The Small Tech Podcast by EC

Explore the world of AI tools in this episode of our podcast! Join our host, Raph, as he discusses the latest advancements in AI, specifically focusing on large language models like ChatGPT and their impact on various industries. From content generation and text restructuring to image generation and search engines, Raph shares his experiences and thoughts on how these tools can help businesses and individuals. Get insights into GitHub Copilot and its code generation capabilities, and learn about Canva and its recent AI-powered additions. Raph also discusses the potential of AI tools in the future and their role in shaping technology. Listen in and discover the exciting developments in AI as Raph delves into the world of this cutting-edge technology.

🗣️ Got feedback? Want to be a part of this? Contact us!

🚀 Subscribe to the podcast:

✉️ Subscribe to the newsletter:

🎥 Watch on YouTube:

📄 Read the EC blog:

🛠️ Need help building something?

Creators & Guests

Raphaël Titsworth-Morin
Trying to do good in the world with tech and design. I also take the occasional photograph. Co-founder of Éphémère Creative. He/him.

What is The Small Tech Podcast by EC?

A podcast about building tech products with Éphémère Creative.

There's a lot of content out there about the big tech companies and how they work. But what about solopreneurs, early stage startups, non-profits, and small businesses building tech? It turns out that building small systems is an art, and in this podcast we'll talk about what it takes.

🗣️ Got feedback? Want to be a part of this? Contact us!

🚀 Subscribe to the podcast:

✉️ Subscribe to the newsletter:

🎥 Watch on YouTube:

📄 Read the EC blog:

🛠️ Need help building something?

Hey folks, and welcome to
the small tech podcast.

I'm your host Raph from Ephemere Creative.

And today we are going to
be talking about AI tools.

So, again, this is going to be a bit
more of a loosey goosey kind of episode,

because, you know, everyone has just
been talking about these AI tools and.

They're pretty neat.

And I think.

In the context of building small
tech, there are a lot of little

like productivity gains that
you can get out of some of them.

I don't think they'll
make sense for everyone.

And I don't think.

I have lots of thoughts about
whether or not this is, this is

a big game-changing thing or not.

Fundamentally, I do think this is a new
way for people to interact with machines,

which I think is pretty game-changing.

Um, specifically thinking about tools
like chat GPT and, and those types

of, uh, large language models and the.

The chat driven interface.

Uh, for interacting with them.

In terms of how all of that stuff
fits into building small tech.

I found that there's a couple
ways in which AI tools fit into

my workflow, our workflow at EDC.


Just generally things that, that, that.

Are involved in building small tech.

So the first one I'm
going to talk about is.

Get hub.


The next is chat GPT itself.


We'll touch on the AI.

Um, Like the image generators that you can
find in, uh, in, in being, and, uh, and.

What am I even talking about?

Uh, like the, the stable diffusion, uh,
mid journey, Dolly, all of those things.

I find those are less relevant.

Um, But maybe in like
how they fit into your.

Potentially into a workflow.

And then, um, finally there's.

Uh, what was the last one?

Oh, well like being, being searched.

So Bing has an image generator, uh, but,
uh, Bing search and another one that is,

I assume, built off of, uh, open AI eyes.

Uh, Chad G well, GBT four.


3.5 or whatever.

Um, it's a tool called perplexity,
which I find kind of interesting.

Um, and I guess we can talk a little
bit about Canva and what they've been

doing, uh, with, uh, with these large
language models and the image generation.

Uh, tools.

So, yeah, little loosey goosey.

Uh, but let's start with.

Get hub, copilot.

Uh, and there are other tools like this.

I think, uh, Microsoft has their own,
there are a few others out there.

But essentially tools
that help you write code.

And there was some controversy.

Actually with all of these tools
that are controversy, right?

Uh, With the large language models,
there is controversy surrounding truth.

What is it?

And how is this going
to flood the zone with.

Things that are hard to.

Determine whether they're true or not.

And whether people are going to
do the homework required of them.

To look into things that, that are
generated by these, by these models, with

the image generators, there's controversy
around IP assignment and who gets to.

Own the images that are generated,
but also should these models be

trained on, uh, on images from
artists who have not given permission?

There's there's all kinds of stuff there.

With get hub and copilot in particular,
there were people who were talking

about how, you know, they were
seeing snippets of code that looked a

whole lot like snippets of code that
were right out of their code bases.

Um, and that is definitely not ideal.

Um, that's, that's not great.

Um, I think some of the examples I saw
were pretty short snippets of code.

And I do think that if you're given.

A certain context and a certain
language there with, with

shorter pieces of code, right.

If something can be done in five lines.

Or even 10 lines.

There are definitely situations
where there's not going to be many

ways to, to put that together.

So that's, that's a,
that's an awkward one.

Um, I think anything longer
is, you know, not ideal.

Uh, and there should be protections
built into these systems for that.

That being said, I have
been finding that get hub.

Copilot is very helpful for me.

For the most part, I actually would
say it doesn't do a whole lot in

terms of generating longer functions.

I just find that the productivity
improvement I can get from it, just

generating one or two lines at a time.

So if I've got a variable and I've got, or
let's say I've got a list, Of something.

And I let's say, okay, I'll take an
example from our project that you restack.

And there's a bunch of components
as part of a chewy stock project.


When we're building the dependency
graph for these different

components and how they interact.

There's a point in the process
where I want to just get the

infrastructure components.

And so if I already have a function where
I've said, here are the components, I've

loaded them using this other function.

And I start typing.

Uh, the name of a variable
called infrastructure components.

I find that.

Um, co-pilot will write
the right code to you.

Usually just a single line or maybe
three lines to say, all right, let's loop

through all of these components and let's
filter by type equals infrastructure.

And it's, it's the sort of thing.

That's just a really, really small.


But when you add that up to like three
seconds or five seconds that it saves me.

When I'm typing that out.

If I do that, I don't know, a
hundred times and a project.

Uh, in a session.

Um, That actually adds up
to quite a lot of time.

500 seconds is a lot.

And I do find that it's every, you
know, every few seconds it'll, it'll

autocomplete, something like that.

Um, so I think there's
a lot of value there.


The next one that I want to talk about.

So that's code and how you can just
build products a little bit faster

using AI tools, like get hub copilot.

I think one of the other
ones out there is a.

Microsoft in Telecode.

I believe.

Um, they have had a product named
IntelliSense for a long time, which

is just their normal code completion.

Uh, but in tele code, I think is this,
uh, uh, predictive, predictive model,

um, So, yeah, I haven't tried it, but
it might be neat to give it a shot.

The next thing I want to talk
about is Chad GBT itself.

Um, I have found actually what I
will likely do with this, uh, podcast

episode when I'm done is I'm going to
take the transcript of the episode.

I'm going to go to Chad GPT.

I now pay for GPT for, um, and.

What I will do is I will tell
Chad GBT, here's the transcript.

Can you clean this up into a blog,
post maintaining a bit of the tone.

And structure, but turn it into something
a bit more clean, few paragraphs.

And I will use that as a base to
sort of expand on, on what I'm

talking about now and turn it
into a, into a, into a blog post.

And I think there's a lot of value there.

In terms of using these large language
models, not to generate anything novel.

I don't think that is their role.

But to take language that you
have, or that exists out in the

world and doing something with it.

So if I feed in a longer chunk of text and
I want to get something out of it, I think

that works out generally pretty decently.

You still have to double-check.

Uh, so in this case, the transcription
and I want to get a blog post.

Um, I find it works well,
not for summarizing, but for

reorganizing and cleaning things up.

Um, so with, with the
transcription it'll, it'll.

I can tell it.

This is a transcription
with transcription errors.

And basically it'll clean up that text.

Um, and it will format it to some
extent, uh, the way that I might

direct it to, uh, which is practical.

It usually takes a few tries.

But I think it's still a lot faster than
me just writing the whole thing from

scratch, especially since I've already
spoken it out in this podcast episode.

So that was number two.

Uh, Chad GBT.

Uh, I find that the 20 bucks a
month is worth it for me, uh,

especially for this kind of thing,
turning a podcast into a blog post.

Or I've had a few other
situations, um, that I find useful.

Oh, actually as a developer myself,
one of the things that I find useful as

well is using it to, uh, remember the
format of a, uh, command in the terminal.

So if I want to use a tool like our
sync, which will, uh, sync files

from one directory to another,
or from one server to another.

And I can't quite remember the flags, for
example, to make it do certain things.

For those of you who aren't, uh,
super technical, um, It will, uh,

basically spit out a command for me.

That will do the things I needed to, and
I can give it sort of a natural language.

So I will say, Hey, what's the arson
command to, uh, Sync this directory

with, uh, this directory on our
remote server with this IP address.

And I want to, uh, make sure that
everything is archived and small.

And I don't know, something like that.

And it'll spit out the right,
the right, uh, command.


That's another neat use of Chad GPT and
there's a bunch of small ways to do that.

I find it's good for, for
code and terminal commands.

Um, if you just need to spit out
little, little chunks that are useful.

The next one, the image generation stuff.

I have had almost zero use for this.

I think it's more of like a curiosity.

Um, I've heard of people who use
those tools like a stapled, a fusion,

uh, and, uh, Dolly and mid journey
to, uh, generate content, to put

into their marketing materials.

I feel like our brand doesn't really
match with that and I still feel.

A bit more weird about the, uh, The art.

I don't know, there's, there's a lot
of stuff going on with artists, uh,

who are very unhappy about it and like,

I kind of see where they're coming from.

It's weird.

Cause I don't like I put a lot of
code up on GitHub, open source and.

From my context, doing that, I'm
perfectly happy for these models to

be trained on my code and my writing.

I have a fair bit of writing out there on
the internet, and that doesn't bother me.

Um, I have some art out there online.

And frankly, it doesn't bother
me that it's trained on my art.

But for someone who's living
is made off of their art.

Uh, and who is perhaps invested a lot
more time into refining their craft.

I can understand where
they're coming from.

Uh, so.

Yeah, anyway, that's that?

I do know some people who do get
a fair bit of value out of, uh,

the image generation tools for
marketing materials, as you know,

the header of their newsletter or
something like that, they want.

Uh, they're talking about something that
will speed up your website and they want.

Uh, computer running.

With shoes on fire or something like
that, something silly and the AI

spits it out and it's kind of fun.

Um, and yeah, it doesn't
need to be perfect.

It's not something that they
would've paid someone for anyways.

So it's just kind of a neat, neat,
uh, additional value for them.

Uh, so that's.


That's that?

And the, oh, actually on the subject
of the image generation tools.

If you haven't used Canva before,
I think Canva is an amazing tool.

I feel like I may have talked
about it in a previous episode.

Um, if you want to do fairly simple
designs for social media posts.

Uh, presentations, documents, uh,
short videos, that sort of thing.

And you want access to a bunch
of royalty free content built

directly into your editor.

And you want to have your branding
materials available, like fonts, colors.

Uh, graphics, other things
that, that use to make your.

Uh, collateral recognizable as your own.

Uh, Canva is great for that.

And one of the things that they added
recently is a bunch of tools built on

both large language models and these,
uh, image generation, uh, tools.

And so now within Canva, you can.

Generate an image.

Uh, that you might want to use
as like the background for,

uh, for a social media post.

Uh, and so you could tell it, Hey,
I want a wavy green fields with

a Hawk flying through the sky.

Or something like that.

And, uh, yeah.

Have it generate that for your background
and throw in some texts and whatever.

Um, so that's kind of neat.

I've tried it once.

It's a little slow as all of these
things are and the quality is.

It's okay.

Um, They also added in some,
uh, tools to generate text.

Uh, directly in there.

I find that a little awkward again,
for me, the primary value that

comes out of these large language
models is not in generating text.

It's in restructuring.

Texts that I have generated.

Um, again, I don't think the
large language models are.

They don't create anything truly novel.

Um, Well, that's not exactly true.

They don't come up with novel ideas.

So they'll come up with novel
texts, but you still have to give.

Them something to work with.

And so if you're trying to
generate text for a, uh, for

marketing material, I don't know.

I feel like you have to have something
pretty good in there to start with.

And it just feels like a
weird, weird place for me.

Um, to, to be doing that in Canva.

But yeah, it is, it is neat.

They have this, uh, this new,
uh, tool that I haven't tried

yet where you're supposed to be
able to describe a presentation.

And it will spit out the presentation.

Um, with your brand or,
uh, and, and your content.

So I am curious to give that a shot,
especially if you can pump in enough

content to say, Hey, I want to do a
presentation on all of these things.

Here's like my.

Intro the stuff that I want to
go over and like a conclusion.

Can you put that together in
a nice cohesive presentation?

For me, that would be kind of neat.

Um, So, yeah, we'll see if
that's actually any good.

Um, and the last thing I wanted to talk
about is the search, uh, and sort of

finding information in sort of a coherent.

Natural language format.


It's weird because I don't
like my instinct was that this

is not something that large
language models would be good at.

And that I wouldn't find much value in it.

But, uh, I have been sort of proven wrong.

Uh, to some extent, uh, especially.

Framing this again, in the context of
building a small, small tech product.


I think.

If you're operating business, there is
some value in being able to quickly parse.

Uh, websites information, whether
you're looking for grants.

Uh, you, uh, are trying to find people
to connect with, uh, potential clients.

I don't know.

Uh, users for your app, and you're
trying to figure out how, how

all of your, your things connect.

Perplexity, I think has been kind of neat.

Uh, it's a tool that is structured
a bit like a search engine, but

it provides a bit more context.

Um, so it's very much the same value
prop as, uh, Bing's a new chat interface.

Which is, you can ask it a
question in natural language and

it will do some searching for you.

Uh, some more traditional search
pull out pages that contain the

information that you want and sort
of summarize and spit that out.

In a.

Uh, natural language format.

Now, one of the things I find
really annoying about being is

you have to use Microsoft edge
browser, which I do not want to use.

Uh, so I will not be using being.

Which is I think a weird move
on Microsoft's part because

I could absolutely see myself
using, being more often.

If it weren't tied to the edge
browser, but I use arc, which is a

wonderful chromium based browser.

And I don't want to move away from it.

It's lovely.

So no being for me.

But perplexity gives me
something very, very similar.

It will pull out, uh, websites
that it uses as, uh, sources.

And so it'll tell you basically where the
information came from, that it summarizes.

But the other thing that I really
like about it is it's got a Chrome

extension that you can use to sort of
summarize or ask questions about a page.

So if you're browsing.

And you come across, uh, for example,
a very long grant application document.

You can ask it things about that document.

You can ask it to summarize and,
uh, figure out what's what's

in there that might be notable.

Um, and that's something that.

I find valuable, even if it's.


The value that I get out of it is, is.

I'd say significant.

Um, yeah.

So those are the AI tools that I've
used that I've poked around that.

I would love to hear more about
what, what you all are using.

Uh, it's, uh, it's it feels like quite
the explosion of AI tools at the moment.

I guess it's been going for a little
while, but especially since chat GBT.

Uh, came, came on the scene.

Uh, things have.

I'd say very much.

Exploded and I'm so, so
curious to see where they go.

I think there's a.

There's a hype cycle.

That happens with any new sort of
technology as it, it captures the

public interest, uh, particularly
the consumer public interest.

But there's something about these
AI tools that strikes me as a little

different than some of the hype that
we've seen over the past five years or so.

A lot of the technologies that I've
seen hyped over the past five years.

I couldn't find any
directive value in them.

Uh, so whether that's a crypto
V R a R, I know some people got

a lot of value out of those.

To me, I'm seeing a lot more people
get value out of the new AI tools.

Very quickly.

Including myself.

And maybe that's just my own
context, but there's something

there that I find very exciting.

Um, And so I'm curious to see where
it, where it lands when the hype

cycle dies down and we're left
with the actual evaluable stuff.

So, yeah.

Anyway, if you want to keep up with
this stuff about building small

tech, then make sure to subscribe.

If you're checking this out on YouTube,
make sure to hit that like button.

I have been your host wrath, and we
all want to do some good in the world.

So go out there and build something.

Good friends.