November 7, 2023 • 41 mins
Welcome to FranPro Resource Podcast. If you would like to access our most recent content and to receive updates, you can register here: https://franpro.com/
Contact us here Anything@FranPro.com if you:
- Want help finding a franchise
- Would like to be featured on our program
- Would like help producing or want a podcast produced for you
- Are a franchise company and want access to our free ROI Tracker dashboard
In this episode, Lance Hood of FranPro interviews Dominik Lambersy, co-founder of TextCortex. Dominik Lambersy is an incredible resource for any franchise organization. If you would like to work with Text.Cortex go here: https://Franpro.vip/GoTextCortex
Covered in this call:
- Text.Cortex has ChatGPT along with its own AI model called Zeno Chat
- Zeno Chat’s browser extension is a companion on over 4,000 platforms
- Your AI twin can be trained to write like you or in company voice
- AI’s output is dependent on your ability to prompt it effectively
- AI is unlimited creativity & needs rules for large complex tasks
- And more
Imagine harnessing the power of AI to fuel your creativity and boost productivity. That's exactly what we tackle in our conversation with Dominik Lambersy, the co-founder of TextCortex. Dominic divulges the genius behind how TextCortex is reshaping AI experiences by offering personalized intelligence that yield tailored results. He talks us through how users can customize their AI assistant, input their own writing examples, and even link their own documents. Dominik takes things up a notch by providing a live demonstration of the TextCortex Chrome extension and its capabilities.
Have you ever pondered about AI's role in creative processes? Generative AI changes the game by automating these processes, and we dive into how this revolution is happening. The importance of communication as a soft skill is discussed, guiding AI to give the right instructions and parameters. We also underline how vital it is to verify AI-generated content and break down larger tasks into manageable ones. Dominik gives us a peek into the world of AI, explaining the current model that uses a window of 100,000 tokens and the ongoing research into AI's short-term memory retention.
Finally, we explore the future of AI - a future where AI experiences are customizable. Dominik shares his insights on how to make these experiences more personal, from using personas to understanding the training and knowledge bases. We stress the importance of educating the new generation to work with AI, not just to search but to create. Join us on this exhilarating journey into the world of AI, its potential to automate creative processes, and how TextCortex is playing a pivotal role in this revolution.
Contact us at Anything@FranPro.com if you:
- Want help finding the right franchise for you
- Would like to be featured on our program
- Would like help to produce or want a podcast produced for you
- Are a franchise company and want Free access to our ROI Tracker dashboard
*Some of the companies we interview compensate us a commission if you purchase something.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Lance Hood (FranPro) (00:09):
Hi everyone.
Today I'd like you to meetDominic Lamberzi, the co-founder
of Text Cortex.
Hi Dominik.
Dominik Lambersy (TextCorte (00:17):
Hey Lance, Thanks for having me.
It's a pleasure to be here.
Lance Hood (FranPro) (00:20):
Well you, I'm so glad that we're on this
call today and enjoying this andhelping out the franchise
companies.
Can you explain what does?
Because it does AI writing, butyou guys also are very
different, I think.
Dominik Lambersy (TextCort (00:37):
Yeah , thanks for the brief
introduction.
Actually, up until the end oflast year, we were known as a
writing assistant, ai writingassistant a little bit like
Grammarly, but much more capablein doing things.
Ever since we released ourversion of chat GBT we call it
Xenochat People have startedseeing us as a virtual assistant
(00:57):
to their work.
Yeah, what I am currentlybuilding on our narrative around
is that we want to buildpersonalized intelligences.
Why personalized?
Because a lot of solutions youcurrently see around they give
you pretty generic outputs andresults, and with , you don't.
You do not only have acompanion with is with you on
over 4,000 websites andplatforms, but it's also
(01:20):
customizable on how you act asan individual or organization,
plus also knowing what you knowas an individual or organization
, and therefore, yeah, ending upin a much more personalized
experience for you.
Think about your own little AIcoworker.
Lance Hood (FranPro) (01:39):
Right?
Well, it sounds like you canload up a bunch of stuff that
your company has written, sothat then, when it does create
copies so let's just say blogposts, emails, responses,
whatever it's going to write itthrough the lens of all the
stuff you've written before, sothat it writes in your unique
(01:59):
voice every time, not a generaloutput, correct?
Dominik Lambersy (TextCort (02:04):
Yeah , exactly, we did this whole
feature built around two pillars.
Yeah, first of all,customization how you act, what
you give is basically sort of abackground you as a person,
where you're coming from, yeah,what is your history and, for
example, the values you areacting after and what you then
can do is also adding a fewwriting examples, such that our
(02:25):
algorithm can map your digitaltwins, so to say, on over 60
variables.
This is one pillar.
The second pillar are knowledgeconnectors yeah, where you, for
example, would can connect,upload any type of PDF files,
connect your notion, connectyour Google Drive.
The knowledge connectors arestill in beta for the for the
time being, but will be alsoreleased very soon.
(02:46):
And then it's just, like yousaid, basically building up on
what you have learned in thepast as an individual or
organization.
Again, because, if you thinkabout the main disadvantage of
many of these, many of theselarge language models, whether
it be chat, gpt or any other AItools, is that they are stuck in
the past.
They are mostly havinginformation from 2021, from the
(03:11):
European continent.
Yeah, if you would ask for afamous soccer, famous, famous
sports event, like the soccerWorld Cup here, who was the
winner of the 2022 soccer WorldCup.
Most AI tools cannot answer it,so what we have done with Zeno
is basically integrating a lotof different knowledge fields,
also including search, including, in the future, own
(03:32):
personalized files, whatever youwant to share, certainly, we
try our best to make it assecure as possible as well.
Lance Hood (FranPro) (03:42):
And so, as we go through the functionality
, we've gone through thepersonal AI assistant that does
writing.
We've talked about how Zenochat I mean there is access to
chat GPT, but we have the Zenochat, which is its own model and
then what are the otherfunctionalities that people
should be aware of?
Dominik Lambersy (TextCorte (04:02):
The other functionalities we have.
We have over 200 use cases bynow.
We have plenty of differentusers from from plenty of
different fields.
Maybe let's roll up thatquestion a little bit
differently and talk about whoare our top users.
The top users are people fromthe creative space, oftentimes
marketers or copywriters.
That's by far the biggest group.
(04:24):
Then we have yeah, we call themcorporate professions.
Oftentimes those are salesfunctions, those are customer
support functions, but alsomanagement functions, because a
lot of people deal withcommunication and prove their
communication of our tools.
And the third one is SMEs, andthat's a combination of what
(04:48):
I've what I've previously said.
Yeah, we help them withmarketing, with sales, general
communication, a lot.
And then we also have a lot ofstudents, which I don't believe
is so important right now forour call here, but yeah, they
are getting some help with theirhomework, more or less, so also
improve their communication,their writing.
Lance Hood (FranPro) (05:04):
Well, did you want to do a brief demo
today, dominic, or?
Dominik Lambersy (TextCortex (05:10):
Oh , I mean absolutely.
I can show people brieflyaround.
Let me just quickly share myscreen and find where I already
I think it's this one.
Let me check.
Yeah, I think that's the right.
Can you see my screen now?
Lance Hood (FranPro) (05:28):
I can yeah .
Dominik Lambersy (Tex (05:29):
Beautiful .
So first of all, yeah, what Iwill, what I will show you today
, is the workings of the Chromeextension.
I can see it here in the topright.
The Chrome extension is active.
You can change it.
I mean, I haven't changed inEnglish.
You can change it in differentlanguages.
That's one of the majorproblems we had recently,
because we have global impact onmany different language
speakers and they want to see itin German, spanish, french.
(05:52):
But let's get down to the usecases here.
For example this is a callsummary when I talked with you,
for example in the past, or withany type of potential customer,
I always note down things inthe most uncanny ways, with
really weird abbreviations, forexample, just like quickly
(06:12):
shutting it down and trying tojust retain knowledge.
What I can do now sorry, that'snow being German basically says
press a shortcut for Zeno.
Why is it German nowInteresting.
Lance Hood (FranPro) (06:32):
Yeah, I like it.
I like it there, I've used itand what I like is, as long as
you see that pink and bluecircular icon there, you know
that you can just click on thatand access it right away.
Dominik Lambersy (TextCorte (06:46):
Yes , exactly, I want to show you
now first the Zeno assistfeature, for example, and what I
can do now is I can say, hey,please summarize me the above.
It works.
It says the demo call was withBen Tobito's restaurant chain,
owner of TechSavvy.
It makes the whole thing alittle bit more understandable.
(07:06):
This is one of the one things.
The same thing you can do withour toolbar, for example.
It's very easily happening justby highlighting anything and
then clicking rewrite, expandingcreative writing, email writing
as well.
So this toolbar is usually ourmost popular features fixed
(07:31):
spelling and grammar completion,for example.
I could also, if I now saygrammar fix, it should fix that
really weird.
Yeah, look, with the code andmy weird abbreviation here we
could bring revenue potential 5X, but it didn't really summarize
it.
For summarization, we wouldclick this year and we would
arrive at a very similar example, like we had with Zeno assist
(07:53):
here.
What you mentioned with thelittle bubble, that's our access
to wait.
I need to move you a little bitto the over here.
This is where our Xenoy chatcomes into effect, yeah, where
we can say, hey, write me theoutline of blog post on
(08:14):
increasing franchiseopportunities via follow up
emails.
Let's see what it will writeusually takes around 10 seconds
up here In no time.
(08:35):
We already have our outline.
Then we can start as a humanbeing.
Then we can go back into thedivide and conquer approach
(08:57):
where you can now say, hey,let's expand on each chapter and
build it step by, step by step.
The whole thing works on yeah, Imentioned someone before 4,000
different platforms.
We have here this content brief, which is part of one of our
YouTube sessions, for example,as well, where you can hear now
also say, hey, we have the localSEO topic here, please expand
(09:23):
on that Can make it a little bitlonger, takes a little while,
and it also starts writing here.
You see again, it startedcreating an outline, basically
filling that with a little bitmore of life.
I click on the format thing isa little bit broken now and then
(09:45):
you can already go step by stepdown the whole the whole road,
also working on Gmail, forexample.
Here's again the bento Beatlesexample.
In Gmail, I can just say, hey,I take these bullet points as I
already have made up my thoughts.
Just highlight it, press on theemail button and it basically
(10:05):
formulates out an email for you.
Maybe we need to have it alittle bit longer.
That's also when I still do ourcustomer support.
A lot of times I just copypaste the original email from
our customers into one of ourtemplates, just drop it in here.
I give some directions, as Ihad it here with the bento
(10:31):
example, because you always have.
If you read an email, youalways have these three, four,
five next steps you want to doup on top of your head.
You can write them down here.
Basically let the AI create.
I won't create it now because Idon't have an original email
here, but then I already have myemail formulated out.
(10:52):
That comes at much higherresponse time, certainly, but
also at less psychologicalburden.
I mean, this is what usuallycustomer support users use text
cortex for exactly this view,and I know it also.
Sometimes you have customerswhich are plainly not nice,
which is a psychological strainfor many users, customers as
(11:15):
well.
There's so much about the textcortex plug-in.
I don't want to go deep, I justwant to show it briefly.
In the customization suite youneed to go to our web app for
that and then you have theKnowledge Connectors Customize
your AI and this is basicallywhere you can start Persona name
, persona block background.
(11:37):
I think we talked two or threetimes in here.
We just type down here's theexample from my co-founder,
jehun some values, a description, and then you would enter
different types of writingexamples.
I can show you with my littledigital twin how I did it.
This is then the result.
After analysis, it tells methat apparently I'm a very
(11:58):
enthusiastic, motivational,inspirational person.
I think I should be, as afounder, that reading is.
I really try to sound as easyas possible such that everybody
can understand that sentencelength, medium, active voice.
You can see the background ofmyself.
I hear values, I like to workafter examples, and then you can
(12:24):
basically use this Personaright within the note chat.
Lance Hood (FranPro) (12:29):
So you've taken the time to fill out all
that information and set it up,and now you're writing in the
Persona of this AI Persona,which is look, that's my comment
(12:52):
pitch Born and raised on Germanfun fair selling sausages and
fries.
Dominik Lambersy (Text (12:57):
Customer service.
Yeah, that's my first pocketmoney in gaming in the early
2010s, before NFTs were a thing,and suddenly I have my virtual
twin, who replies a lot like meand can give advice not fully
like me because it doesn't haveall the knowledge I have, but
(13:17):
that will soon come here as wellwith our Knowledge Connectors
where you can integrate allsorts of different things and,
before overloading your viewersmore, if somebody's interested,
we have a lot of blog postsaround how you can use all
features videos around them aswell to best educate you.
Lance Hood (FranPro) (13:41):
Yeah.
So I think that this custom AIis really a great way to get
around the general genericresponses people would get.
But also, not only are theygeneric, but they're not going
to really follow along with yourcore values or anything like
that and your more humanisticpersonality driven writing
styles.
This is kind of the shortcut orhack, for that is to set up
(14:04):
your persona.
They would also let people knowthat the people who are writing
stuff for the company, or ifyou're a marketer that's doing
work for a company, you canupload stuff on that specific
company in here and then, oh,today I'm writing for franchise
A, so I use this persona to that.
You know, now I'm writing forfranchise B, I'll use this
(14:26):
persona.
Dominik Lambersy (Te (14:27):
Absolutely , absolutely.
Lance Hood (FranPro) (14:29):
Yeah, so it helps you produce stuff much
better.
So Well, this is excellent.
You know, one of the thingsthat I like about the usability
of this is that you have thebrowser extension, so wherever
you're working, the XenoChat canstart writing things for you.
You know, it's very handy.
Dominik Lambersy (TextCort (14:50):
Yeah , it's very handy for 100,000 of
people who use it actively on4,000 platforms now, and I can
see our competitors and they cansee how people use it in Google
Docs.
They use it on Outlook, gmail,even chat, gpt, for example.
So when our belief and ourvision was back in the days,
(15:10):
nobody wants another web app totap in between all the time.
We just want to be next to ouruser, right where they need us.
Lance Hood (FranPro) (15:21):
Right.
Well, you know as I think aboutusing and implementing this
stuff, because I've always lovedcopywriting, done copywriting
and this tool when I first foundit was just absolutely amazing
because it overcomes the whole.
I don't know where to start.
If you just give me a whitesheet of paper and say write
this, it's like there's too muchcreativity there.
(15:43):
I don't know where to startwithout having you know some
sort of preexisting tests ortemplates or baseline of some
kind.
I can improve very easily, butimproving on nothing is hard.
So this is a great way to getstarted, but you know to talk
about how to talk to it andimprove it.
I think, first of all, it'sgood to understand what is AI,
(16:05):
because I think people have somemisconceptions or
non-understandings on what it is, and you have to know what it
is and how it works so that youcan talk to it the best and get
the best results.
So what is AI?
Is it this big computer systemthat's going to take over the
world?
Dominik Lambersy (TextCort (16:21):
It's a great question.
I've also, last week, have beenpart of a finance summit as a
keynote speaker for GenerativeAI and there was similar
misconceptions.
I mean, ai is nothing so new.
Generative AI is now new.
What is Generative AI basicallydoing?
It tries to create something.
I like to call it also creativeAI because from what it has
(16:43):
learned from all the tens ofthousands of texts or, in terms
of mid-journey tens of millionsof pictures it has seen, it can
draw and learn patterns which weusually combine with some sort
of textual structure and canrecreate on top of it.
What happens under the hood ismore or less an auto completion.
(17:04):
So if you would ask an AI, hey,please write me the blog
outline for a blog around thetopic of how I can improve my
marketing ops for my franchise.
The AI basically takes this asinput and improves.
Or hello, Siri, I'm talkingabout AIs why?
(17:24):
Why does it come up now?
It tries to auto complete yourprompt.
That's why it's also soimportant that that's what I've
seen with many users.
One or two years ago from ourproduct, they talked to Zeno, to
text cortex, like it was amachine.
But ultimately, what you need toreally think about is how you
(17:47):
communicate and I believe thiswhole thing, the soft skill of
communication, will be much moreimportant now in the upcoming
future in the collaboration withAI systems.
Because, in comparison to if wetalk with our team, for example,
and I tell one of my teammembers, hey, it would be great
if I have a result to topic XYZin one to two days, I get a
(18:09):
response in one to two days withan AI, have instant feedback.
So if I ask it to write anoutline for a blog post, I will
have a result within 10 seconds.
If I don't like the output,maybe it was more or less about
me, because write a blog outlinefor a topic X is not very
defined.
So, for example, expand on theprompt.
(18:30):
Hey, the topic is generative AIin the franchise ecosystem.
I want to talk about themarketing operations and how to
decrease costs, for example, orhow can I scale up for more
opportunities if I add somesources, some real time data,
(18:51):
that really helps the AI tounderstand where you want to
arrive at.
And these basic instructionsare incredibly important and
that's why I say communicationas a soft skill will be much
more important in ourcollaborative ways around AI in
the future.
Lance Hood (FranPro) (19:10):
So you need to talk to it like you talk
to people and just like you'rehaving a conversation with
another person.
That's how you talk to AI.
But also one thing I've learnedfrom talking with you in the
past is that AI is creative.
It's made to be creative and soit can go off all over the
place because it's just designedto be creative.
(19:32):
So the more you give it moreinformation and you type in more
stuff or to give it direction,it's going to be more of what
you want.
So if you just say, write thisand you get a general response,
that's because what you'resaying is too general and so
it's just going okay and itanswers it.
But if the more precise you get, you'll get it.
(19:52):
And that's why you also saidthat if AI is doing, if you were
using AI to do some sort ofresearch, you'd want to just
double check it, because it'svery creative and it can try to
autocomplete and not necessarilyhave autocomplete.
Be 100% accurate with facts.
Dominik Lambersy (TextC (20:12):
Correct .
Yeah, exactly, that's a reallygood point from our former
conversations Always validatewhat the AI is creating for you.
I mean, six months ago,everybody believed what AI has
written, which is extremelydangerous.
It's just like how you interactwith other human beings.
Take everything with a grain ofsalt and do your own critical
(20:32):
thinking around it.
There are plenty of people whotry to sell you something and
lie to you.
Similar things can also happenwith any type of AI generation.
So always double check facts.
There was another point which Iwanted to add to that, but
sincerely it flew out of my mindnow.
Maybe it comes in a little bitlater down the road, but
(20:55):
validation of information isextremely important by an AI
model and it's kind of thereason why AI's generally have a
limited output, because you'll,you'll, uh, you can.
Lance Hood (FranPro) (21:08):
it can go out too far off in the wrong
direction if you don't Overseeit, and I've always kind of
looked at.
After I started using it, IStarted looking at AI as more of
like a race car, but it needs areally good human driver, you
know like it's a little course.
What's that?
Dominik Lambersy (TextCortex) (21:26):
a really good pilot to it, huh.
Lance Hood (FranPro) (21:28):
Yeah, yeah , exactly.
Somebody's got to be in controlof it, Otherwise it's not gonna
have as much value.
So if you're not getting thevalue out of your AI and it's
not producing what you want, youjust have to learn how to talk
to it and you have to learn howto drive it and direct it, and
then you have to oversee theoutput.
So it's not necessarilyEliminating a lot of humans.
(21:48):
What it's doing is, is it justit produces way more, so you can
get more, way more output perperson.
Dominik Lambersy (TextCort (21:55):
Yeah , that's what I also said in
that in that speech recentlylast week.
We basically go now from an agefrom Unlimited access to
information to an unlimited, tothe age of unlimited, unlimited
creation.
Basically, what happens is thatwe do not need to think about
all these information whichcomes to us at the moment and
(22:16):
then need to produce something.
The production is alreadyHappening, also on the AI level,
but still it's our job tovalidate whether it's right and
proper, fitting Our use case andgenerally true, and I also just
got my point break which I justlost a minute ago.
One concept which we're alsopraising to our users is called
divide and conquer.
It's nothing new.
Yeah, with many big problemsyou have in life, you should
(22:39):
divide and conquer them intosmaller ones, and we have this
beautiful YouTube video as wellwhen we basically show people
through how they can collaboratewith AI to create blog posts.
Now, if your goal is a 2000word blog post, your
expectations shouldn't be thatthe AI is Just snapping the
finger and you have a 2000 wordblog post in front of you, which
is perfectly fitting your case.
(22:59):
That's the.
The short-term memory of an AIis not enough to keep that
concise and to keep, keep andkeep the truthfulness behind it.
So what you do?
You divide and conquer theproblem.
For example, instead of 2000words, I say, okay, let me first
write the outline of, let mefirst write the outline with
(23:21):
Zino chat about the blog post.
And then for each of thechapters, yeah, if we divide the
problem into 10 chapters, wesuddenly have 200 words per
chapter and we have arrived atour 2000 word blog post.
And that's how you shouldapproach it.
And this also gives a much moretime for you to personalize
everything, to validate whetherthe facts are right and also to
(23:43):
put in your own research.
Still, because ultimately,these, these contents, needs to
be, needs to have some, someform of originality, and we also
, with, with that age ofunlimited creation, we've been
already living in in the floodof information before.
It will be even moreinformation now and a lot of
(24:03):
more Generic information, so itwill be even more important to
find these few little pieceswhich Still touch a human's
heart more than you do that byproper research, by creating
proper value for somebody with.
Yeah, beyond that, beyond theplay, thinking, more or less, we
would say Right.
Lance Hood (FranPro) (24:24):
And so if you're gonna do something more
complex, let's just say you'rewriting in a follow-up sequence
for your email autoresponder offollowing up with customers, and
you're you're trying togenerate Like a, you know, like
a one month or a one yearsequence or whatever.
You're gonna, instead of justtelling it to create that,
because you know who knows whatwould happen by the end.
(24:47):
You're gonna start with writethe framework, write the outline
, and then, within eachComponent, you're gonna look at
what needs to happen in thatspecific unit.
So if it's a book, you'd you'dwrite the, the premise and
everything else, and then you'dmaybe write the chap, you'd list
out the chapters and then youstart writing the chapters.
(25:08):
But if you're doing a follow-upsequence, you might say what's
you know how many days is isideal, how many Emails is ideal
for this amount of time for afollow-up, what are the things
that we should discuss?
And then you pick each one ofthose and start writing each one
of those, and so you'reproducing it Kind of in that
organized way, but not all atonce, so it can do a good job.
Dominik Lambersy (TextCort (25:31):
Yeah , exactly, there's a great
customer case study as well.
We have on the web page fromfaith.
She herself has dyslexia andshe was struggling with writing
such a.
You know, I think it was athree month long follow-up
sequence for anybody whoSubscribed for a newsletter.
As she was struggling with thatfor seven months, she found our
extension, started workingaround it exactly how you
(25:53):
described it and within twohours she had nine emails over
three months.
The problem was gone Now.
But this divide and conquerapproach is so important and
that's why you still always willneed a human.
No, because for this originalthought or for the basic
validation, you still need somehuman creativity, while you know
everything else building thenarrative out of your strategy
(26:16):
can be done with an AI, forexample, and filling, filling,
filling, filling a chapter withmore meaning for, for example,
can be done by an AI.
Lance Hood (FranPro) (26:26):
Right, because if you just do it it's
just gonna fill it up.
It'll be more general andthat's why I love the things
about you know, having havingthe AI assistant that has a
baseline of information to keepit, you know, in line so that it
writes like you or writes morelike a human, because you do get
(26:47):
that general Kind ofemotionalist writing.
So can you talk about how to,maybe how people can tweak their
writing to make it write morelike an individual, make it like
write more like?
Some of the things that I havetested is you know and write
this like a copywriter with tenyears experience and blah, blah,
blah blah.
Can you give us an idea on howprompts work and how to make
(27:09):
them better?
Dominik Lambersy (TextCortex) (27:11):
I mean, we take quite a quite a
bit away from it with ourcustomers, where I feature
already More.
You basically define once thebackground, the values, for
example, and some writing styleExamples where the algorithm
then can build up upon it suchthat you know whatever Later
collaboration you have.
You basically chat with yourvirtual twin there.
(27:32):
But what does it mean?
I oftentimes take an examplefrom how organizations put brand
scripts.
Yeah, you have the backgroundof the company, you define which
type of writing style.
You want to have a text context.
We clearly say we are aninformed, more informal company.
Yeah, we are not a formalcooperation where everybody
needs to speak veryprofessionally.
(27:53):
No, we are informal, we want tobe educated and we simply want
to create value without havingany expectations from whoever
consumes our content.
And Basically, what you try todo is building those prompts
step by step by step, addingthat information to the model or
into our customized very Ifeature ultimately and Then work
(28:14):
around that and this willalways think about it.
It is sort to say theshort-term memory of the AI.
By giving that into theshort-term memory you can deal
with a lot of the limitations,large language models or, sort
to say, the crystallineknowledge base that has, which
is from 2021, more or less, ordepending on the models.
Yet it takes quite a while toretrain, retrain those models,
(28:38):
but to keep them in line, youuse the prompt, the short-term
memory, to insert moreinformation.
Lance Hood (FranPro) (28:45):
And and about how far?
Because if you're talking withthe chat, right it?
It looks at the previous chatand applies it to what you're
doing too.
So how many characters backdoes it typically go?
Dominik Lambersy (TextCorte (29:00):
And it really depends on the models
.
The issue is now there'sthere's now there's a new model
around which says we have ahundred thousand, a hundred
thousand tokens of Window, ofshort-term memory window.
I'm not quite sure.
I haven't tested it out thatfully, because a hundred
thousand tokens are seventy fivethousand words, which is quite
(29:23):
a long text, and to fullyvalidate whether the AI has not
hallucinated anywhere within somuch information Is is a longer
task.
No, I think there's a lot ofresearch happening around that
model.
Now I wouldn't trust it at themoment.
I know what works rather well isif you have somewhere between
Two, three thousand words.
Yeah, that's where I know thatalso our users are very happy.
(29:46):
I tested it out myself as well.
There's research which whichcan basically say, hey, yeah, it
keeps information.
Well, you can see with modelslike GPT for the biggest one is
what thirty, two thousand tokens, I believe, and that's around
what?
Twenty six to twenty eightthousand words, maybe what
twenty five thousand words, andnot quite sure about the exact
math now takes a, takes a lotmore time, certainly to compute
(30:12):
as well, but also currently I'mnot quite sure how, how much
information is still retained inthe short-term memory.
But yet again, I would keep itconcise.
Yeah, ultimately, it's justlike if you overload a human
with information, they getconfused.
Now I'm getting confused if youask me too many questions.
(30:32):
If you ask me more than 10questions, I'm just like hey,
wait, before we go deeper downthe rabbit hole of questions,
let's first answer the firstfive questions and then we can
move on, because otherwise thequality of communication will
suffer.
It's same.
It's the same thing with a icemm-hmm.
Lance Hood (FranPro) (30:51):
And what do you?
I Know that you guys have allthe Templates where if whatever
you're trying to do, you justclick on it and it's already pre
kind of designed for you andready to go, making it easy.
This is really kind of a greatway for people to just free up
their day and start using thesetasks to just get things going.
(31:12):
I like how we discuss that.
You have to keep an eye on it.
I think people didn't mayberealize that there's a reason
that you do Words at a certainlength because you know you work
at a certain length becauseotherwise it's unlimited.
Creativity can take it down arabbit hole versus focus on what
you want, and then we have tobe more specific when we use it.
(31:33):
What is the?
I was just gonna say what sortof market research can we do
that would be effective?
You know, asking about, youknow competitor analysis or To
help develop a?
You know an avatar, a customeravatar or anything like that Is
(31:57):
there?
How reliable would it be to askit to do some stuff like that
to give us a baseline to startfrom?
Dominik Lambersy (TextCortex) (32:03):
I Mean from our last conversation
.
We actually made great progresson our knowledge connectors.
So one of our engineers he'soriginally from Turkey, he's
immigrating to Germany now andas a special project and beta
Piloting project, he put in thewhole German law into Guess
what's their knowledge base,uploaded via via PDF, and now he
(32:27):
uses text, cortex and Zenobasically as a Communication
layer to talk to that knowledge.
This is a feature which will berolled out in the next month
and I'm very excited to see whathappens there, because
basically, what we havedeveloped there is Sort of a
knowledge search through a lot,and I'm talking about German law
(32:50):
.
Well, german laws, one of themost complex law systems on this
planet.
Yeah, that's also why we knowfor an exporter of bureaucracy
and regulation was more or less.
But yeah, he asked questionslike hey, how long do I have to
be in Germany before I can applyfor working permit, how long
(33:11):
for residency, and it basicallytook the legal information and
made it human understandable.
Yeah, not German legal speak,which is a nightmare I mean,
english legal speak is also anightmare but makes it much more
understanding for for thecommon, for the common person,
these type of new developments.
I mean, we are also not theonly one who work on on that.
(33:32):
To be fully transparent, thiscould make a very, very great
research Retainer tool as well.
So, basically, what you puttogether is maybe one word.
It could be a word document, itcould be a notion page, it
could be just a PDF documentwhich you upload to text cortex
and then Our system is searchingthrough your knowledge,
(33:54):
integrating this into yourgeneration.
That's still heavily better.
Yeah, I hope that this Comesout to the point how we envision
it now, and it certainly willbe in one year from now.
Currently, what your best, bestwith us do your research on a
separate paper you would need todo it anyways and then you
start dividing, conquering a bigproblem and basically writing
(34:17):
down whatever you want toachieve if it's a blog post
around how to Improveprofitability in your marketing
operations for franchise, andyou have three key facts you
want to talk about.
Lance Hood (FranPro) (34:27):
At those three key facts, to your prompt
you talked about Fivereoccurring problems around
marketing, sales andcommunications that most people
are using this for, and I justwanted Check with you what.
What are those five things thatare really being used the most?
Dominik Lambersy (TextCorte (34:44):
Hmm , yeah, first pillar text
modification.
Yeah, rewrite text.
I think you you know what.
You start writing an email, youstart writing a blog post you
already mentioned and you don'tFully feel fulfilled with what
you've written.
Yeah, you can use our rewritefunction to rewrite it.
You can use the tone changer tomake it sound informal, formal,
more enthusiastic that's one ofmy favorite features because,
(35:07):
as a German, I'm not usually themost enthusiastic writer, so
that can help me to make it alittle bit more friendly.
But also summarization, forexample, boiling down heavy
corpus of texts where we alreadyhave this, this.
That's where the wholeknowledge connector Feature has
(35:28):
started, with us summarizing alot of information down such
that you know it's truthful andyou can work with it.
That's the second point.
Front point improving yourproductivity with an AI agent.
We talked a lot about it today.
Yeah, you can use it for Forimproving your communication in
emails.
We can see how people love touse it in the Communication with
(35:49):
colleagues on slack, on discord, even.
You can use it to align, forexample, to different writing
styles as well.
If you know that your customeris more of an analytical
communicator, you can basicallyalso work together with Zino and
saying hey, please.
My receiver of that message isa very analytical person.
(36:11):
How can I improve my writing ofmy email now such that it fully
resonates with that personimpersonates somebody else?
Translation is also a huge issue.
I mean, we are a German companyor European company, the
continent of any type oflanguage.
It's very difficult.
(36:32):
You just go around 100 to 100kilometers.
You're in a completelydifferent country speaking a
completely different language.
I know from my visits to the USyou don't have that problem.
Everybody speaks English, maybea different accent, but somehow
it's somewhere similar.
But that translation isdefinitely something where we're
helping a lot the dyslexia andhelp for neurodiversity.
(36:53):
I told you I'm very proud ofthe problem we are solving there
.
A lot of people use thetext-to-speech feature there.
There will be sooner voicecommand features where you can
basically just talk out yourprompt and whatever you want to
do.
You don't need to write it down.
That comes from your voice.
What else am I very proud of interms of the features we are
(37:15):
offering?
Yeah, also the templates.
Currently we offer, I believe,around 80 default templates.
We're currently in the makingof increasing that to over 1,000
templates.
It takes a little bit more time, also for the user.
There will be a full editorwhere you can save whatever
template you constantly workaround, because it gets very
(37:37):
tedious at one point.
If you always ask a chat, gptor a ZenoChat, hey, please take
the following, rewrite it in amore academic way and then you
copy-paste everything in there.
If you do that five times,always using the same prompt,
you want to have a template forit.
Similar things also around thecustomization.
I have my virtual twin calledDominik Lamperse on
(38:01):
text-to-speech.
I feed it with my backgroundborn and raised on German
funfairs, venture capital,always tech businesses, the
eight values I'm following aftermy writing style.
Then I also use that persona tobasically answer customer
support emails, also salesemails, ultimately write some
(38:22):
content for us around thecompany, around our narrative or
whatever type of value I wantto provide, or LinkedIn posts,
for example.
That's also a thing, right.
The last point I want to finishon is to make a customizable
experience out of the verygeneric AI world you currently
have.
Lance Hood (FranPro) (38:44):
Are there any last insights that you want
to share with everyone todayabout AI or using ZenoChat?
Dominik Lambersy (TextCort (38:53):
What I can see in a lot of people
currently is a lot of doubt, alot of conspiracy as well.
Ultimately, I had a beautifulslide, said that I don't have
the slide that can now preparesomewhere around, but we kept on
talking that machines can onlybe as smart and as useful as the
human who use it.
(39:13):
That's pretty much also thecase.
Even with as creative systemswe are getting now, they're only
as good as the user who'sbehind it, who thinks about what
they want to create.
I really hope that also oureducational system is picking up
and educating the newgeneration how to work with it.
(39:34):
You can see already now youngergenerations who work with Zeri
to solve their math issues.
In the next five years, we willall need to adapt to AI
collaborators, like we did adoptto the personal computer, for
example, to all the informationwe have of our smartphones.
This will be common now, andthe earlier we adopt to it, the
(39:58):
more we can flourish.
I think that's a good word toend.
Lance Hood (FranPro) (40:02):
Instead of just searching with AI, we can
now use it to create.
We have unlimited creativity,so now it's going to be it's
also going to have to help ussift through all this new
creativity that's being thrownout into the world to find what
we're looking for.
Dominik Lambersy (Te (40:16):
Absolutely .
Lance Hood (FranPro) (40:18):
Well, I just tell everybody that right
here next to the video, is alink to Text Cortex.
I think that if you apply whatyou've learned here today, you
use those personas, set those upand use those and then go in
and you'll have that browserextension.
You can start using thisinformation to chunk it.
Make sure that you're clear inyour writing style and take
(40:40):
advantage of all the training,the blog posts and the knowledge
bases here, because if you justpause for a second and go
through the training andunderstand it, you'll get an
even better result.
I mean, I've found that resultsare great from the beginning,
but if I keep tweaking with howI talk to it, I actually get
what I really want, and thenonce I know that I can use it
(41:02):
forever, so I would takeadvantage of it.
Thank you, Dominik, I reallyappreciate it.
Thanks, lance, talk to you soon.
https://Franpro.
vip/GoTextCortex.
Dominik Lambersy (TextCortex (41:11):
It was a pleasure.
Hi everyone.
Today I'd like you to meetDominic Lamberzi, the co-founder
of Text Cortex.
Hi Dominik.
Dominik Lambersy (TextCorte (00:17):
Hey Lance, Thanks for having me.
It's a pleasure to be here.
Lance Hood (FranPro) (00:20):
Well you, I'm so glad that we're on this
call today and enjoying this andhelping out the franchise
companies.
Can you explain what does?
Because it does AI writing, butyou guys also are very
different, I think.
Dominik Lambersy (TextCort (00:37):
Yeah , thanks for the brief
introduction.
Actually, up until the end oflast year, we were known as a
writing assistant, ai writingassistant a little bit like
Grammarly, but much more capablein doing things.
Ever since we released ourversion of chat GBT we call it
Xenochat People have startedseeing us as a virtual assistant
(00:57):
to their work.
Yeah, what I am currentlybuilding on our narrative around
is that we want to buildpersonalized intelligences.
Why personalized?
Because a lot of solutions youcurrently see around they give
you pretty generic outputs andresults, and with , you don't.
You do not only have acompanion with is with you on
over 4,000 websites andplatforms, but it's also
(01:20):
customizable on how you act asan individual or organization,
plus also knowing what you knowas an individual or organization
, and therefore, yeah, ending upin a much more personalized
experience for you.
Think about your own little AIcoworker.
Lance Hood (FranPro) (01:39):
Right?
Well, it sounds like you canload up a bunch of stuff that
your company has written, sothat then, when it does create
copies so let's just say blogposts, emails, responses,
whatever it's going to write itthrough the lens of all the
stuff you've written before, sothat it writes in your unique
(01:59):
voice every time, not a generaloutput, correct?
Dominik Lambersy (TextCort (02:04):
Yeah , exactly, we did this whole
feature built around two pillars.
Yeah, first of all,customization how you act, what
you give is basically sort of abackground you as a person,
where you're coming from, yeah,what is your history and, for
example, the values you areacting after and what you then
can do is also adding a fewwriting examples, such that our
(02:25):
algorithm can map your digitaltwins, so to say, on over 60
variables.
This is one pillar.
The second pillar are knowledgeconnectors yeah, where you, for
example, would can connect,upload any type of PDF files,
connect your notion, connectyour Google Drive.
The knowledge connectors arestill in beta for the for the
time being, but will be alsoreleased very soon.
(02:46):
And then it's just, like yousaid, basically building up on
what you have learned in thepast as an individual or
organization.
Again, because, if you thinkabout the main disadvantage of
many of these, many of theselarge language models, whether
it be chat, gpt or any other AItools, is that they are stuck in
the past.
They are mostly havinginformation from 2021, from the
(03:11):
European continent.
Yeah, if you would ask for afamous soccer, famous, famous
sports event, like the soccerWorld Cup here, who was the
winner of the 2022 soccer WorldCup.
Most AI tools cannot answer it,so what we have done with Zeno
is basically integrating a lotof different knowledge fields,
also including search, including, in the future, own
(03:32):
personalized files, whatever youwant to share, certainly, we
try our best to make it assecure as possible as well.
Lance Hood (FranPro) (03:42):
And so, as we go through the functionality
, we've gone through thepersonal AI assistant that does
writing.
We've talked about how Zenochat I mean there is access to
chat GPT, but we have the Zenochat, which is its own model and
then what are the otherfunctionalities that people
should be aware of?
Dominik Lambersy (TextCorte (04:02):
The other functionalities we have.
We have over 200 use cases bynow.
We have plenty of differentusers from from plenty of
different fields.
Maybe let's roll up thatquestion a little bit
differently and talk about whoare our top users.
The top users are people fromthe creative space, oftentimes
marketers or copywriters.
That's by far the biggest group.
(04:24):
Then we have yeah, we call themcorporate professions.
Oftentimes those are salesfunctions, those are customer
support functions, but alsomanagement functions, because a
lot of people deal withcommunication and prove their
communication of our tools.
And the third one is SMEs, andthat's a combination of what
(04:48):
I've what I've previously said.
Yeah, we help them withmarketing, with sales, general
communication, a lot.
And then we also have a lot ofstudents, which I don't believe
is so important right now forour call here, but yeah, they
are getting some help with theirhomework, more or less, so also
improve their communication,their writing.
Lance Hood (FranPro) (05:04):
Well, did you want to do a brief demo
today, dominic, or?
Dominik Lambersy (TextCortex (05:10):
Oh , I mean absolutely.
I can show people brieflyaround.
Let me just quickly share myscreen and find where I already
I think it's this one.
Let me check.
Yeah, I think that's the right.
Can you see my screen now?
Lance Hood (FranPro) (05:28):
I can yeah .
Dominik Lambersy (Tex (05:29):
Beautiful .
So first of all, yeah, what Iwill, what I will show you today
, is the workings of the Chromeextension.
I can see it here in the topright.
The Chrome extension is active.
You can change it.
I mean, I haven't changed inEnglish.
You can change it in differentlanguages.
That's one of the majorproblems we had recently,
because we have global impact onmany different language
speakers and they want to see itin German, spanish, french.
(05:52):
But let's get down to the usecases here.
For example this is a callsummary when I talked with you,
for example in the past, or withany type of potential customer,
I always note down things inthe most uncanny ways, with
really weird abbreviations, forexample, just like quickly
(06:12):
shutting it down and trying tojust retain knowledge.
What I can do now sorry, that'snow being German basically says
press a shortcut for Zeno.
Why is it German nowInteresting.
Lance Hood (FranPro) (06:32):
Yeah, I like it.
I like it there, I've used itand what I like is, as long as
you see that pink and bluecircular icon there, you know
that you can just click on thatand access it right away.
Dominik Lambersy (TextCorte (06:46):
Yes , exactly, I want to show you
now first the Zeno assistfeature, for example, and what I
can do now is I can say, hey,please summarize me the above.
It works.
It says the demo call was withBen Tobito's restaurant chain,
owner of TechSavvy.
It makes the whole thing alittle bit more understandable.
(07:06):
This is one of the one things.
The same thing you can do withour toolbar, for example.
It's very easily happening justby highlighting anything and
then clicking rewrite, expandingcreative writing, email writing
as well.
So this toolbar is usually ourmost popular features fixed
(07:31):
spelling and grammar completion,for example.
I could also, if I now saygrammar fix, it should fix that
really weird.
Yeah, look, with the code andmy weird abbreviation here we
could bring revenue potential 5X, but it didn't really summarize
it.
For summarization, we wouldclick this year and we would
arrive at a very similar example, like we had with Zeno assist
(07:53):
here.
What you mentioned with thelittle bubble, that's our access
to wait.
I need to move you a little bitto the over here.
This is where our Xenoy chatcomes into effect, yeah, where
we can say, hey, write me theoutline of blog post on
(08:14):
increasing franchiseopportunities via follow up
emails.
Let's see what it will writeusually takes around 10 seconds
up here In no time.
(08:35):
We already have our outline.
Then we can start as a humanbeing.
Then we can go back into thedivide and conquer approach
(08:57):
where you can now say, hey,let's expand on each chapter and
build it step by, step by step.
The whole thing works on yeah, Imentioned someone before 4,000
different platforms.
We have here this content brief, which is part of one of our
YouTube sessions, for example,as well, where you can hear now
also say, hey, we have the localSEO topic here, please expand
(09:23):
on that Can make it a little bitlonger, takes a little while,
and it also starts writing here.
You see again, it startedcreating an outline, basically
filling that with a little bitmore of life.
I click on the format thing isa little bit broken now and then
(09:45):
you can already go step by stepdown the whole the whole road,
also working on Gmail, forexample.
Here's again the bento Beatlesexample.
In Gmail, I can just say, hey,I take these bullet points as I
already have made up my thoughts.
Just highlight it, press on theemail button and it basically
(10:05):
formulates out an email for you.
Maybe we need to have it alittle bit longer.
That's also when I still do ourcustomer support.
A lot of times I just copypaste the original email from
our customers into one of ourtemplates, just drop it in here.
I give some directions, as Ihad it here with the bento
(10:31):
example, because you always have.
If you read an email, youalways have these three, four,
five next steps you want to doup on top of your head.
You can write them down here.
Basically let the AI create.
I won't create it now because Idon't have an original email
here, but then I already have myemail formulated out.
(10:52):
That comes at much higherresponse time, certainly, but
also at less psychologicalburden.
I mean, this is what usuallycustomer support users use text
cortex for exactly this view,and I know it also.
Sometimes you have customerswhich are plainly not nice,
which is a psychological strainfor many users, customers as
(11:15):
well.
There's so much about the textcortex plug-in.
I don't want to go deep, I justwant to show it briefly.
In the customization suite youneed to go to our web app for
that and then you have theKnowledge Connectors Customize
your AI and this is basicallywhere you can start Persona name
, persona block background.
(11:37):
I think we talked two or threetimes in here.
We just type down here's theexample from my co-founder,
jehun some values, a description, and then you would enter
different types of writingexamples.
I can show you with my littledigital twin how I did it.
This is then the result.
After analysis, it tells methat apparently I'm a very
(11:58):
enthusiastic, motivational,inspirational person.
I think I should be, as afounder, that reading is.
I really try to sound as easyas possible such that everybody
can understand that sentencelength, medium, active voice.
You can see the background ofmyself.
I hear values, I like to workafter examples, and then you can
(12:24):
basically use this Personaright within the note chat.
Lance Hood (FranPro) (12:29):
So you've taken the time to fill out all
that information and set it up,and now you're writing in the
Persona of this AI Persona,which is look, that's my comment
(12:52):
pitch Born and raised on Germanfun fair selling sausages and
fries.
Dominik Lambersy (Text (12:57):
Customer service.
Yeah, that's my first pocketmoney in gaming in the early
2010s, before NFTs were a thing,and suddenly I have my virtual
twin, who replies a lot like meand can give advice not fully
like me because it doesn't haveall the knowledge I have, but
(13:17):
that will soon come here as wellwith our Knowledge Connectors
where you can integrate allsorts of different things and,
before overloading your viewersmore, if somebody's interested,
we have a lot of blog postsaround how you can use all
features videos around them aswell to best educate you.
Lance Hood (FranPro) (13:41):
Yeah.
So I think that this custom AIis really a great way to get
around the general genericresponses people would get.
But also, not only are theygeneric, but they're not going
to really follow along with yourcore values or anything like
that and your more humanisticpersonality driven writing
styles.
This is kind of the shortcut orhack, for that is to set up
(14:04):
your persona.
They would also let people knowthat the people who are writing
stuff for the company, or ifyou're a marketer that's doing
work for a company, you canupload stuff on that specific
company in here and then, oh,today I'm writing for franchise
A, so I use this persona to that.
You know, now I'm writing forfranchise B, I'll use this
(14:26):
persona.
Dominik Lambersy (Te (14:27):
Absolutely , absolutely.
Lance Hood (FranPro) (14:29):
Yeah, so it helps you produce stuff much
better.
So Well, this is excellent.
You know, one of the thingsthat I like about the usability
of this is that you have thebrowser extension, so wherever
you're working, the XenoChat canstart writing things for you.
You know, it's very handy.
Dominik Lambersy (TextCort (14:50):
Yeah , it's very handy for 100,000 of
people who use it actively on4,000 platforms now, and I can
see our competitors and they cansee how people use it in Google
Docs.
They use it on Outlook, gmail,even chat, gpt, for example.
So when our belief and ourvision was back in the days,
(15:10):
nobody wants another web app totap in between all the time.
We just want to be next to ouruser, right where they need us.
Lance Hood (FranPro) (15:21):
Right.
Well, you know as I think aboutusing and implementing this
stuff, because I've always lovedcopywriting, done copywriting
and this tool when I first foundit was just absolutely amazing
because it overcomes the whole.
I don't know where to start.
If you just give me a whitesheet of paper and say write
this, it's like there's too muchcreativity there.
(15:43):
I don't know where to startwithout having you know some
sort of preexisting tests ortemplates or baseline of some
kind.
I can improve very easily, butimproving on nothing is hard.
So this is a great way to getstarted, but you know to talk
about how to talk to it andimprove it.
I think, first of all, it'sgood to understand what is AI,
(16:05):
because I think people have somemisconceptions or
non-understandings on what it is, and you have to know what it
is and how it works so that youcan talk to it the best and get
the best results.
So what is AI?
Is it this big computer systemthat's going to take over the
world?
Dominik Lambersy (TextCort (16:21):
It's a great question.
I've also, last week, have beenpart of a finance summit as a
keynote speaker for GenerativeAI and there was similar
misconceptions.
I mean, ai is nothing so new.
Generative AI is now new.
What is Generative AI basicallydoing?
It tries to create something.
I like to call it also creativeAI because from what it has
(16:43):
learned from all the tens ofthousands of texts or, in terms
of mid-journey tens of millionsof pictures it has seen, it can
draw and learn patterns which weusually combine with some sort
of textual structure and canrecreate on top of it.
What happens under the hood ismore or less an auto completion.
(17:04):
So if you would ask an AI, hey,please write me the blog
outline for a blog around thetopic of how I can improve my
marketing ops for my franchise.
The AI basically takes this asinput and improves.
Or hello, Siri, I'm talkingabout AIs why?
(17:24):
Why does it come up now?
It tries to auto complete yourprompt.
That's why it's also soimportant that that's what I've
seen with many users.
One or two years ago from ourproduct, they talked to Zeno, to
text cortex, like it was amachine.
But ultimately, what you need toreally think about is how you
(17:47):
communicate and I believe thiswhole thing, the soft skill of
communication, will be much moreimportant now in the upcoming
future in the collaboration withAI systems.
Because, in comparison to if wetalk with our team, for example,
and I tell one of my teammembers, hey, it would be great
if I have a result to topic XYZin one to two days, I get a
(18:09):
response in one to two days withan AI, have instant feedback.
So if I ask it to write anoutline for a blog post, I will
have a result within 10 seconds.
If I don't like the output,maybe it was more or less about
me, because write a blog outlinefor a topic X is not very
defined.
So, for example, expand on theprompt.
(18:30):
Hey, the topic is generative AIin the franchise ecosystem.
I want to talk about themarketing operations and how to
decrease costs, for example, orhow can I scale up for more
opportunities if I add somesources, some real time data,
(18:51):
that really helps the AI tounderstand where you want to
arrive at.
And these basic instructionsare incredibly important and
that's why I say communicationas a soft skill will be much
more important in ourcollaborative ways around AI in
the future.
Lance Hood (FranPro) (19:10):
So you need to talk to it like you talk
to people and just like you'rehaving a conversation with
another person.
That's how you talk to AI.
But also one thing I've learnedfrom talking with you in the
past is that AI is creative.
It's made to be creative and soit can go off all over the
place because it's just designedto be creative.
(19:32):
So the more you give it moreinformation and you type in more
stuff or to give it direction,it's going to be more of what
you want.
So if you just say, write thisand you get a general response,
that's because what you'resaying is too general and so
it's just going okay and itanswers it.
But if the more precise you get, you'll get it.
(19:52):
And that's why you also saidthat if AI is doing, if you were
using AI to do some sort ofresearch, you'd want to just
double check it, because it'svery creative and it can try to
autocomplete and not necessarilyhave autocomplete.
Be 100% accurate with facts.
Dominik Lambersy (TextC (20:12):
Correct .
Yeah, exactly, that's a reallygood point from our former
conversations Always validatewhat the AI is creating for you.
I mean, six months ago,everybody believed what AI has
written, which is extremelydangerous.
It's just like how you interactwith other human beings.
Take everything with a grain ofsalt and do your own critical
(20:32):
thinking around it.
There are plenty of people whotry to sell you something and
lie to you.
Similar things can also happenwith any type of AI generation.
So always double check facts.
There was another point which Iwanted to add to that, but
sincerely it flew out of my mindnow.
Maybe it comes in a little bitlater down the road, but
(20:55):
validation of information isextremely important by an AI
model and it's kind of thereason why AI's generally have a
limited output, because you'll,you'll, uh, you can.
Lance Hood (FranPro) (21:08):
it can go out too far off in the wrong
direction if you don't Overseeit, and I've always kind of
looked at.
After I started using it, IStarted looking at AI as more of
like a race car, but it needs areally good human driver, you
know like it's a little course.
What's that?
Dominik Lambersy (TextCortex) (21:26):
a really good pilot to it, huh.
Lance Hood (FranPro) (21:28):
Yeah, yeah , exactly.
Somebody's got to be in controlof it, Otherwise it's not gonna
have as much value.
So if you're not getting thevalue out of your AI and it's
not producing what you want, youjust have to learn how to talk
to it and you have to learn howto drive it and direct it, and
then you have to oversee theoutput.
So it's not necessarilyEliminating a lot of humans.
(21:48):
What it's doing is, is it justit produces way more, so you can
get more, way more output perperson.
Dominik Lambersy (TextCort (21:55):
Yeah , that's what I also said in
that in that speech recentlylast week.
We basically go now from an agefrom Unlimited access to
information to an unlimited, tothe age of unlimited, unlimited
creation.
Basically, what happens is thatwe do not need to think about
all these information whichcomes to us at the moment and
(22:16):
then need to produce something.
The production is alreadyHappening, also on the AI level,
but still it's our job tovalidate whether it's right and
proper, fitting Our use case andgenerally true, and I also just
got my point break which I justlost a minute ago.
One concept which we're alsopraising to our users is called
divide and conquer.
It's nothing new.
Yeah, with many big problemsyou have in life, you should
(22:39):
divide and conquer them intosmaller ones, and we have this
beautiful YouTube video as wellwhen we basically show people
through how they can collaboratewith AI to create blog posts.
Now, if your goal is a 2000word blog post, your
expectations shouldn't be thatthe AI is Just snapping the
finger and you have a 2000 wordblog post in front of you, which
is perfectly fitting your case.
(22:59):
That's the.
The short-term memory of an AIis not enough to keep that
concise and to keep, keep andkeep the truthfulness behind it.
So what you do?
You divide and conquer theproblem.
For example, instead of 2000words, I say, okay, let me first
write the outline of, let mefirst write the outline with
(23:21):
Zino chat about the blog post.
And then for each of thechapters, yeah, if we divide the
problem into 10 chapters, wesuddenly have 200 words per
chapter and we have arrived atour 2000 word blog post.
And that's how you shouldapproach it.
And this also gives a much moretime for you to personalize
everything, to validate whetherthe facts are right and also to
(23:43):
put in your own research.
Still, because ultimately,these, these contents, needs to
be, needs to have some, someform of originality, and we also
, with, with that age ofunlimited creation, we've been
already living in in the floodof information before.
It will be even moreinformation now and a lot of
(24:03):
more Generic information, so itwill be even more important to
find these few little pieceswhich Still touch a human's
heart more than you do that byproper research, by creating
proper value for somebody with.
Yeah, beyond that, beyond theplay, thinking, more or less, we
would say Right.
Lance Hood (FranPro) (24:24):
And so if you're gonna do something more
complex, let's just say you'rewriting in a follow-up sequence
for your email autoresponder offollowing up with customers, and
you're you're trying togenerate Like a, you know, like
a one month or a one yearsequence or whatever.
You're gonna, instead of justtelling it to create that,
because you know who knows whatwould happen by the end.
(24:47):
You're gonna start with writethe framework, write the outline
, and then, within eachComponent, you're gonna look at
what needs to happen in thatspecific unit.
So if it's a book, you'd you'dwrite the, the premise and
everything else, and then you'dmaybe write the chap, you'd list
out the chapters and then youstart writing the chapters.
(25:08):
But if you're doing a follow-upsequence, you might say what's
you know how many days is isideal, how many Emails is ideal
for this amount of time for afollow-up, what are the things
that we should discuss?
And then you pick each one ofthose and start writing each one
of those, and so you'reproducing it Kind of in that
organized way, but not all atonce, so it can do a good job.
Dominik Lambersy (TextCort (25:31):
Yeah , exactly, there's a great
customer case study as well.
We have on the web page fromfaith.
She herself has dyslexia andshe was struggling with writing
such a.
You know, I think it was athree month long follow-up
sequence for anybody whoSubscribed for a newsletter.
As she was struggling with thatfor seven months, she found our
extension, started workingaround it exactly how you
(25:53):
described it and within twohours she had nine emails over
three months.
The problem was gone Now.
But this divide and conquerapproach is so important and
that's why you still always willneed a human.
No, because for this originalthought or for the basic
validation, you still need somehuman creativity, while you know
everything else building thenarrative out of your strategy
(26:16):
can be done with an AI, forexample, and filling, filling,
filling, filling a chapter withmore meaning for, for example,
can be done by an AI.
Lance Hood (FranPro) (26:26):
Right, because if you just do it it's
just gonna fill it up.
It'll be more general andthat's why I love the things
about you know, having havingthe AI assistant that has a
baseline of information to keepit, you know, in line so that it
writes like you or writes morelike a human, because you do get
(26:47):
that general Kind ofemotionalist writing.
So can you talk about how to,maybe how people can tweak their
writing to make it write morelike an individual, make it like
write more like?
Some of the things that I havetested is you know and write
this like a copywriter with tenyears experience and blah, blah,
blah blah.
Can you give us an idea on howprompts work and how to make
(27:09):
them better?
Dominik Lambersy (TextCortex) (27:11):
I mean, we take quite a quite a
bit away from it with ourcustomers, where I feature
already More.
You basically define once thebackground, the values, for
example, and some writing styleExamples where the algorithm
then can build up upon it suchthat you know whatever Later
collaboration you have.
You basically chat with yourvirtual twin there.
(27:32):
But what does it mean?
I oftentimes take an examplefrom how organizations put brand
scripts.
Yeah, you have the backgroundof the company, you define which
type of writing style.
You want to have a text context.
We clearly say we are aninformed, more informal company.
Yeah, we are not a formalcooperation where everybody
needs to speak veryprofessionally.
(27:53):
No, we are informal, we want tobe educated and we simply want
to create value without havingany expectations from whoever
consumes our content.
And Basically, what you try todo is building those prompts
step by step by step, addingthat information to the model or
into our customized very Ifeature ultimately and Then work
(28:14):
around that and this willalways think about it.
It is sort to say theshort-term memory of the AI.
By giving that into theshort-term memory you can deal
with a lot of the limitations,large language models or, sort
to say, the crystallineknowledge base that has, which
is from 2021, more or less, ordepending on the models.
Yet it takes quite a while toretrain, retrain those models,
(28:38):
but to keep them in line, youuse the prompt, the short-term
memory, to insert moreinformation.
Lance Hood (FranPro) (28:45):
And and about how far?
Because if you're talking withthe chat, right it?
It looks at the previous chatand applies it to what you're
doing too.
So how many characters backdoes it typically go?
Dominik Lambersy (TextCorte (29:00):
And it really depends on the models
.
The issue is now there'sthere's now there's a new model
around which says we have ahundred thousand, a hundred
thousand tokens of Window, ofshort-term memory window.
I'm not quite sure.
I haven't tested it out thatfully, because a hundred
thousand tokens are seventy fivethousand words, which is quite
(29:23):
a long text, and to fullyvalidate whether the AI has not
hallucinated anywhere within somuch information Is is a longer
task.
No, I think there's a lot ofresearch happening around that
model.
Now I wouldn't trust it at themoment.
I know what works rather well isif you have somewhere between
Two, three thousand words.
Yeah, that's where I know thatalso our users are very happy.
(29:46):
I tested it out myself as well.
There's research which whichcan basically say, hey, yeah, it
keeps information.
Well, you can see with modelslike GPT for the biggest one is
what thirty, two thousand tokens, I believe, and that's around
what?
Twenty six to twenty eightthousand words, maybe what
twenty five thousand words, andnot quite sure about the exact
math now takes a, takes a lotmore time, certainly to compute
(30:12):
as well, but also currently I'mnot quite sure how, how much
information is still retained inthe short-term memory.
But yet again, I would keep itconcise.
Yeah, ultimately, it's justlike if you overload a human
with information, they getconfused.
Now I'm getting confused if youask me too many questions.
(30:32):
If you ask me more than 10questions, I'm just like hey,
wait, before we go deeper downthe rabbit hole of questions,
let's first answer the firstfive questions and then we can
move on, because otherwise thequality of communication will
suffer.
It's same.
It's the same thing with a icemm-hmm.
Lance Hood (FranPro) (30:51):
And what do you?
I Know that you guys have allthe Templates where if whatever
you're trying to do, you justclick on it and it's already pre
kind of designed for you andready to go, making it easy.
This is really kind of a greatway for people to just free up
their day and start using thesetasks to just get things going.
(31:12):
I like how we discuss that.
You have to keep an eye on it.
I think people didn't mayberealize that there's a reason
that you do Words at a certainlength because you know you work
at a certain length becauseotherwise it's unlimited.
Creativity can take it down arabbit hole versus focus on what
you want, and then we have tobe more specific when we use it.
(31:33):
What is the?
I was just gonna say what sortof market research can we do
that would be effective?
You know, asking about, youknow competitor analysis or To
help develop a?
You know an avatar, a customeravatar or anything like that Is
(31:57):
there?
How reliable would it be to askit to do some stuff like that
to give us a baseline to startfrom?
Dominik Lambersy (TextCortex) (32:03):
I Mean from our last conversation
.
We actually made great progresson our knowledge connectors.
So one of our engineers he'soriginally from Turkey, he's
immigrating to Germany now andas a special project and beta
Piloting project, he put in thewhole German law into Guess
what's their knowledge base,uploaded via via PDF, and now he
(32:27):
uses text, cortex and Zenobasically as a Communication
layer to talk to that knowledge.
This is a feature which will berolled out in the next month
and I'm very excited to see whathappens there, because
basically, what we havedeveloped there is Sort of a
knowledge search through a lot,and I'm talking about German law
(32:50):
.
Well, german laws, one of themost complex law systems on this
planet.
Yeah, that's also why we knowfor an exporter of bureaucracy
and regulation was more or less.
But yeah, he asked questionslike hey, how long do I have to
be in Germany before I can applyfor working permit, how long
(33:11):
for residency, and it basicallytook the legal information and
made it human understandable.
Yeah, not German legal speak,which is a nightmare I mean,
english legal speak is also anightmare but makes it much more
understanding for for thecommon, for the common person,
these type of new developments.
I mean, we are also not theonly one who work on on that.
(33:32):
To be fully transparent, thiscould make a very, very great
research Retainer tool as well.
So, basically, what you puttogether is maybe one word.
It could be a word document, itcould be a notion page, it
could be just a PDF documentwhich you upload to text cortex
and then Our system is searchingthrough your knowledge,
(33:54):
integrating this into yourgeneration.
That's still heavily better.
Yeah, I hope that this Comesout to the point how we envision
it now, and it certainly willbe in one year from now.
Currently, what your best, bestwith us do your research on a
separate paper you would need todo it anyways and then you
start dividing, conquering a bigproblem and basically writing
(34:17):
down whatever you want toachieve if it's a blog post
around how to Improveprofitability in your marketing
operations for franchise, andyou have three key facts you
want to talk about.
Lance Hood (FranPro) (34:27):
At those three key facts, to your prompt
you talked about Fivereoccurring problems around
marketing, sales andcommunications that most people
are using this for, and I justwanted Check with you what.
What are those five things thatare really being used the most?
Dominik Lambersy (TextCorte (34:44):
Hmm , yeah, first pillar text
modification.
Yeah, rewrite text.
I think you you know what.
You start writing an email, youstart writing a blog post you
already mentioned and you don'tFully feel fulfilled with what
you've written.
Yeah, you can use our rewritefunction to rewrite it.
You can use the tone changer tomake it sound informal, formal,
more enthusiastic that's one ofmy favorite features because,
(35:07):
as a German, I'm not usually themost enthusiastic writer, so
that can help me to make it alittle bit more friendly.
But also summarization, forexample, boiling down heavy
corpus of texts where we alreadyhave this, this.
That's where the wholeknowledge connector Feature has
(35:28):
started, with us summarizing alot of information down such
that you know it's truthful andyou can work with it.
That's the second point.
Front point improving yourproductivity with an AI agent.
We talked a lot about it today.
Yeah, you can use it for Forimproving your communication in
emails.
We can see how people love touse it in the Communication with
(35:49):
colleagues on slack, on discord, even.
You can use it to align, forexample, to different writing
styles as well.
If you know that your customeris more of an analytical
communicator, you can basicallyalso work together with Zino and
saying hey, please.
My receiver of that message isa very analytical person.
(36:11):
How can I improve my writing ofmy email now such that it fully
resonates with that personimpersonates somebody else?
Translation is also a huge issue.
I mean, we are a German companyor European company, the
continent of any type oflanguage.
It's very difficult.
(36:32):
You just go around 100 to 100kilometers.
You're in a completelydifferent country speaking a
completely different language.
I know from my visits to the USyou don't have that problem.
Everybody speaks English, maybea different accent, but somehow
it's somewhere similar.
But that translation isdefinitely something where we're
helping a lot the dyslexia andhelp for neurodiversity.
(36:53):
I told you I'm very proud ofthe problem we are solving there
.
A lot of people use thetext-to-speech feature there.
There will be sooner voicecommand features where you can
basically just talk out yourprompt and whatever you want to
do.
You don't need to write it down.
That comes from your voice.
What else am I very proud of interms of the features we are
(37:15):
offering?
Yeah, also the templates.
Currently we offer, I believe,around 80 default templates.
We're currently in the makingof increasing that to over 1,000
templates.
It takes a little bit more time, also for the user.
There will be a full editorwhere you can save whatever
template you constantly workaround, because it gets very
(37:37):
tedious at one point.
If you always ask a chat, gptor a ZenoChat, hey, please take
the following, rewrite it in amore academic way and then you
copy-paste everything in there.
If you do that five times,always using the same prompt,
you want to have a template forit.
Similar things also around thecustomization.
I have my virtual twin calledDominik Lamperse on
(38:01):
text-to-speech.
I feed it with my backgroundborn and raised on German
funfairs, venture capital,always tech businesses, the
eight values I'm following aftermy writing style.
Then I also use that persona tobasically answer customer
support emails, also salesemails, ultimately write some
(38:22):
content for us around thecompany, around our narrative or
whatever type of value I wantto provide, or LinkedIn posts,
for example.
That's also a thing, right.
The last point I want to finishon is to make a customizable
experience out of the verygeneric AI world you currently
have.
Lance Hood (FranPro) (38:44):
Are there any last insights that you want
to share with everyone todayabout AI or using ZenoChat?
Dominik Lambersy (TextCort (38:53):
What I can see in a lot of people
currently is a lot of doubt, alot of conspiracy as well.
Ultimately, I had a beautifulslide, said that I don't have
the slide that can now preparesomewhere around, but we kept on
talking that machines can onlybe as smart and as useful as the
human who use it.
(39:13):
That's pretty much also thecase.
Even with as creative systemswe are getting now, they're only
as good as the user who'sbehind it, who thinks about what
they want to create.
I really hope that also oureducational system is picking up
and educating the newgeneration how to work with it.
(39:34):
You can see already now youngergenerations who work with Zeri
to solve their math issues.
In the next five years, we willall need to adapt to AI
collaborators, like we did adoptto the personal computer, for
example, to all the informationwe have of our smartphones.
This will be common now, andthe earlier we adopt to it, the
(39:58):
more we can flourish.
I think that's a good word toend.
Lance Hood (FranPro) (40:02):
Instead of just searching with AI, we can
now use it to create.
We have unlimited creativity,so now it's going to be it's
also going to have to help ussift through all this new
creativity that's being thrownout into the world to find what
we're looking for.
Dominik Lambersy (Te (40:16):
Absolutely .
Lance Hood (FranPro) (40:18):
Well, I just tell everybody that right
here next to the video, is alink to Text Cortex.
I think that if you apply whatyou've learned here today, you
use those personas, set those upand use those and then go in
and you'll have that browserextension.
You can start using thisinformation to chunk it.
Make sure that you're clear inyour writing style and take
(40:40):
advantage of all the training,the blog posts and the knowledge
bases here, because if you justpause for a second and go
through the training andunderstand it, you'll get an
even better result.
I mean, I've found that resultsare great from the beginning,
but if I keep tweaking with howI talk to it, I actually get
what I really want, and thenonce I know that I can use it
(41:02):
forever, so I would takeadvantage of it.
Thank you, Dominik, I reallyappreciate it.
Thanks, lance, talk to you soon.
https://Franpro.
vip/GoTextCortex.
Dominik Lambersy (TextCortex (41:11):
It was a pleasure.
Popular Podcasts
Bookmarked by Reese's Book Club
Welcome to Bookmarked by Reese’s Book Club — the podcast where great stories, bold women, and irresistible conversations collide! Hosted by award-winning journalist Danielle Robay, each week new episodes balance thoughtful literary insight with the fervor of buzzy book trends, pop culture and more. Bookmarked brings together celebrities, tastemakers, influencers and authors from Reese's Book Club and beyond to share stories that transcend the page. Pull up a chair. You’re not just listening — you’re part of the conversation.
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com