatGPT and the various tools into their workflows.

00:32:23.180 --> 00:32:25.799
Yeah, I mean, these things don't surprise me

00:32:25.799 --> 00:32:27.740
now. Maybe they should. I think on an average

00:32:27.740 --> 00:32:30.039
day, I probably save more than that. I save more

00:32:30.039 --> 00:32:31.700
time than that just dictating to my computer

00:32:31.700 --> 00:32:34.880
instead of typing. So it would be nice to know

00:32:34.880 --> 00:32:38.519
a bit more detail about what they mean by saving

00:32:38.519 --> 00:32:43.140
time. Are they sending out five more emails a

00:32:43.140 --> 00:32:46.079
day that their colleagues may or may not even

00:32:46.079 --> 00:32:49.460
be reading? I don't know. Or are they able to

00:32:49.460 --> 00:32:54.220
do meaningful work that genuinely moves the needle

00:32:54.220 --> 00:32:58.970
forward? I don't know. In my own work, yes. I

00:32:58.970 --> 00:33:01.150
would love to have just a bit more detail. So

00:33:01.150 --> 00:33:03.970
in the report, they do give a little bit more

00:33:03.970 --> 00:33:07.329
information. So we can see that it says workers

00:33:07.329 --> 00:33:10.910
report AI is contributing to value across departments.

00:33:11.089 --> 00:33:15.230
It says 87 % of IT workers report faster IT resolution.

00:33:15.430 --> 00:33:19.250
So that's a nice specific use case. 85 % of marketing

00:33:19.250 --> 00:33:22.430
and product users report faster campaign execution.

00:33:22.950 --> 00:33:27.670
75 % of HR professionals report. improved employee

00:33:27.670 --> 00:33:32.250
engagement and 73 % of engineers report faster

00:33:32.250 --> 00:33:35.349
code delivery. It's very helpful. Okay. Another

00:33:35.349 --> 00:33:38.109
question, which we cannot answer is what does

00:33:38.109 --> 00:33:40.930
faster campaign execution mean? Like that's what

00:33:40.930 --> 00:33:43.690
I want to know. Is it, they, is this basically

00:33:43.690 --> 00:33:45.849
writing? Like for us, when we do a campaign,

00:33:45.970 --> 00:33:47.890
we always write five blogs and it used to take

00:33:47.890 --> 00:33:51.049
eight weeks and now it takes two weeks and the

00:33:51.049 --> 00:33:54.369
first draft is drafted by ChatGPT or what have

00:33:54.369 --> 00:33:57.190
you. I'd love to get a bit more information on

00:33:57.190 --> 00:33:59.210
that. The other thing I'd love is HR professionals

00:33:59.210 --> 00:34:02.630
report improved employee engagement. This is

00:34:02.630 --> 00:34:05.869
actually, of all of them, could be kind of interesting.

00:34:05.970 --> 00:34:09.070
Like, why? Is it because your HR team are able

00:34:09.070 --> 00:34:12.570
to come up with initiatives or communications?

00:34:13.070 --> 00:34:16.239
Like, is the communication just stronger? Because

00:34:16.239 --> 00:34:18.900
they can put a message through ChatGPT and say,

00:34:19.119 --> 00:34:23.000
we're suffering with employee engagement. Help

00:34:23.000 --> 00:34:25.099
us build a strategy to execute on and help us

00:34:25.099 --> 00:34:26.800
execute. Like that would be really interesting

00:34:26.800 --> 00:34:29.260
as well. I'd love to know. Because whenever I've

00:34:29.260 --> 00:34:31.800
asked, I haven't run this for a while actually,

00:34:31.900 --> 00:34:34.500
but whenever I ask a model to build a content

00:34:34.500 --> 00:34:38.079
strategy for me around a particular keyword or

00:34:38.079 --> 00:34:41.340
an audience and a theme, it over eggs the pudding

00:34:41.340 --> 00:34:44.989
to such degree. that I'll be lucky if I can use

00:34:44.989 --> 00:34:48.590
20 % of the plan because it's so desperate to

00:34:48.590 --> 00:34:51.329
give so many different ideas. And I'm like, no,

00:34:51.429 --> 00:34:53.730
what's the 20 % of the effort that gets the 80

00:34:53.730 --> 00:34:57.250
% of the result? So it does occasionally give

00:34:57.250 --> 00:34:59.250
me ideas that I don't think I would have come

00:34:59.250 --> 00:35:01.869
up with myself, but it's miles away from just

00:35:01.869 --> 00:35:03.889
giving me a marketing strategy and a plan that

00:35:03.889 --> 00:35:06.630
I can execute, at least in my experience so far.

00:35:06.750 --> 00:35:08.889
So I don't know. What do you think? What does

00:35:08.889 --> 00:35:11.349
this report tell us? What you and I already know.

00:35:12.030 --> 00:35:13.889
people who are not using the tools might not

00:35:13.889 --> 00:35:16.130
though i don't know yeah i think this this report

00:35:16.130 --> 00:35:18.110
has clearly been written for a particular audience

00:35:18.110 --> 00:35:20.469
right and this is people that are on the fence

00:35:20.469 --> 00:35:24.150
about making the business case to invest in chat

00:35:24.150 --> 00:35:27.190
gpt enterprise licenses when they could be spending

00:35:27.190 --> 00:35:31.929
their money with co -pilot 365 or google gemini

00:35:31.929 --> 00:35:34.590
workspace accounts and all the rest of it um

00:35:34.590 --> 00:35:36.590
i do think there are some interesting takeaways

00:35:36.590 --> 00:35:38.889
in terms of adoption trends though so one thing

00:35:38.889 --> 00:35:43.579
that it talks about is More people are using

00:35:43.579 --> 00:35:49.579
workflows, sorry, kind of using projects and

00:35:49.579 --> 00:35:57.400
custom GPTs. That's increased 19 times year to

00:35:57.400 --> 00:36:02.780
date. So people are using it more for integrated

00:36:02.780 --> 00:36:05.579
repeatable processes is how they describe it

00:36:05.579 --> 00:36:08.579
rather than casual querying. And I recognize

00:36:08.579 --> 00:36:11.989
that for my own use case. The number of projects

00:36:11.989 --> 00:36:16.369
that I now have is just expanding weekly almost.

00:36:16.670 --> 00:36:19.489
I'm just finding new use cases to go, oh, there's

00:36:19.489 --> 00:36:21.949
my custom prompt. There's a bit of knowledge

00:36:21.949 --> 00:36:24.150
that I need that to know. I find myself doing

00:36:24.150 --> 00:36:27.150
this thing every month, every week. So I'm going

00:36:27.150 --> 00:36:29.809
to create a project for it. With enterprise and

00:36:29.809 --> 00:36:33.389
ChatGPT for business, you can share project folders.

00:36:34.050 --> 00:36:37.670
I can see that being a very useful feature for

00:36:37.670 --> 00:36:40.400
people. I agree. I use projects quite a lot.

00:36:41.179 --> 00:36:45.460
I have found that you can over -project. My first

00:36:45.460 --> 00:36:48.219
approach to projects was I'd just give it loads

00:36:48.219 --> 00:36:51.800
of information about maybe our business, our

00:36:51.800 --> 00:36:55.860
customers, our services. And I think a bit like

00:36:55.860 --> 00:36:59.480
a human, you can over -saturate them with information

00:36:59.480 --> 00:37:02.719
because they're having to read it all as the

00:37:02.719 --> 00:37:06.039
start of every chat goes into the context. pull

00:37:06.039 --> 00:37:08.059
the wrong detail out of the wrong document. So

00:37:08.059 --> 00:37:09.960
I think I made the mistake of thinking I could

00:37:09.960 --> 00:37:12.199
have one project to rule them all. And it's actually

00:37:12.199 --> 00:37:14.480
better to be a bit more niche -y, right? Like

00:37:14.480 --> 00:37:17.320
here's a project for this project, ironically,

00:37:17.480 --> 00:37:19.719
I guess the clue is in the name, but like I've

00:37:19.719 --> 00:37:22.539
got a particular campaign, let's say, rather

00:37:22.539 --> 00:37:25.099
than every business discussion with one project.

00:37:25.260 --> 00:37:28.360
But I haven't shared them much with my team.

00:37:28.460 --> 00:37:30.840
I haven't proactively gone out of my way to not

00:37:30.840 --> 00:37:33.559
share them, but I also have found myself wanting

00:37:33.559 --> 00:37:36.420
to give the projects not just information about

00:37:36.420 --> 00:37:39.900
the business but my own learnings or my own thoughts

00:37:39.900 --> 00:37:42.179
about a particular topic maybe even specifications

00:37:42.179 --> 00:37:45.860
related to my own job role which then by definition

00:37:45.860 --> 00:37:48.480
leads someone else in the business to not get

00:37:48.480 --> 00:37:50.719
the same or not necessarily get the output they

00:37:50.719 --> 00:37:53.000
want because maybe the things they're focused

00:37:53.000 --> 00:37:55.760
on are different to me and maybe my custom instructions

00:37:55.760 --> 00:37:58.260
if you will will bias the outputs a little bit

00:37:58.260 --> 00:38:00.900
so um i'm finding very useful for me haven't

00:38:00.900 --> 00:38:03.230
really shared them all uh with anyone i would

00:38:03.230 --> 00:38:05.969
love to know if a listener has had some really

00:38:05.969 --> 00:38:09.510
good examples of where they've created projects

00:38:09.510 --> 00:38:11.550
and they're sharing them and the whole team uses

00:38:11.550 --> 00:38:13.969
the project i i'm sure there are loads of great

00:38:13.969 --> 00:38:16.570
use cases and i just haven't had the time um

00:38:16.570 --> 00:38:19.070
or need really to like lean into setting one

00:38:19.070 --> 00:38:21.949
up but i'd love some inspiration from our from

00:38:21.949 --> 00:38:23.909
our passionate listeners if they wanted to send

00:38:23.909 --> 00:38:26.230
us a put something on linkedin or what have you

00:38:26.230 --> 00:38:28.139
i think what you talked there about That your

00:38:28.139 --> 00:38:31.019
use of them is quite interesting. You're giving

00:38:31.019 --> 00:38:35.900
the project detail about your role and your personal

00:38:35.900 --> 00:38:39.440
context, which would lessen the value and utility

00:38:39.440 --> 00:38:42.460
for someone else to use. Interesting you look

00:38:42.460 --> 00:38:44.940
at the prompt engineering courses that OpenAI

00:38:44.940 --> 00:38:47.659
have on their academy. And dear listeners, if

00:38:47.659 --> 00:38:50.500
you've not gone over to academy .openai .com

00:38:50.500 --> 00:38:52.739
and checked out their prompt engineering courses,

00:38:52.880 --> 00:38:55.699
I would highly recommend it. It's good for beginners

00:38:55.699 --> 00:38:59.130
and for... intermediate users as well. But in

00:38:59.130 --> 00:39:02.429
their prompt engineering 101, like the most basic

00:39:02.429 --> 00:39:04.869
introduction to it, they talk about context,

00:39:05.110 --> 00:39:08.730
role, and expectation. That's the very simple

00:39:08.730 --> 00:39:12.530
framework that they put out for users to get

00:39:12.530 --> 00:39:14.769
started with prompting. And in the role piece,

00:39:14.949 --> 00:39:16.610
what I found interesting is that everybody always

00:39:16.610 --> 00:39:19.889
talks about give the role, give the prompt a

00:39:19.889 --> 00:39:23.769
role for the AI. But in their description in

00:39:23.769 --> 00:39:26.590
the academy, they say, tell it your role. Give

00:39:26.590 --> 00:39:29.150
it a role, but tell it who you are so it has

00:39:29.150 --> 00:39:31.710
an understanding of its relationship to you and

00:39:31.710 --> 00:39:35.050
your context. It's very easy to say to the model,

00:39:35.170 --> 00:39:38.889
you're an expert marketing blah, blah, blah.

00:39:39.489 --> 00:39:42.230
But tell it who is working with and that will

00:39:42.230 --> 00:39:45.369
understand how to collaborate with you better.

00:39:45.590 --> 00:39:47.150
Yeah, I think it makes a lot of sense. I mean,

00:39:47.190 --> 00:39:50.030
we in the early days, it was very, very helpful

00:39:50.030 --> 00:39:52.750
to be quite prescriptive about what the role

00:39:52.750 --> 00:39:55.650
the model needed to fill. Interestingly, I saw

00:39:55.650 --> 00:39:57.949
a bit of research released maybe this week or

00:39:57.949 --> 00:40:00.570
last week from Ethan Mollick, and this was not

00:40:00.570 --> 00:40:03.090
just some anecdotal testing. I think they actually

00:40:03.090 --> 00:40:04.989
published a paper on this, that you basically

00:40:04.989 --> 00:40:07.489
don't really need to do that anymore. If you

00:40:07.489 --> 00:40:11.289
say you are a world -class SEO expert, that doesn't

00:40:11.289 --> 00:40:13.349
somehow make its outputs better. But in the early

00:40:13.349 --> 00:40:16.949
days, it did. You had to get it to lean into

00:40:16.949 --> 00:40:20.550
a specific statistical part of its memory, if

00:40:20.550 --> 00:40:23.510
you like. the things it knows. But yeah, this

00:40:23.510 --> 00:40:25.090
is one of the main reasons I did it. I found

00:40:25.090 --> 00:40:28.590
that I get very generic responses unless I give

00:40:28.590 --> 00:40:31.570
it the widest context I can. Some of that is

00:40:31.570 --> 00:40:33.750
about our business, some is about our customers,

00:40:33.869 --> 00:40:36.190
but some is honestly about me and the things

00:40:36.190 --> 00:40:38.889
I'm trying to solve, how I'm getting measured,

00:40:39.110 --> 00:40:42.050
especially because I often use those projects

00:40:42.050 --> 00:40:45.059
for more high -level brainstorming. And the more

00:40:45.059 --> 00:40:47.179
context you can give it about the problems you're

00:40:47.179 --> 00:40:49.019
trying to solve and the reasons that they're

00:40:49.019 --> 00:40:51.760
important now and all this other stuff, the less

00:40:51.760 --> 00:40:53.400
back and forth you've got to have. Like I could

00:40:53.400 --> 00:40:56.039
probably get there in like eight messages, but

00:40:56.039 --> 00:40:57.760
we're thinking on, you have to wait for your

00:40:57.760 --> 00:41:00.000
messages to come in. I'd like, no, that's just,

00:41:00.019 --> 00:41:02.179
let's get this done in three messages if we can.

00:41:02.360 --> 00:41:03.760
And the other thing I love about enterprise,

00:41:03.920 --> 00:41:06.179
I'm sure it's not cheap on a per user basis,

00:41:06.300 --> 00:41:09.960
but I hammer the deepest thinking capabilities.

00:41:10.239 --> 00:41:15.360
I am not shy about rolling out. um gpt 5 .1 in

00:41:15.360 --> 00:41:18.519
this case pro um because it's hard for me to

00:41:18.519 --> 00:41:20.699
max out you know how many uses have we've got

00:41:20.699 --> 00:41:22.920
of all of this stuff deep research throwing them

00:41:22.920 --> 00:41:25.519
off into ether into the ether like nobody's business

00:41:25.519 --> 00:41:27.239
in fact the only thing that really bothers me

00:41:27.239 --> 00:41:29.260
about enterprise is you do get the new products

00:41:29.260 --> 00:41:32.380
the slowest like we still have 5 .1 we don't

00:41:32.380 --> 00:41:35.179
have 5 .2 meanwhile the rest of the world you

00:41:35.179 --> 00:41:37.139
know as it turns out that might be a blessing

00:41:37.139 --> 00:41:39.519
in disguise but if that wasn't the case i'd be

00:41:39.519 --> 00:41:42.250
extremely frustrated that i everyone's like the

00:41:42.250 --> 00:41:44.550
creative writing of 5 .2 is amazing i'm like

00:41:44.550 --> 00:41:46.550
well i can't use it for at least seven to ten

00:41:46.550 --> 00:41:48.670
days after it comes out and that's always the

00:41:48.670 --> 00:41:51.070
thing that then makes me think i'll give claude

00:41:51.070 --> 00:41:53.690
opus ago do you know what i mean because uh i

00:41:53.690 --> 00:41:56.150
hate waiting but um maybe that's just me but

00:41:56.150 --> 00:41:57.690
when things are moving this quickly you want

00:41:57.690 --> 00:41:59.710
the latest tech right the minute it comes out

00:41:59.710 --> 00:42:01.730
you don't have to wait for it you do and that's

00:42:01.730 --> 00:42:04.829
exactly why i ended up subscribing to claude

00:42:04.829 --> 00:42:09.250
max so that i could get more usage of opus 4

00:42:09.250 --> 00:42:11.570
.5 when they raised the limit to the same as

00:42:11.570 --> 00:42:14.909
sonnet 4 .5 previously and i've yet to hit the

00:42:14.909 --> 00:42:17.210
limit since i've made that upgrade and yeah i

00:42:17.210 --> 00:42:20.170
would say it was plenty worth it they lovely

00:42:20.170 --> 00:42:22.769
stuff the only other thing that that i spotted

00:42:22.769 --> 00:42:25.110
this week i think is interesting is another ethan

00:42:25.110 --> 00:42:26.949
mollock post martin that's just made me think

00:42:26.949 --> 00:42:31.190
of where he was talking about how he expected

00:42:31.190 --> 00:42:35.519
he's very big on business use cases for ai I

00:42:35.519 --> 00:42:38.519
think he's a business professor over at the Wharton

00:42:38.519 --> 00:42:42.079
Business School or something. And so he's always

00:42:42.079 --> 00:42:43.639
looking at business use cases. And the one that

00:42:43.639 --> 00:42:46.599
he's felt has taken a while to emerge is really

00:42:46.599 --> 00:42:49.460
the ability to create great reports and presentations,

00:42:49.780 --> 00:42:51.880
right? PowerPoint presentations. And he said

00:42:51.880 --> 00:42:53.500
something interesting this week, Martin, about

00:42:53.500 --> 00:42:56.800
expecting... the models, the text -based models

00:42:56.800 --> 00:42:59.260
to eventually solve this by coming up with what

00:42:59.260 --> 00:43:01.099
the presentation should have, but basically then

00:43:01.099 --> 00:43:04.380
coding it in XML or whatever it is, is the underlying

00:43:04.380 --> 00:43:07.599
code. But being surprised that Nano Banana Pro

00:43:07.599 --> 00:43:10.780
is so good at infographics that it can actually

00:43:10.780 --> 00:43:13.780
also be very good at slides. And if you go into

00:43:13.780 --> 00:43:16.280
a tool like Notebook LM, which is a Google product,

00:43:16.400 --> 00:43:18.380
if you've got Google Workspace, you have access

00:43:18.380 --> 00:43:20.400
to this. If you have a free Google account, I

00:43:20.400 --> 00:43:22.079
think you've even got access to it with some

00:43:22.079 --> 00:43:25.059
limits. You can chuck a report in. or a brief,

00:43:25.219 --> 00:43:27.039
and it will produce a slide deck for you. And

00:43:27.039 --> 00:43:30.039
it's using the latest nano banana model. And

00:43:30.039 --> 00:43:33.179
apparently these slide decks are really much,

00:43:33.280 --> 00:43:36.400
much better. You can't edit them yet, but I think

00:43:36.400 --> 00:43:38.400
Ethan said in his post that a little birdie told

00:43:38.400 --> 00:43:40.099
him that that's something they're working on

00:43:40.099 --> 00:43:42.260
that will make it much easier to manipulate those

00:43:42.260 --> 00:43:45.179
presentations. So I'm quite excited about that

00:43:45.179 --> 00:43:46.699
because I agree. I do feel like that's the sort

00:43:46.699 --> 00:43:48.900
of thing that should have been better by now,

00:43:48.960 --> 00:43:51.199
but it hasn't been. Yeah, creating slides has

00:43:51.199 --> 00:43:54.630
proven to be tricky we've we've known that chat

00:43:54.630 --> 00:43:58.170
gpt can do it for ages by using a python script

00:43:58.170 --> 00:44:01.650
to create a basic version but they've always

00:44:01.650 --> 00:44:04.829
been pretty bad unless you just wanted a simple

00:44:04.829 --> 00:44:07.909
layout where it adds a little bit of text on

00:44:07.909 --> 00:44:10.889
a slide and maybe some speaker notes if you are

00:44:10.889 --> 00:44:13.630
really pushing the boundaries of the the model

00:44:13.630 --> 00:44:18.010
whereas recently we've seen it embedded in copilot

00:44:18.010 --> 00:44:20.329
pro so if you've got a copilot pro subscription

00:44:20.329 --> 00:44:23.300
you can actually prompt it to create a slide

00:44:23.300 --> 00:44:26.059
deck and that does a pretty good job i would

00:44:26.059 --> 00:44:28.679
i would say i enjoy using that and i've had good

00:44:28.679 --> 00:44:32.320
success using that i've been using manus recently

00:44:32.320 --> 00:44:36.139
and i really like what manus is doing it's not

00:44:36.139 --> 00:44:39.679
without its challenges their slide builder interface

00:44:39.679 --> 00:44:42.840
you can't drag and drop and reorder the slide

00:44:42.840 --> 00:44:45.940
so you have to prompt it to do that and yes that's

00:44:45.940 --> 00:44:49.019
that's not always the the best experience and

00:44:49.019 --> 00:44:51.670
then weirdly i had an experience with Gemini

00:44:51.670 --> 00:44:54.030
3 the other day, where even though I hadn't asked

00:44:54.030 --> 00:44:57.230
it to do this, it created a slide deck for me

00:44:57.230 --> 00:45:00.070
as part of an ongoing discussion. I was like,

00:45:00.110 --> 00:45:03.789
oh, great. Cheers. Can you make some edits to

00:45:03.789 --> 00:45:06.210
that deck? Instead of editing the deck that it

00:45:06.210 --> 00:45:10.010
had produced, it got rid of that entirely, rewrote

00:45:10.010 --> 00:45:13.530
a new one, but using a HTML file. And then when

00:45:13.530 --> 00:45:15.909
I said, oh no, I want it in Google Slides like

00:45:15.909 --> 00:45:18.420
you did the first time, it's... started to gaslight

00:45:18.420 --> 00:45:20.420
me and told me that it didn't have that capability

00:45:20.420 --> 00:45:25.099
of creating slides in Google Slides. And I haven't

00:45:25.099 --> 00:45:28.199
been able to recreate it since. So yeah, overall,

00:45:28.340 --> 00:45:30.019
I would say the experience of trying to use AI

00:45:30.019 --> 00:45:33.179
to create presentations is inconsistent and frustrating.

00:45:33.500 --> 00:45:34.840
Yeah, it's funny you should say that. Do you

00:45:34.840 --> 00:45:36.679
remember that workshop that we went to where,

00:45:36.780 --> 00:45:39.219
as part of what we were showing, because most

00:45:39.219 --> 00:45:40.639
people at that point didn't even know you could

00:45:40.639 --> 00:45:43.280
get a PowerPoint file out of ChatGPT. We're talking

00:45:43.280 --> 00:45:45.639
maybe even two years ago now, right? And we tried

00:45:45.639 --> 00:45:48.699
to do it in the live demo. And then the chat

00:45:48.699 --> 00:45:51.039
chief was like, no, I can't. And you were like,

00:45:51.719 --> 00:45:53.579
no, you have. You've done it for me in the past.

00:45:53.699 --> 00:45:55.280
It was like, no, I can't. You were like, yeah,

00:45:55.360 --> 00:45:57.219
do you remember this time you did this and this?

00:45:57.340 --> 00:45:59.519
And then you gave me the PowerPoint file. Like,

00:45:59.519 --> 00:46:01.400
I really insist that you give me my PowerPoint

00:46:01.400 --> 00:46:03.519
file. And they went, here's your PowerPoint file.

00:46:04.019 --> 00:46:06.500
So we're still in the age of having to twist

00:46:06.500 --> 00:46:09.179
the arm of a tool to remember its capabilities,

00:46:09.559 --> 00:46:13.780
unfortunately. Yeah, it is frustrating, but for

00:46:13.780 --> 00:46:18.019
sure. One other story from me this week, this

00:46:18.019 --> 00:46:21.199
is a personal story that I don't think I'm going

00:46:21.199 --> 00:46:23.420
to get a great deal of sympathy from people,

00:46:23.519 --> 00:46:27.539
but this little device in my hand, listeners

00:46:27.539 --> 00:46:32.880
may remember, some time ago I ordered the Limitless

00:46:32.880 --> 00:46:36.380
Pendant. So Limitless were a company that were

00:46:36.380 --> 00:46:40.019
creating AI tools and they had a meeting note

00:46:40.019 --> 00:46:41.679
assistant and then they said they were going

00:46:41.679 --> 00:46:44.219
to move into hardware. And you could pre -order

00:46:44.219 --> 00:46:46.679
this and they were going to retail it for $100.

00:46:46.760 --> 00:46:49.059
But if you pre -ordered, it was a bit like Kickstarter.

00:46:49.800 --> 00:46:54.059
Paid $50. They'd ship it out to you. And I paid

00:46:54.059 --> 00:46:56.599
that money about 18 months ago. It feels like

00:46:56.599 --> 00:46:59.699
now it was some time ago. This was finally delivered

00:46:59.699 --> 00:47:01.840
to me. This little pendant. Great little gadget.

00:47:02.079 --> 00:47:05.920
It was delivered in September. I only used it

00:47:05.920 --> 00:47:08.599
a few times. A bit of a tricky device. It's always

00:47:08.599 --> 00:47:12.369
on. listening device which you know people rightly

00:47:12.369 --> 00:47:14.869
have some privacy concerns when you when you

00:47:14.869 --> 00:47:16.769
say oh do you mind if i just record everything

00:47:16.769 --> 00:47:19.369
you say to me um and have that on a permanent

00:47:19.369 --> 00:47:22.269
record forever is that a problem some people

00:47:22.269 --> 00:47:25.510
did have a problem with that and yeah anyway

00:47:25.510 --> 00:47:30.269
long story short meta has acquired the limitless

00:47:30.269 --> 00:47:35.449
company and in doing so they've also ceased all

00:47:35.449 --> 00:47:39.190
business operations in europe so they they shut

00:47:39.190 --> 00:47:41.650
down the service in the eu and it seems like

00:47:41.650 --> 00:47:45.170
they've rolled up the uk into that i'm presuming

00:47:45.170 --> 00:47:48.190
that's got something to do with gdpr and our

00:47:48.190 --> 00:47:51.630
implementation so yeah i waited about 18 months

00:47:51.630 --> 00:47:54.449
to get my hands on the limitless pendant got

00:47:54.449 --> 00:47:57.510
it used it a few times much to people's disgust

00:47:57.510 --> 00:48:01.389
and now i've got a bricked device what a 50 investment

00:48:01.389 --> 00:48:03.929
that was so you've got to speculate to accumulate

00:48:03.929 --> 00:48:06.250
mine so i think it was a good try but um The

00:48:06.250 --> 00:48:09.389
irony being your initial exposure to this has

00:48:09.389 --> 00:48:11.949
been people are maybe not super keen on getting

00:48:11.949 --> 00:48:14.090
recorded. And then you're like, oh, no, don't

00:48:14.090 --> 00:48:17.050
worry. The company's owned by Meta now. So you

00:48:17.050 --> 00:48:21.769
can absolutely rely on the data not being misused.

00:48:21.949 --> 00:48:24.030
It's just going to add more fuel to the fire

00:48:24.030 --> 00:48:27.070
of people seeing ads on Facebook or Instagram.

00:48:27.309 --> 00:48:29.349
Like, my phone listens to me. My phone listens

00:48:29.349 --> 00:48:31.610
to me. No, it's not your phone. It's Martin's

00:48:31.610 --> 00:48:37.019
pendant. It's Martin. Yeah, I mean, yeah. When

00:48:37.019 --> 00:48:39.260
you want to be on the cutting edge of hardware

00:48:39.260 --> 00:48:42.400
acceleration and stuff, I guess this is going

00:48:42.400 --> 00:48:45.420
to happen. But pretty sucky to wait 18 months,

00:48:45.579 --> 00:48:47.659
get basically two months of use, and then your

00:48:47.659 --> 00:48:51.039
device gets bricked for you. Any sign of any

00:48:51.039 --> 00:48:56.059
compensation or get your money back? No, I emailed

00:48:56.059 --> 00:48:58.800
them, but I dare say that's just fallen into

00:48:58.800 --> 00:49:04.300
a big... bin called do not respond meta hq yeah

00:49:04.300 --> 00:49:06.539
they'll just look at your ip address and go don't

00:49:06.539 --> 00:49:09.500
bother with that area of the world anymore one

00:49:09.500 --> 00:49:12.260
last thing for me um if you can find this online

00:49:12.260 --> 00:49:16.340
it's pretty cool there is a new like robot from

00:49:16.340 --> 00:49:19.539
one of the founders of boston dynamics um mark

00:49:19.539 --> 00:49:21.440
raybert his name is he works a different company

00:49:21.440 --> 00:49:24.619
now but they've been using reinforcement learning

00:49:24.619 --> 00:49:28.260
to have this it's like a two -wheel bike with

00:49:28.969 --> 00:49:32.269
the equivalent of like a human body like counterweight

00:49:32.269 --> 00:49:33.829
on the top of it so it looks kind of bizarre

00:49:33.829 --> 00:49:37.809
but it's been teaching itself to do like crazy

00:49:37.809 --> 00:49:41.730
bike stunts and parkour and stuff in a simulated

00:49:41.730 --> 00:49:44.670
environment so what it did is it's been in this

00:49:44.670 --> 00:49:46.969
computer simulated environment figuring out what

00:49:46.969 --> 00:49:48.909
it would need to do to be able to do these tricks

00:49:48.909 --> 00:49:52.090
and not lose balance and then zero shot the actual

00:49:52.090 --> 00:49:54.530
robot in the real world could do all those tricks

00:49:54.530 --> 00:50:00.960
without There's a podcast that I listen to called

00:50:00.960 --> 00:50:02.980
Moonshots. The main reason I like it is just

00:50:02.980 --> 00:50:05.780
because it's just so sort of almost ridiculously

00:50:05.780 --> 00:50:08.280
optimistic and positive. Sometimes in a world

00:50:08.280 --> 00:50:11.320
of most news networks not being that, it's quite

00:50:11.320 --> 00:50:12.880
nice to have something like that in your life.

00:50:13.039 --> 00:50:16.380
But one of the folks on that podcast was at the

00:50:16.380 --> 00:50:19.260
big AI conference last week, the week before.

00:50:19.679 --> 00:50:22.840
And among AI and machine learning researchers,

00:50:23.440 --> 00:50:26.460
there is now a growing excitement that after

00:50:26.460 --> 00:50:30.269
agents, robots are are maturing at a very quick

00:50:30.269 --> 00:50:33.130
rate and this feels like it's already getting

00:50:33.130 --> 00:50:37.670
some of that early boom bust discussion like

00:50:37.670 --> 00:50:39.369
there's quite a lot of money going into robotics

00:50:39.369 --> 00:50:43.849
obviously um elon musk is talking about it a

00:50:43.849 --> 00:50:46.130
lot and whenever he makes predictions about how

00:50:46.130 --> 00:50:48.070
quickly a product can be in your hands you just

00:50:48.070 --> 00:50:50.789
need to times it by 58 and you've probably got

00:50:50.789 --> 00:50:53.639
the correct timeline But there seems to be really

00:50:53.639 --> 00:50:55.420
some cool stuff going on there. So if you are

00:50:55.420 --> 00:50:58.079
interested in AI and robotics, I would suggest

00:50:58.079 --> 00:50:59.800
you go find that video because it is kind of

00:50:59.800 --> 00:51:02.159
cool. And if it's true that videos are, sorry,

00:51:02.320 --> 00:51:05.699
robots are training themselves in simulated environments

00:51:05.699 --> 00:51:09.179
with this type of delivery, like this is not

00:51:09.179 --> 00:51:10.980
like I'm training myself to make a coffee, which

00:51:10.980 --> 00:51:12.840
would be impressive enough in a simulated environment.

00:51:13.059 --> 00:51:16.059
This is doing parkour on two wheels, bouncing

00:51:16.059 --> 00:51:18.579
your bike all over the place. Like it's pretty

00:51:18.579 --> 00:51:21.530
cool video. Go find it. It's interesting. Right.

00:51:21.630 --> 00:51:23.710
I think we'll wrap up this week there then, Martin.

00:51:23.750 --> 00:51:25.590
Thanks very much as always for joining me and

00:51:25.590 --> 00:51:27.750
I'll look forward to you on our next one. Meet

00:51:27.750 --> 00:51:31.429
you next time. Cheers. Bye. Thank you for listening

00:51:31.429 --> 00:51:34.550
to Artificially Intelligent Marketing. To stay

00:51:34.550 --> 00:51:37.929
on top of the latest trends, tips and tools in

00:51:37.929 --> 00:51:41.130
the world of marketing AI, be sure to subscribe.

00:51:41.630 --> 00:51:44.230
We look forward to seeing you again next week.