Using CHATGPT offline? How to use AI in B2B SaaS and lots more with Kyle Legare- Account Exec
Summary
TLDRIn this episode of 'Unsupervised Learning', host Renee interviews Carl Leir, a B2B sales professional who explores the intersection of AI with his non-technical background. They delve into Carl's personal projects, such as creating a chatbot for sales calls and a Santa phone service for his nephews, showcasing the accessibility of AI technology. Carl shares his experiences with local language models and the potential of AI in home automation, emphasizing the importance of keeping data local. The conversation also touches on the broader AI community, the misconceptions about AI capabilities, and the impact of AI on various industries, including Carl's work in the financial sector.
Takeaways
- đ€ The guest, Carl Leir, discusses the use of AI in home automation and making technology more accessible.
- đĄ Carl shares his personal projects, such as creating a chatbot for sales calls and a Santa phone number for his nephews.
- đ± Carl's interest in AI was sparked by the release of Chat GPT and his desire to improve his coding skills.
- đ Carl emphasizes the importance of local models and data privacy, particularly in the context of home automation.
- đ Carl's experience with 11 Labs and the challenges of creating accurate voice models, such as an Australian accent for Santa.
- đ ïž Carl's exploration of low-code/no-code interfaces like Jan and LM Studio, which make AI more accessible to non-technical users.
- đ€ Carl's involvement in the AI community, particularly through the Vancouver AI Meetup, and his observations on the intersection of technical and non-technical skills.
- đ Carl's perspective on the current state of AI, suggesting that people often overestimate or underestimate AI capabilities.
- đ Carl's excitement about the potential of AI agents and the future of generative AI in specific use cases.
- đ Carl's work in the financial sector, focusing on improving customer service through AI without generating responses, to mitigate risks.
- đą Carl's thoughts on the evolving public perception of AI, influenced by widespread exposure through platforms like Chat GPT.
Q & A
What is the main topic of discussion in this podcast episode?
-The main topic of discussion is the field of B2B, home automation, making technology more accessible, and the use of AI tools like chat GPT for non-technical people.
How did Carl Leir use AI in his sales role?
-Carl Leir used AI to improve his performance on sales calls by processing his speech through a prompt and receiving relevant feedback to react better during the call.
What was the purpose behind Carl's creation of a Santa phone number for his nephews?
-The purpose was to create a fun and interactive experience for his nephews where they could call a phone number and have a conversation with a voice-cloned Santa Claus.
What challenges did Carl face while trying to replicate an Australian accent for Santa using 11 Labs?
-Carl faced challenges in getting the Australian accent right, as it ended up sounding more like The Grudge. He had to find YouTube videos of English-style Santa and strip out background noises to get the voice right.
How did Carl get started with local language models (LLMs)?
-Carl got started with LLMs after the release of chat GPT, which sparked his interest in using AI for coding and other projects without needing extensive technical knowledge.
What is the significance of Jan for Carl's AI projects?
-Jan is significant because it allows Carl to run models locally, making it easier for him to work with AI without relying on internet connectivity and providing a more accessible way to experiment with AI models.
What is Carl's vision for home automation and AI?
-Carl's vision is to have a smart home system that operates locally, similar to the Star Trek computer, where he can dump personal data and have AI models running within his house to perform various tasks.
How does Carl perceive the current state of AI in relation to its capabilities and public perception?
-Carl believes that people often overestimate AI's current capabilities while underestimating its potential impact. He sees AI as a powerful tool but also acknowledges that it's not yet perfect and has limitations.
What are Carl's thoughts on the future of AI in sales and customer service?
-Carl sees AI as a tool that can help sales and customer service agents by handling repetitive tasks, allowing them to focus on core interactions with customers and providing better service.
How does Carl stay updated on AI news and developments?
-Carl stays updated on AI news through Reddit, Twitter, newsletters, and attending AI meetups where he can learn about creative and innovative uses of AI.
Outlines
đŁïž Podcast Introduction and Guest Discussion
Renee introduces the podcast episode, focusing on unsupervised learning and open source technology. She speaks with Carl Leir, discussing B2B, home automation, making technology accessible, and local model running. They also touch on the use of AI in coding and non-technical individuals setting up systems like chatbots.
đ§ AI in Sales and Personal Projects
The conversation delves into how AI can assist in sales calls and personal projects. Renee shares her experiences with voice recognition and creating a Santa Caller for her nephews. The discussion also includes the challenges of working with accents and the limitations of current AI technology.
đ€ Exploring AI and Local Models
The guest shares his interest in low-code or no-code interfaces and his exploration of AI tools like Hugging Face and Jan. They discuss the potential of running AI models locally for home automation and the appeal of having a smart, offline home system, inspired by Star Trek.
đ Involvement in the AI Community
The guest talks about his non-technical background and how he got involved in the AI community, particularly through the Langchain framework. They discuss the formation of a Vancouver AI Meetup group and the diverse range of people and projects involved, highlighting the crossover between technical and non-technical individuals.
đĄ AI's Role in Communication and Fine-Tuning
The conversation explores the impact of AI on communication skills and the fine-tuning of AI models. They discuss the challenges of prompting AI effectively and the evolving nature of AI interfaces. The guest also shares his thoughts on the potential of AI agents and the excitement around their development.
đ AI in Banking and Client Adoption
The guest talks about his work in the banking industry, using AI to improve customer service through virtual assistants. They discuss the importance of not generating answers in the financial sector and the use of AI to assist customer service agents with repetitive tasks. The conversation also touches on the changing perceptions of AI in the industry.
đ Future of AI and Keeping Up with Developments
The guest shares his thoughts on the future of AI, particularly in the context of virtual assistants and the importance of making AI tools accessible and user-friendly. They discuss the challenges of keeping up with the rapid developments in the AI field and the various ways they stay informed, including Reddit, Twitter, and AI meetups.
đ€ Closing Remarks and Contact Information
Renee wraps up the episode, thanking the guest for their insights and contributions. She invites listeners to reach out with questions or feedback and provides information on how to connect with the guest and the podcast.
Mindmap
Keywords
đĄUnsupervised Learning
đĄB2B
đĄHome Automation
đĄChat GPT
đĄLocal Models
đĄLow Code/No Code
đĄHugging Face
đĄLang Chain
đĄAI Community
đĄFine-Tuning
đĄAI Misconceptions
Highlights
Renee interviews Carl Leir, discussing B2B, home automation, and making tech more accessible.
Carl shares his experience with using Chat GPT for coding and non-technical applications.
The conversation delves into the challenges of setting up home automation for non-technical individuals.
Carl's creative project of creating a phone number for his nephews to call Santa Claus using a cloned voice.
The difficulties Carl faced with replicating Australian accents for his voice projects.
Carl's interest in local running models and the potential for personal data privacy.
The discussion touches on the ease of using Jan, an open-source tool, for local model running.
Carl's perspective on the balance between low-code/no-code interfaces and traditional coding environments.
The importance of local AI for home automation and the desire for a Star Trek-like computer experience.
Carl's involvement in the AI community in Vancouver and the formation of the AI Meetup group.
The observation that AI can help both technical and non-technical individuals improve their skills.
Carl's thoughts on the current state of AI, including its capabilities and limitations.
The potential of AI agents and the excitement around their development.
Carl's work with Gia, a tool that helps financial institutions communicate with their customers.
The challenges of adopting AI in conservative industries like banking and credit unions.
The impact of Chat GPT on public perception and understanding of AI.
Carl's preference for AI tools that make life easier rather than just automating tasks.
The importance of keeping up with AI news and developments through various channels like Reddit, Twitter, and newsletters.
Carl's recommendation for resources to stay informed about AI, including specific newsletters and communities.
Transcripts
Renee here at unsupervised learning your
easy listening podcast for bleeding edge
open source Tech this episode saw me
speaking to an awesome guest in the
field of B2B Carl leir we spoke about
home automation making Tech more
accessible keeping models locally
running and how you even go about
setting that up as a non-technical
person it's one you'll enjoy let's get
into it when you're using chat GPT to
code or when you're starting like coding
courses like what what are you making
the yeah I don't even remember what I
would have been making back in the day
or why just I think it was an area of
interest but like with Chad gbt like
some of the stuff I was focusing on
was like one of the things I made was I
got really interested in voice like
having
it hear me and then like feed stuff back
and so back to that comment when you
said a and maybe for the people
listening a is like account executive
meaning in sales so it was like how can
this help me be better on a sales call
and that was like the big idea I was
trying to do was as I talk can you chunk
it out run it through a prompt and then
feedback something relevant so so I can
react better on a call so I kind of was
tinking around with that of course in
this space there's like a dozen
companies that are doing this now but
that's just the way it goes and then one
of the other fun ones was around
Christmas was I built a a phone number
where my nephews could call Santa
basically or have Santa call them so it
has like a clone Santa voice that I used
like with 11 labs and it was set up
Santa prompt style so they could just
dial Santa and talk to Santa basically
so I thought that was a really cute idea
from the nephews really really cute yeah
but 11 Labs is really interesting to me
like I messed around with it but the
Australian accents are really really
tricky to get so it ended up sounding a
bit like The Grudge like I tried to do
like a podcast intro but it was like the
uh the voice thing was odd like I I
don't know when you think of Santa's
voice like what do you think does he
have an Australian accent or does he
have kind of a okay he's English for
sure he's English she like hello I'm s
wow that was bad but yeah yes okay I
agree that's exactly how I think about
Santa but it was hard to get the voice
so I'd like find YouTube videos of in
English style Santa talking and then you
got to like strip out the because all
the videos of s online are like there's
like elf noises in the background and
workshop noises so you have to like
those out and there's like some software
that helped but there however that
technology works there was just like
weirdness with the voice at some point
so sometimes when the kids are talking
to Santa Santa's like it just goes
really awkward and then goes back to
normal but the kids didn't really call
it out which I was thankful for but it's
certainly weird if you talk to Santa
long enough I found you through a post
about Jan right someone pinged me and
said hey this person's doing like
they've they've spoken about local llms
I was like why and then I looked at your
account and you're you're an AE and I
stereotyped you really hard because I
was like what's he doing what's he doing
with software yes I know what's this
Tech bro doing should G be on a sales
call so how did you get into
llms well I guess like yeah into LMS in
general so I think I don't know what the
date like November 30th I think when
chat GPT came out and you know I think I
saw like on Twitter like people were
just like sharing some little stuff and
I think like our first interaction is
always
like you know sing me a song well
Pirates like something really stupid
right and then you're kind of like wow
this is really neat and then I asked it
a question on HTML and it like spit out
formatted code and I was like like I
just feel like the gears start spinning
in my head because I Bally like my the
history of my online education with like
coding is littered with starting and
stopping uh coding courses like you get
to like you set up your environment
valuables or whatever and you finally
get to say hello world and then like
well that's probably it you never come
back to it and then you pick it up later
this was like a whole new thing help me
like Leap Frog what I thought was
probably would have taken years of
actual learning to just like test out
ideas or Tinker really I'm thinking like
gong and civil have you tried civil AI
I've tried C was it real time that
you're looking at well that yeah so that
I mean that's really like the trickiness
you run into what so like what I had
built out and it was a really cud
prototype but it was like almost like a
information window that's like if you're
talking and I'm talking it's trying to
like surface relevant chunks there is a
inherent difficulty in understanding how
you chunk people talking is it like is
it at is it at a pause can you look for
like a natural like period in point so
it's it's not perfect and it was quite
hard and there's smart people that are
working on this but that was just an
idea of stuff that the Delta between me
having that idea and then actually being
able to have something you could kind of
show to people prior to like chat gbt
was got you know like I don't know I
still wouldn't say I know how to code
but I know how to like massage the llms
to get stuff to work for me so it's
stuff like that and there's lots of
silly ideas like the Santa caller as one
example so yeah I think I was gonna TI
that up I think it's just like I have a
lot of ideas some dumb some bad some
good but either way it just allows you
to kind of iterate and test stuff out
without you know some hacky way to get
it or trying to convince someone who
knows how to code why your ideas is
worth executing and oftentimes they
aren't so that that's like one of my
main interests so far yeah it's a lot of
the I would say like low code no code or
the interfaces like have you tried like
super agent or those those interfaces
that the uis that aren't just vs code
because vs code for me is very it's like
a learning curve but if I see something
with a with a interface that's very like
a like a chatbot like Jan or like um
LM Studio or something that's like not
it's not as Technical and so are you
you're current kind of not workflow but
like in terms of tinkering and stuff do
you head to hugging face for a specific
use case or are you what are you doing
well yeah so that was I mean that's how
Jan I guess came on my radar because all
the work before was just with like open
AI API and it was just fine it's
accessible they had very good
documentation but then I was always like
watching huging face there's always like
cool stuff going on right there's like
interest models and the but there's
always like a technical limitation
on like my computer like I just have a a
Macbook I think it's like an M2 which
actually surprisingly now aren't bad for
a lot of this stuff but you know a few
months ago even there wasn't a lot of
ways to run some of these models so this
is something I just looked at from the
sidelines and
then a random Reddit post talked about
Jan and I was like oh very cool so
download that and I was like yes this is
ex exactly it right it's an easy app you
can just pick and load your models so I
guess my interest is maybe it wasn't
just open source specific but it was
just like cool new stuff and a lot of
the models are these open source models
are the cool new stuff but from that
there's like two things that I got very
excited about so the first I was I fly a
lot for work and I was on a plane I had
work to do and I used these LMS for like
everything all the time just like
bouncing ideas off there's no Wi-Fi on
the plane well guess what Jan's fine
right boot it up the laptop and then you
can just keep going with that that to me
is amazing and then it goes to like this
other area of Interest so I'm a big Star
Trek fan and have you watched Star Trek
yeah yeah okay so you know they just go
they go to like the computer and they'll
just say computer and they ask anything
and it and it'll give an answer
obviously it's very hard to do you know
op I kind of do that but what does it
look like to run that completely locally
within your house and this is like you
could now start having your data
accessible you're not scared to start
dumping your own personal data in I do a
lot of like home automation stuff so
those all run locally on their own and I
hate that some of them the stuff I have
set up has to call out to open AI so
what does it look like where you have
your own llm running in your house to to
do stuff whatever that stuff might be
and that's like what I I'm very
interested about some of these models is
you know your house offline is still
smart it's like the start Treck computer
I think it's like my end goal but that's
why I got really excited Jam because
they have the like you can just start
the server and have the API calls
running to do a lot of this this made
the stuff so easy for someone like me
who has the support you know of an lm2
code but still I would not call myself
technical but the J team making this
like so easy with like stuff that I do
know how to do versus I mean for anybody
who is you know I'll just say like us I
guess it's a complicated world to wait
into to do the littlest stuff there's so
much implicit knowledge that
someone might look and go oh I of course
that's how you do like sometimes when I
first started people like oh here you go
and they just send you a GitHub link
like what am I supposed to do with this
like you're trying to go to the read me
and I'm like I does this on the problem
so I've never been to GitHub before like
before prior to you know that that was
one of my issues was I I hadn't used
GitHub but like in your day today at
work specifically are you considered
technical by like colleagues and peers
or are you kind of like like where do
you sit because I I want to get to your
involvement with the AI community in
it's Vancouver right Vancouver yeah you
got it nice I know a lot of people in
Saskatoon Canada which is apparently
like the middle of nowhere it is indeed
yes that is even for Canada which has a
lot of empty space sason is quite it's
unusual I would say to know no people
there yeah let's just on them yeah
you suck hey I'm from the Prairie these
are my people I get it but you know I
think the people from sasun would say
the same I hope hopefully you don't have
a large uh sasonian listenership but
sorry to my fellow
Canadians So currently that's
self-driven that
experience yeah like I would my RO is
not technical
we like the company itself I guess it's
always like this in software right the
underlying knowledge is obviously very
technical and has a very a lot of very
very smart people working on it but
those aren't the people selling it call
like you know a salesperson's world just
understand the problems it solves not
necessarily a technology so my work my
day-to-day is very
non-technical however everything else is
self-driven so my my interest and my
life has always kind of been like that
I've always been in sales and and
working at SAS companies I guess makes
you a little more technical
than other people actually I don't know
if that's true we just use a lot of
software I don't know maybe that makes
us feel like we're technical but no my
day-to-day it's very untechnical but
it's just all personal Pursuits where
you do this and love to talk about the
ad Community when we get to that but
yeah because that that's been an
interesting world because your initial
reaction like when A's involved in here
that also happens in real life at all
these meetups I'm still an enigma I
think or an anomaly I should say at a
lot of these technical
meetups yeah and so before we get so
that I wanted to talk about like are you
using llms at work or are you using like
you setting your day-to-day like home
automation I get like are you using like
is it NFC is that what they're called
the no you know the little
chips yeah yeah yeah I tried to buy some
chips off Amazon so that I could just
touch it and my Xbox would like connect
to my and yeah again my husband was like
what are you doing I'm like I'm from the
future I'm Judy Jetson leave me alone
yeah there's if you look carefully
there's little stickers actually all
over this enironment for for stuff like
that and so yeah there's that and
there's just like the motion sensors and
and you know one one of the projects was
building a like a Google Home Smart
speaker but that runs off runs locally
so
there's that's the end State and I want
Jan to be powering that essentially
there's just a few hurdles I got to jump
over to get that but the idea being you
can like have your little computer with
a speaker ask it a question say about
your house or ask it to do an Automation
and have that executed like completely
offline so that's like one of the the
end States so um NFC tags are just one
layer to that but it's that's the thing
that I'm trying to do and that's it's
very complicated if you think software
is hard bring in anything to do with
hardware and it's like G to the most
insane level of
complication so there's again a lot of
SM people online to help out
unfortunately but yeah I'm thinking
about different use cases for that in
specific that I've seen in a lot of
Discord communities people trying to
make AI girlfriends like local AI
girlfriends that like kind of like the
movie Her have you seen that yes like
this disembodied voice and I'm like you
know what I think there are some things
that we look back on and we're like oh
that's not such a crazy idea but at the
time it's like what like it's it's easy
for me to punch down and be like oh got
a real girlfriend but like I can see the
appeal kind of but it's not something
that I'm like actively
championing yeah yeah I don't know where
I like the local aspect though because I
don't even know you know you don't want
whatever weirdness you're talking to
about it ever being online so at least
you can just keep that right within the
local network rather than risk anything
yeah yeah I think that's a big draw so
AI Community how did you get into that
is the Lang the langang community
specifically it was yeah so we started
as so I have a few colleagues I guess
well I'll call friends actually some
colleagues depersonalize it so much
they're we're like both tinkering
together I guess and you know one of my
friends my shout out Reed Robinson he
works at zapier on their AI side he's
also not technical but kind of it is so
it's like both of us are just like wow
this is really cool we can do stuff that
we couldn't do before and so just like
us sharing ideas is he through his work
he would speak to Lang chain and like
are you familiar with Lang chain if you
come across it very briefly like I know
that the Jan team is working with Lang
chain and I have just started using Lang
chain tools to try and mess around with
stuff but I I'm a few steps behind you I
think in terms of technical capability
and awareness so yeah and if I'm coming
across is very confident in this case
again it's just the L in fact it's right
in front of me feeding me answers no I'm
just joking um the um well Lang chain is
like it's just a framework to help you
work with like large language models so
like when I talked about before about
building like the voice projects and and
chunking Lang chain like brings in ways
to chunk they help you with embeddings
and vectors in like a very easy way so
they are you know at least at the time I
I don't know the state of it now a lot
of people's projects involve Lang chain
somehow because it just made a lot of
stuff easier to do when you were
building and he got connected with
Harrison the co-founder and then you
started finding out well there's a lot
of people in Vancouver doing this as
well right you know it's happening on
Twitter they're happening across at
other tech companies it's not work stuff
just people having fun I guess or
pushing the limits of these llms at the
time so we thought it'd be fun to have a
Meetup and so the first Meetup was Lang
chain and we met another gentleman who
he had won some competition with Lang
chain and well so three Vancouver people
connected and we just put out this
invite honestly not knowing what to
expect and like gosh within like like a
very small amount of time a few days it
was sold out so I think it was 60
something people we're like so it just
shows you kind of like the the interest
that existed across Vancouver at the
time and that's morph into just an AI
Meetup so while we did focus on Lang
chain we had a lot of people coming out
of the woodwork that were like like guys
with p phds and data science who were
working on like strange biochemistry AI
projects who wanted to share stuff and
they felt limited by a same Lang chain
so now it's just honestly open open meet
up come show cool stuff and meet other
like-minded people but it makes me laugh
your comment cuz like you know
everyone's like what do you do software
developer data scientist all that it's
what do you do like I'm in sales
like so yeah back yeah it is still an
anomaly but just speaks of this new
world this is what I get very excited
about is I can sit in a room with lot of
these people and share stuff and
understand Kno how again with there's so
much more educated and you know have a
different almost way of thinking
but it's like
we're one guy call one guy brought up
something really interesting and I want
to see what you think about this so on
my side I would call myself like a
fairly good communicator but I'm not I'm
not very
technical LMS have like helped me become
more Technical and understand
that and then on the flip side you have
a lot of people who are very technical
but they're they're poor communicators
you know I I don't want to paint a broad
brush for engineers or software
developers but you probably kind of get
where I'm going their their world is a
little different than like a world where
you're engaging with people to people
all day what they called out was like
when you a good prompting is like a good
clear a good clear way to communicate to
get the answers you ultimately want so
it was like well I'm saying it helps me
be technical they're saying it helps me
become better communicators I thought
there was like this interesting two
sides of the coin but in this the middle
is you know how to work with llms which
is still helping us so I don't know if
that sounded like really strange but it
was just like an interesting observation
I had from two very different
personality types but 100% like I think
it's helping
me scale my technical capability but
also I feel like I'm losing a little bit
of my ability to communicate but I think
for for other people they're kind of
realizing they're building all of these
really cool things but they need a way
to be able to bring it was like a I
think a good example for me was like if
you've had the chance to try those gpts
like the GPT Builder it's like you talk
about what you're trying to do and it's
like a different way to build and I
think it's really interesting because
you're like I want it to do this and
then it doesn't do that you're like
actually here's what I want to do it's
like this like back and forth
conversation you're having to build well
essentially an app which it's just
fundamentally different than you know
the very strict well this is what you
called in your function in Python so
this is what it does of course this is
what it does but this is like you're
talking like not not exactly what I
meant so it's like this different way to
communicate where I think sometimes you
know people in I know people who are
often in personto person jobs
are better at communicating I want to
say always good I think we all work with
people who probably could improve their
communication but a lot of people who
are used to very matter of fact
technical speaking it's not the same I
don't know I think I'm spiring out here
as well but it's like it's just
interesting Clarity that's giving me
some clarity because I'm thinking now
about that did you say Za or is
zpo Reed's gonna kill me if he listens
to this I don't know it's like API is
the key in it so it's
like zap that's like the that's where it
comes from so is it so the the interface
with AI and it's like hey give me a
workflow and I'll do it I find myself
struggling with it because I think at
first I was too prescriptive and then I
wasn't prescriptive enough and it's kind
of the same as when you're say you need
to delegate work you need to know what
you're first doing and that's like the
biggest issue that people have so
they're like I'm just going to hold on
to this and I'll just do it all because
I do it right it's like you don't
necessarily do it right I think you just
don't know how to communicate and I'm
talking about myself this is a self-
roast because I have so many things that
I could communicate and then give to
somebody else but I'm like I don't want
to write it down over explain something
to somebody and then like it actually
doesn't work that well because you
you've gone so incredibly descriptive
for it so yeah yeah it's like learning
how to talk to these things it's it's
like a different tool set and it changes
all the time like prompting from six
seven months ago is already different
now for a lot of these but it is a skill
in itself I think for um it's also weird
there's not a lot of things you do in
life where like you ask something and
you get immediately you get to see
immediate feedback on how bad you just
communicated or how well you
communicated right it did something like
was way off on that I really got to
change how I just asked it to do
something so your I think I tried to ask
this in a really non cinct way but are
you
currently just taking the llms and like
using them out of the box or are you do
you know how to find tune or what do you
where's your interest yeah they're
they're out of the box and the fine
tuning like your podcast with the geman
from ons sloth where it was just funny
because I was like
you know they have well one they have
like the adorable little animals on
their website which almost like makes
you think it's going to be easy to
understand and I was like trying to read
them like I'm like this sounds like a
great problem to solving but I have no
idea what that problem actually is
because I I've never actually fine tuned
and I would see if there's a thing I
would like to understand more it's like
what fine tuning can do or like I
understand the process what it means
it's just
like I don't know if in any of my work
so far I've felt like I was disappointed
with what it's doing and thus
necessitate fine-tuning so I guess again
that was also a long way of saying no
they're out of the box is basically what
I use for at best I'll use something
maybe more purpose built that's probably
already been fine tuned so like in Jan
there's a couple good C really it's like
either deep seek or wizard to help with
coding so I've used that so and I guess
in a sense that's a little more purpose
built but I personally haven't
fine-tuned anything yeah neither have I
uh I am experimenting with like AI
agents so I saw that you had a post on
baby AGI and I think I'm testing out
super agent I tested out crew AI which
is there's no interface so that scared
me so it was just vs code and I reached
out to I'm going to butcher his name uh
Jo joah the founder and he we did like
an async interview and I I just I think
it's one of those things you just need
to kind of get in there but I totally
get what you're saying like it's so
difficult as a non-technical person to
read through documentation you're like I
don't know what an end point is
like what do you
mean yeah it's the little stuff I call
it like the implicit knowledge and stuff
that anybody who's like they just know
the stuff but like sometimes I find my
projects all get stuck on the silliest
problem because it's like I don't know I
just didn't set up an environment
variable correctly or something I
understand the file
structure but that's not like the
project it's just like a thing they
probably learned in like programming 101
and but I like kind of skipped all the
this basic stuff so you know maybe
that's a slight against what I'm doing
but regardless I'm still going to keep
doing it the agent thing is really
interesting and I think maybe that sort
of fine-tuning is interesting
because like that baby AGI those are all
just like cool little experiments that
came out of like an yeah like watching
it Chain of Thought
but if you get to see or play with
agents where or even like try to do
something this is where you kind of see
the AI is like it's not good at a lot of
stuff it has to be like a very specific
use case and I think this where fine
tuning probably would come in but a lot
of tasks that you ask it to do and if
there's like maybe a few steps involved
it can like f i want f up at a lot of
different steps there whereas like would
if you just do like the one shot
prompting or your shot it's like these
chains of stuff you can really see where
it breaks down and I I it's like a
prompting thing it's the right data and
and maybe fine tuning but the agent
space I'm it's probably an area that I'm
most excited about
cuz this is like when people think about
AI that's what we're talking about you
know it's out there doing stuff
completely on its own which I find very
exciting it's also what people are
scared about but it's also very exciting
yeah a question that I had for you was
what do you think people get wrong about
AI yeah I would say I have a lot of hot
take maybe but the one I think I find
most interesting is a lot of people I
think are really
overestimating what you know the is
capable
today but conversely I think a lot of
people are underestimating and and when
I say this like there's a lot of people
probably like me who are like super just
bullish on everything and like oh my God
yeah I can do that and then there's like
the really negative pessimistic people
like I was talking to a guy who was he
was just he was starting school to be a
graphic designer and I was kind of like
is that
I I don't know I could be totally way
off base here seems like a rough place
to enter right now as when you're there
and you have all these different career
pass I don't know if that's the right
path because their thing was like no AI
is really bad at like hands and stuff
and I think there's still room for
graphic designers I'm like yeah but like
but I had this conversation three months
ago it's like already I think a problem
that's almost been solved so I think
people are underestimating some of the
impacts but then there's also a really
big school people who are like vastly
overestimating where we are so I I just
I just find it funny because that's like
if I'm in a community everyone is like
so excited about this and then you talk
to like people in the real world and
they just kind of are very pessimistic
or they like it's not that good but I
think people on both sides are wrong
yeah so I think the Enlighten centrism
position I think we have very similar
views on that because there's people
that yeah there's definitely people that
very pessimistic or
I think someone it was Daniel at Jan
said to me that it's like a 9-year-old
who's hallucinating sometimes like a
very precocious nine-year-old but that's
the kind of level that it's at at the
moment it's like you wouldn't trust it
to do your taxes but it knows a lot of
stuff like that's it still knows a lot
and that's like I guess if I can do one
more thing that it's like when people
downplay it a bit and they're like well
it's just like fancy
autocomplete and I'm like well sort of I
guess but also it's the most
mind-blowing autocomplete ever if that's
the analogy you want to I don't know why
that's like a negative thing for them to
say like it's just a fancy autocomplete
I'm like okay sure I guess but have you
used it like God I don't know I think
that there's two different types of
people that say that they I think a way
the reason I say it is to kind of uh not
have people be too worried about it
because I'm generally trying to speak to
an audience who are not technical are
using it like they might have been
exposed to chat gbt
but might not have been don't know you
know the capabilities they think oh it's
a robot it's like no it's literally like
a fancy autocomplete but that being said
it's still like amazing I think there is
definitely like a way to kind of neg it
and be like it's just auto complete it's
like okay bro like you create ai go and
do it if it's so easy yeah that's it's
funny actually it is a good way to
explain it to like a new person but like
i' I've come across where people are
like dismissing it or like they'll so
like my the company I'm at now I used to
work at a company called fin they got
acquired by this company but our our
whole sole product was a chapot like a
virtual assistant for banks and credit
unions and people always like like well
it's not really AI it's machine learning
I was like I guess I just like don't
know what you're dismiss I don't know
it's just like that's just people's
behavior but they'll always like
like that or they're like is it a real
Ai and I'm like well philosophically
speaking what are we talking about here
when you say this like is it a
completely artificial intellig you know
like the topic you're like well are we
getting philosophy here or is it yeah
you're right actually it is machine
learning we's just move on from the
topic so it's that I don't people it's
just like a it's just human behavior
some people just knee-jerk poo poo stuff
and the autocomplete one just always
gives me a laugh I'm like okay you know
maybe it is in a sense in a verye
elementary sense but like my God it's
the craziest auto complete ever you can
ask it seven different questions in
different order referencing back and
it'll just like answer them all for you
like my AO complete still does ducking
so I don't know if this is like the
great analogy
but you're making me think about the
funny things that my older complete has
said very recently you just mentioned
about Banking and finance and I want to
speak about is Gia Gia you say yeah
you're working currently yeah correct
yeah well we like at its core G is just
a tool that helps like there's credit
unions in Australia right I believe so
yeah sort of mostly just Banks just I
was gonna make a jerk yeah I was gonna
be like well around my
Hut Canada's like there's like five
banks in Canada and a bunch of really
training to Credit Union but I I work
with the US market where they have
thousands of banks thousands of credit
units but um our TR just helps customers
or members because they're called
members when you're cedit Union like
more easily communicate with their
financial institution through digital so
it's could be video chat you know just
regular chat or audio just like we're
speaking through the computer so that's
like the original platform they acquired
the company I was at before so they have
a virtual assistant now so that it's
like that's not really generative AI
because like obviously one of the
biggest scariest risks in financial
services is like something generating
answers right and I know there's like a
lot of good ways to reduce
hallucinations but you know even if
there's a
99.9% chance the answer is always fine
that 0.1% is a very scary to a bank so
our tool actually like has the natural
language understanding so if you ask it
questions it'll pull from a preset
answer so it can't generate an answer
that's usually what banks want but on
the flip side there's a lot of like
great stuff we're doing with generative
ey on like the agent side so helping
them get contextual information and the
conversations they're having and just
making their life easier so it's like
one of that stuff like that I think is
really interesting because there's been
a lot of tools where they sandwich
generative Ai and it's like so useless
right it's like I don't know I don't
even have an example off the top of my
head but it's like you know that little
star icon that everything has now and
click it and it's like it was probably a
press release and that was it but on we
glaz on a lot of stuff where just making
the agents so someone that works in
customer service make like the
repetitive stuff at their job not part
of the job so they can just focus on
talking to the customer is like maybe
the core thing that we're trying to
deliver you know similar to like gong
actually you mentioned gong so like gong
records my calls it does summaries and
helps me with notes without me doing
anything so I can just focus on selling
I think that right now that's like the
lwh hanging fruit for a lot of this and
I hope companies keep adopting that
mindset rather than like generate an AI
post or something like LinkedIn has one
that I think is so dumb it's like it's
like useless stuff like that so but I
like it's like make make my life easier
don't just like make it rewrite
something or come up with it I know that
was a bit of a rant so no right that's
fine rant are fine are you on client
side are you like client adoption of Gia
would you get a lot of objections I
guess I'm saying like about AI
specifically yeah we I would say less so
actually so it's like funny that the
company has that before like you know is
there I don't know what what year is it
now but it was like maybe 2019 when I
started or something it was like 19 and
there were chat Bots obviously around
there was lots they just like they're
all so bad and like they're so so bad or
they took extraordinary amounts of work
to even get good like Google dialogue
flow it's like a bot Builder but it
takes just tons of work to even make it
halfway decent the one other company I
was at was actually very good only in
the context of Banks and Credit Unions
because that was the other thing like
this general purpose knowledge that we
have access to now that wasn't a thing
like you had to have very vertical
specific tools because that was the only
way to have it good again technologist
wasn't quite there so you had it was a
very hard conversation people didn't
know about it they had a really bad
taste with AI so it was like honestly
quite hard and Banks and credit unions
are a very conservative industry so it
was not the easiest thing to pitch so
your objections would yeah like the
objection might be like I just have no
idea what you're talking about so it was
kind of like that and you know honestly
since chat gbt came out life's been a
lot easier because people understand it
they get it they have exposure to it and
sometimes the objections are well I
don't want this thing generating answers
we're scared of that you're like perfect
it doesn't do that so it's been a great
like upleveling collectively of
knowledge that the tool is less scary I
guess the the world of AI is less scary
because you know PE all PE most people
not all have been able to use chat GPT
in some capacity which just been like
free Marketing in a sense because when I
talk about it they they get the general
idea rather than you know this scary
chatot thing that they either think
is going to be crappy because that's the
biggest objection is like well I called
you know CVS in the states it's like a a
pharmacy thing and it's got the worst
virtual assistant it just like it
doesn't want you to get to a human
that's the problem with it so just like
railroads you doesn't give you helpful
answers so that's like been almost
anti-marketing because that's people's
bad exposure to it we're in a very
different world now with just people
having so much more knowledge it's like
splattered all over the news and
everywhere we hear about it whereas
before it was like this weird little
thing like as an example here's a funny
story I was at a conference and we had a
booth and so we're like it was all just
Bankers so again assume like maybe tech
people knew about chap Bots had a better
understanding but go to a very
conservative industry like banking where
they stuff just hasn't even hit their
radar and I had a big banner behind me
that said like AI powered virtual
assistant and like the lady comes up she
like puts her glasses she's like who's
Al and I was like I was like what I'm
like what she's like Al powered because
it's just like a with a capital I and I
was like what I'm like oh no it
doesn't it's not even worth explaining
anything here so oh Al yeah he's not
here right now he's the guy that powers
the conversation when you're chatting
with it but yeah yeah sorry sorry about
that that was like the old world I think
and now it's less weird to see AI
powered
something in fact we probably too much
now i rate a lot yeah I'm just thinking
about my friends in like marketing
product marketing specifically that have
like a vendetta against AI powered I'm
like yeah that's fair it is the Super
Bowl today tonight how long does the
Super Bowl go for I about three hours
roughly I know what kind of extra
commercials we'll have slotted in but do
people like the commercials like is that
Well here here's the thing so my
American friends are probably watching
some really great commercials that we'll
all see on Twitter later Canada doesn't
get those we get hyper Regional
commercials so in fact just before I
hopped on like the king of floring from
a small town near here was advertising
their floring sale with the with the
Super Bowl in mind I always ask everyone
if you could have invented anything what
would it be like a thing that's already
invented and I wish I came up with the
idea or oh okay you can pick two so you
can pick the intellectual property
rights for something and then also you
can pick just having invented
something okay well I'm going to go to
an ebook reader and why I picked this is
hilariously I found my mom shared like a
do you know what a duang is do they have
those in Australia just like a folder I
don't know if it's just a Canadian term
just a place to hold your paper it's got
little bendy metal Clips in it the three
holes we don't have any paper over here
so yeah just like writing in the dust on
the sand kind of thing yeah know usually
it's slate
okay good all right well it's a place to
keep your little slates that you have I
guess but I had this whole mockup for
like a little
e-reader with like little cartridges and
this would have been like you know like
early early 90s so I was like wow that
was pretty good
idea again no technical knowledge to
have done it so I I'll just pick eBook
reader for this one I think I had the
idea first yeah we got to go find them
and fight them clearly who would you
interview in the AI
space that yeah I would see the the guy
we're talking about before like Jane
Harrison Chase only like he just so
incredibly helpful because I think when
you know chat gbg hit there's the open
Ai apis and stuff but if you ever get to
building stuff there's a lot of like
other stuff to make it work like you
know vectors and embeddings like how to
take a big chunk of text and make that
searchable by my AI or context length
Windows right you can't just puke in
10,000 words you have to break up stuff
stuff how do you easily break something
up the the tools exist but you again
have to go find out which tool it is how
it works how do I make those API calls
like Lang Jan just made it so you could
do all that stuff with like their little
recipe books they call their cookbooks
to do that and it's just like I have a
very soft spot in my heart for that
because it was just so tremendously
helpful because I don't like when you
start a project you like oh I have such
a good idea oh I didn't think about that
I didn't think about that but L already
thought of all that stuff so I just got
to like follow along and that's always
like kind of at The Cutting Edge for
where a lot of this is going and again
shout out to the open source right so
they're doing a lot of really great
accessible stuff that leads me to my
last question which is like where do you
keep up with everything with like the AI
news and like do you have specific
places like you said your you're
redditor yeah Reddit Twitter and then a
couple newsletters so Ben bites uh is
one yeah he's quite famous now on
supervis learning is in the repertoire
here thank you plug for the show but
that's it yeah Twitter is probably the
biggest one in Reddit and then just you
know the AI meetups are very helpful
because you get get a lot of people that
are doing weird or creative stuff I
guess I'll say and it's like I'm sure
you feel the same way it's an industry
that's literally impossible to keep up
with even like the major developments
it's insane and like a friend I was
killing myself just trying
to I didn't want to miss anything now
it's just like you know what I'm happy
stuff's moving along and you know
try yeah the fomo is very real I don't
think people sleep very much in the
space but where can I will put it in the
show notes but where are you most
reachable for people that want to chat
to you definitely LinkedIn sorry your
the Vancouver AI Meetup is it just
called yeah well we we sh it up a it
it's the van AI
Meetup oh that's right and people show
up with Van and like
color yeah no no van wids coming up yet
offering you know free free llm tokens
on the side out of the side of their van
yeah it's probably LinkedIn is probably
where I'm most active you know that
definitely speaks back to the sales part
of me thank you very much for yeah being
on the show and having such great
insights and conversation that is so
sarcastic it's really not but oh thanks
I thought that was okay yeah thank you
so much I love what you're doing here
and I feel I mean you had a couple like
real you know heavier hitters already
and now you're coming to the sales
people so we appreciate you thinking of
us in this llm world that wraps up this
week's episode of unsupervised learning
I'm your host Renee and I've had a great
time chatting with you as always links
to everything we discussed will be in
the show notes make sure you reach out
to our guests questions or feedback
reach out to pod unsupervised learning.
until then leave a like follow or rating
on Spotify Apple podcast or YouTube and
until next week stay curious
oh
Voir Plus de Vidéos Connexes
Innovation in Presales
Episode 1- Efficient LLM training with Unsloth.ai Co-Founder
"I tested 400 AI Agents, these are the best" - Adam Silverman
"Who Will Make Better Music - People Or AI?" will.i.am vs Piers Morgan
Questa AI fa il mestiere di 700 persone, robotica che cambia la vita, l'AI arriva nelle tue orecchie
This $50M/Yr Side Hustle Is On Track To Make $1 Billion By 2030
5.0 / 5 (0 votes)