WEBVTT

00:00:07.400 --> 00:00:10.419
Hey, it's Mukundan from Data and AI with Mukundan.

00:00:10.619 --> 00:00:13.380
Have you ever written something that looked right,

00:00:13.539 --> 00:00:18.199
sounded right, but just didn't feel right? Like

00:00:18.199 --> 00:00:21.179
you hit publish on a blog post and instead of

00:00:21.179 --> 00:00:24.739
that warm glow of, yes, I nailed this, you feel

00:00:24.739 --> 00:00:28.800
kind of hollow? I've been there multiple times.

00:00:31.219 --> 00:00:34.859
So one day I just stopped guessing and I built

00:00:34.859 --> 00:00:38.359
an AI to tell me the truth. This is a story of

00:00:38.359 --> 00:00:41.219
how I exposed the emotional blind spots in my

00:00:41.219 --> 00:00:45.280
writing using artificial intelligence. It's a

00:00:45.280 --> 00:00:48.320
part emotional wake -up call, part tech deep

00:00:48.320 --> 00:00:52.420
dive and all very very personal. So stay with

00:00:52.420 --> 00:00:58.479
me here. So I'd just written a 1500 word blog

00:00:58.479 --> 00:01:01.920
post. I did everything right. I outlined it.

00:01:02.000 --> 00:01:05.620
I edited it. Reworked the sentences. My grammar

00:01:05.620 --> 00:01:09.469
was solid. My ideas were clear. But after all

00:01:09.469 --> 00:01:13.870
that work, something felt off. You know that

00:01:13.870 --> 00:01:15.969
feeling when someone smiles at you, but their

00:01:15.969 --> 00:01:18.549
eyes don't? That's what my writing felt like.

00:01:19.469 --> 00:01:24.370
On paper, it was fine. But emotionally, disconnected.

00:01:25.670 --> 00:01:30.129
Distant, in fact. And it scared me. Because why

00:01:30.129 --> 00:01:34.030
wouldn't it, right? Because I wasn't trying to

00:01:34.030 --> 00:01:40.870
be cold. I was trying to connect. Deeply. So

00:01:40.870 --> 00:01:43.549
I asked myself the question, that would change

00:01:43.549 --> 00:01:48.870
everything. What if I'm not writing too poorly,

00:01:49.010 --> 00:01:55.730
but too impersonally? And that thought haunted

00:01:55.730 --> 00:02:02.129
me. Until I did something kind of crazy. So I

00:02:02.129 --> 00:02:05.230
built an AI app to... Tell me the emotional truth.

00:02:05.450 --> 00:02:09.530
My words were hiding. So if you don't know me

00:02:09.530 --> 00:02:14.069
already from my previous work, I built a ton

00:02:14.069 --> 00:02:17.169
of little AI apps, I would call it. So these

00:02:17.169 --> 00:02:21.090
are like just simple AI apps that I use to make

00:02:21.090 --> 00:02:24.310
my life better. Because I think AI is here to

00:02:24.310 --> 00:02:30.030
help make us reach unbelievable heights. It's

00:02:30.030 --> 00:02:35.439
here to make our life really easy. And, you know,

00:02:35.460 --> 00:02:40.039
building an application using AI, that is just

00:02:40.039 --> 00:02:43.800
my way of, you know, using AI to benefit my life.

00:02:44.300 --> 00:02:46.919
And I'm hoping to help others do the same, right?

00:02:47.000 --> 00:02:49.080
So that's what I did here. I built an AI app

00:02:49.080 --> 00:02:51.819
to tell me the emotional truth that my words

00:02:51.819 --> 00:02:57.300
in my blog post were hiding. So here's what I

00:02:57.300 --> 00:03:02.740
built. It's a simple Python based app. It uses

00:03:02.740 --> 00:03:06.800
streamlit library in the backend, but it analyzes

00:03:06.800 --> 00:03:10.639
the tone of what you write, paragraph by paragraph.

00:03:12.000 --> 00:03:18.259
One model is from Hugging Face, which is a website

00:03:18.259 --> 00:03:21.139
where you get these amazing open -sourced models

00:03:21.139 --> 00:03:25.120
to use. So it could be something like image -to

00:03:25.120 --> 00:03:28.099
-text classifier. So basically, I mean, so image

00:03:28.099 --> 00:03:30.219
-to -text conversion, so something where you

00:03:30.219 --> 00:03:33.500
want to... convert an image to text or text to

00:03:33.500 --> 00:03:39.039
an image the hugging face website would give

00:03:39.039 --> 00:03:42.819
you these models to use which are open source

00:03:42.819 --> 00:03:45.740
and it could be someone from facebook or could

00:03:45.740 --> 00:03:48.180
be someone from airbnb for example right they

00:03:48.180 --> 00:03:52.900
could be open sourcing these models and making

00:03:52.900 --> 00:03:55.139
them available to you so that you can use it

00:03:55.139 --> 00:03:58.639
to build that amazing text to image classifier

00:03:58.639 --> 00:04:02.400
or text to image converter that I know you can.

00:04:04.319 --> 00:04:07.500
So in this case, I used this model from Hugging

00:04:07.500 --> 00:04:10.759
Face, like I mentioned. It detects emotions.

00:04:11.780 --> 00:04:18.199
It detects joy, frustration, sadness, excitement.

00:04:19.819 --> 00:04:26.680
And the other I used was GPT -4, which is the

00:04:26.680 --> 00:04:31.160
latest, one of the latest at this point. um gpt

00:04:31.160 --> 00:04:34.899
is from chat gpt it at this point i think it

00:04:34.899 --> 00:04:37.579
would be a bit more older but i mean yeah it

00:04:37.579 --> 00:04:43.420
is a gpt by chat gpt it tells me what tone it

00:04:43.420 --> 00:04:49.459
hears whether it's confident apologetic casual

00:04:49.459 --> 00:04:55.879
or robotic yeah if those don't match up the app

00:04:55.879 --> 00:05:02.170
flags it That's the emotional mismatch. And that's

00:05:02.170 --> 00:05:05.089
where my writing is betraying me. I ran my blog

00:05:05.089 --> 00:05:08.730
through it. And paragraph by paragraph, I saw

00:05:08.730 --> 00:05:13.870
it. My intro was meant to be uplifting. But it

00:05:13.870 --> 00:05:17.209
read uncertain. My conclusion was supposed to

00:05:17.209 --> 00:05:20.889
feel powerful. But the AI called it resigned.

00:05:22.329 --> 00:05:26.509
And I had to sit with that. Because it was right.

00:05:28.620 --> 00:05:31.300
So I used that feedback to rewrite the post,

00:05:31.500 --> 00:05:35.180
not just for clarity, but for that connection.

00:05:37.139 --> 00:05:40.480
That's when my writing started to feel like me

00:05:40.480 --> 00:05:46.060
again. So here's where I went deeper, right?

00:05:46.819 --> 00:05:51.800
I wanted the tone detector to know me, to know

00:05:51.800 --> 00:06:00.889
my quirks, my voice, my style. So I trained a

00:06:00.889 --> 00:06:05.269
model to use my own writing. So I labeled these

00:06:05.269 --> 00:06:09.610
paragraphs as this is neutral or this feels warm

00:06:09.610 --> 00:06:16.509
or this is trying too hard. Then I fine -tuned

00:06:16.509 --> 00:06:23.189
a model on those labels. Now, when I run new

00:06:23.189 --> 00:06:27.029
writing through it, it doesn't just judge me.

00:06:27.970 --> 00:06:31.970
Based on these generic emotion categories, it

00:06:31.970 --> 00:06:36.750
checks if my writing sounds like me. Not like

00:06:36.750 --> 00:06:41.389
a perfect writer, but like a real one. And that's

00:06:41.389 --> 00:06:44.810
when something magical happened. My blog started

00:06:44.810 --> 00:06:47.790
getting more comments like, this really spoke

00:06:47.790 --> 00:06:52.709
to me. I wasn't trying harder. See, I wasn't

00:06:52.709 --> 00:06:55.050
trying harder. It's just, I was trying to be

00:06:55.050 --> 00:06:59.430
real. And being real is what helps you connect

00:06:59.430 --> 00:07:03.290
with people, right? And surprisingly, AI got

00:07:03.290 --> 00:07:09.430
me there. So here's what I learned. We spend

00:07:09.430 --> 00:07:13.850
hours tweaking grammar, structure, word choice.

00:07:14.990 --> 00:07:20.889
But if the tone's off, none of it lands. Tone

00:07:20.889 --> 00:07:26.829
is invisible, but it is everything. It's... Why

00:07:26.829 --> 00:07:34.250
a helpful email feels condescending. It's why

00:07:34.250 --> 00:07:39.209
a passionate tweet sounds arrogant. It's why

00:07:39.209 --> 00:07:44.329
your heartfelt blog sounds robotic. Yeah, I just

00:07:44.329 --> 00:07:49.629
do this robot voice. So we are blind to our own

00:07:49.629 --> 00:07:54.889
emotional inconsistencies. But AI, it doesn't

00:07:54.889 --> 00:08:00.939
flinch. It reflects. It shows us our blind spots.

00:08:03.019 --> 00:08:07.819
And if we are brave enough to look, it can help

00:08:07.819 --> 00:08:11.000
us connect in ways we didn't even know we were

00:08:11.000 --> 00:08:16.120
missing. So let me ask you something. When was

00:08:16.120 --> 00:08:20.399
the last time you asked your words how they feel?

00:08:21.100 --> 00:08:23.300
You asked how your words felt exactly, right?

00:08:25.879 --> 00:08:28.000
Previously, I was just trying to give these words

00:08:28.000 --> 00:08:33.960
a person. I mean, a feeling of a person. But

00:08:33.960 --> 00:08:36.580
like if you asked how your words felt to people

00:08:36.580 --> 00:08:38.620
or maybe to yourself, just ask that to yourself.

00:08:40.559 --> 00:08:44.399
Not just what you said, but what you meant and

00:08:44.399 --> 00:08:50.240
how it actually landed. So you can try building

00:08:50.240 --> 00:08:52.679
the app that I have linked in the show notes.

00:08:55.179 --> 00:08:57.559
So the whole code is on there. You should be

00:08:57.559 --> 00:09:00.179
able to build it yourself. It's pretty easy and

00:09:00.179 --> 00:09:03.799
it's free, obviously. Free with the medium membership.

00:09:04.759 --> 00:09:08.200
Just copy paste a paragraph of your writing into

00:09:08.200 --> 00:09:13.080
ChatGPT and ask, what is the emotional tone of

00:09:13.080 --> 00:09:17.620
this paragraph? You might be shocked. You might

00:09:17.620 --> 00:09:21.159
be validated. But either way, you'll be more

00:09:21.159 --> 00:09:26.929
aware. Because that's what makes great communication

00:09:26.929 --> 00:09:32.730
today. Not perfection. Not even cleverness. But

00:09:32.730 --> 00:09:38.870
honest, emotionally resonant connection. Hey,

00:09:38.889 --> 00:09:42.330
so that's all for this episode. But if this resonated

00:09:42.330 --> 00:09:44.549
with you, please share this episode with a friend

00:09:44.549 --> 00:09:49.230
who writes, blogs, or speaks for a living. Better

00:09:49.230 --> 00:09:52.590
yet, share it with someone whose words matter.

00:09:54.000 --> 00:09:56.299
Because the world doesn't need more content.

00:09:56.480 --> 00:10:01.320
It needs more felt truth. It needs more truth.

00:10:01.379 --> 00:10:05.820
It needs people to be more themselves. And thanks

00:10:05.820 --> 00:10:08.019
for listening to Data Anya with Mukundan. And

00:10:08.019 --> 00:10:08.860
see you in the next one.
