AI Hustle: News on Open AI, ChatGPT, Midjourney, NVIDIA, Anthropic, Open Source LLMs: Rabbit Challenges Adept with $20M AI Software Model

Welcome to the OpenAI podcast, the podcast that opens up the world of AI in a quick and

concise manner.

Tune in daily to hear the latest news and breakthroughs in the rapidly evolving world

of artificial intelligence.

If you've been following the podcast for a while, you'll know that over the last six

months I've been working on a stealth AI startup.

Of the hundreds of projects I've covered, this is the one that I believe has the greatest

potential, so today I'm excited to announce AIBOX.

AIBOX is a no-code AI app building platform paired with the App Store for AI that lets

you monetize your AI tools.

The platform lets you build apps by linking together AI models like chatGPT, mid-journey,

and 11Labs.

Eventually, we'll integrate with software like Gmail, Trello, and Salesforce so you

can use AI to automate every function in your organization.

To get notified when we launch and be one of the first to build on the platform, you

can join the wait list at AIBOX.AI, the link is in the show notes.

We are currently raising a seed round of funding.

If you're an investor that is focused on disruptive tech, I'd love to tell you more

about the platform.

You can reach out to me at jaden at AIBOX.AI, I'll leave that email in the show notes.

Welcome to the AIChat podcast, I'm your host, Jayden Schaefer.

Today on the podcast, we have the pleasure of being joined by Joshua Panner, who is

the CEO of Inquisio.AI, which is a firm focused on optimizing government efficiency

through AI.

He has a background in both public service as the mayor of Orting, Washington, and the

private sector.

He has a lot of leadership background there.

He uniquely bridges kind of that gap between technology and governance.

His strategic expertise has driven scalable success in both of these areas.

Welcome to the show today, Joshua.

Hey, thank you so much for having me.

The thing I would super excited, of course, to have you on the show today, the thing I

would love to kick this off with is I'd love to ask you a little bit about your background

in a sense that did you always know you were interested in this technology and this AI space?

Was this something that came to you throughout your career, or I know you went to school

in Harvard?

Is this something you saw there?

Tell us a little bit about your background and your journey.

Oh, man.

Well, I can't say I always was interested, right?

Because I grew up and I went to high school in the 90s and started going to school around

the dot-com era.

And so a significant portion of my being is feeling like I missed out on the dot-com revolution.

I was too young to know what was happening.

My first computer was, I think, in 1999 as I was going into my senior year in high school.

And so for years, I mean, ever since then, of course, I've been interested in technology.

But as I started to explore, I'd say machine learning, right?

Because we talk about AI as if AI is just AI.

But in reality, we're talking about a constellation of technologies, machine learning, and just

a number of different ways to engage things.

And as we, I assume, talk about productizing AI, that's going to be kind of a recurrent

theme, is it's more than just tell the AI to do this and it does it.

I think my company and what we're trying to solve it, it came together through, let's

say, 15 years of experience with the problem that we're trying to solve.

And then at the same time, kind of an alignment of my own ability to understand ways to utilize

some of this machine learning technology in the public space, a space that traditionally

hasn't been very innovative.

And that's how it all came together, was just experiments, understanding the problem

deeply, working together with a really stellar team of folks who understood the problem as

deeply as I did, but also some of them probably understood the technology better than me.

Very cool.

Okay.

Well, for those that are listening, give us an overview of what Inquisio.ai is, what

problem you're solving, and yeah, a little bit about kind of your vision for it.

So Inquisio.ai, in a nutshell, we're leveraging emerging technology to make government better.

And our focus is on what we call small G government.

So the federal government would be big G government, states governments would probably

be big G government, but local governments are small G that's cities, counties, school

districts, fire districts, sewer districts, the, the folks that you run into every day

at the supermarket, or if you have a pothole that you need to fix in front of your house,

you call them, or if you're trying to start a new business, build a, build an apartment

complex, deal with your driver's license issue, all of that is small G government problems.

And we are, we believe through experience that fundamentally the reason why so many

people have conflict driven experiences with small G government comes down to an information

management problem.

There's a lot of reasons that small G governments are super risk averse.

And so what that means is they introduce additional steps of bureaucracy and, and intercedent requirements.

So that way they avoid risk at all costs, but it makes them very inefficient and fundamentally

that's a, that's an information flow problem and a perfect problem for say AI machine learning

and a number of other tools to jump in and improve processes.

Okay, very cool.

Can you talk me through like a specific use case of a way that, you know, your, your company

or, you know, one of these local governments would be able to leverage your platform and

you know, like give me a specific use case of how this would be helpful.

I mean, you mentioned the pothole, I'm not sure if that's relevant.

Is this like a faster way to, you know, get a pothole filled outside your house or is

this kind of something different?

It could be, you know, our solutions aim to sit at the nexus of a bunch of different

problems and local government, but I think when we look at what we're trying to build,

we think of it this way that say that you do drive over that pothole and you, you have

now have to get a new tire, your rims are bent, whatever, there's some sort of damage

to your car.

Right now, what do you do to find out who like, what do you do at that point?

You just say, oh shucks, I talked to my insurance company or I go pay for it.

I mean, that's, go ahead.

What would you do?

Oh yeah.

I mean, yeah, if it's me, I've heard somewhere that somebody is liable.

So yeah, probably would hit up my insurance company to be honest, because I don't know

how to get in contact with local government, anything, and I would ask them who's supposed

to pay for it or what I'm supposed to contact that probably my method, which is also horrible

method because I hate getting on the phone and I hate contacting insurance or customer

service, anything.

Well, we'll let somebody else solve the insurance problem, but I, you know, fundamentally what

you're going to have to do in most cases is contact your local government, they'll tell

you to fill out a tort form, you'll fill it out, you'll hear back from six to eight weeks

after you've got your receipt and paid for everything and you might get it paid for

by the local government.

You might not depending on the pothole and the cause or the government may turn around

and talk to the, maybe the owner of the road, if it's private road, things like that, but

these natural questions, you know, that question of what would you do can be quickly answered

if there are tools that have access to the policy, to the laws, to the things where you

could find the information and you can ask that simple question.

Not only should you be able to ask it at your level of understanding, I ran into a pothole

with my car.

What can I do?

That should be the extent of your question.

You shouldn't have to know what a tort form is.

You shouldn't have to know any of the legalese language.

In fact, you shouldn't even have to speak English very well or at all.

You could be monolegal, non-English and you should be able to get your answer in Spanish

or whatever language that you speak at home.

And so that's, that's just one example of reducing the barriers.

I think to us, when we look at that, that is, that's a very clear use case, maybe a more

challenging one is that every record that a government creates is what's called a public

record and you, you as a citizen have a right to those records.

However, to get access to those records, quite often what you need to do is under, know exactly

what you're looking for and you need to be very specific in what you're looking for.

And even then it's going to be a challenge to get it in a timely way.

You look at just a Google search of records delays, police departments.

You will see Seattle Police Department two year long record delay in getting records

requests fulfilled and you'll see this across the US.

And it's not just for something like police department records, it could be for, say you

want to build a, build an apartment complex or do some, do some construction in your city.

Quite often there's a discovery phase of that sort of thing that takes years to figure

out what permits are needed, what, what the building requirements are in this city versus

that city versus the county.

Do I exist in the county?

Do I need to get what's called a variance because my property doesn't perfectly align

with the policy of the given jurisdiction and which jurisdictions are, do I need to

seek variances from all of this is a very long complex process that adds, adds quite

a bit of cost to the initial idea of developing a property, but also is an information search

process that you engage with the jurisdiction and likely the person you're engaging with

doesn't have all the information.

So they have to do research and we can sit in the middle of that and you can query a

natural language against building codes against building codes of various jurisdictions and

it should be able to sort that out for you and identify what is likely to be your answer.

And then you can ask follow up questions of, well, how do I apply for a variance or what

is a variance or what is, you know, how do I apply to change this, right?

Which might be more natural language way of saying it and as you'd understand, it should

be more capable of that flexibility and eliminate, you know, 90% of the slow back and forth emails

with somebody in an office somewhere that you're one of a thousand emails and they'll

get to you when they get to you, so.

I love that.

That's super cool.

It definitely sounds like I'm now thinking of my dad in Canada.

He had some building permit stuff.

He was trying to get worked out and most of, I think what he ends up doing is literally

going down to the city hall, trying to talk to someone there and getting all the answers

and definitely a high barrier to entry that is not, you know, for everyone.

So super cool.

Okay.

A question I would love to ask you about all this because I think it's super fascinating.

What is like the inception story of Inquisio.ai?

How did you get started?

I know you're the mayor and so is this kind of born out of that and seen some of the inefficiencies

in government there?

Like, tell me a little bit about the beginning.

It's a little bit of all that.

For years, I've had this idea that there's got to be ways to implement machine learning

in some of our processes and I say at the city, but I'm not saying I'm using my city

as a petri dish.

It's just, I experience it with my staff.

I see them overworked in information challenges and I wonder if there's a better way it just

doesn't exist yet.

And I would say, I don't know, maybe 18 months ago, I started playing around and it sounds

like it's just yesterday, 18 months ago, doesn't it?

I really started to focus machine learning on this idea that, so there's a tweak classification

is, I don't know if you can call tweak classification a traditional machine learning problem, but

it's something that you can in a university class on machine learning, probably entry

level machine learning, you would go through tweak classification of intent, like a positive

negative neutral intent on tweets or whatever they're called now under mask.

And the idea is you get a statement, 140 characters, 140 words or whatever it is, and you chop

it up and you say, okay, eliminate all these words that are it, the and we call them stop

words.

And then you would build a corpus, a library of words that are probably positive intent

like happy, you know, something like that.

And corpus of words, they're probably negative intent like hate.

And you, you then would start to construct different models of Corpi, I guess that would

be the plural against a training set or against a known set like a hand scored set of this

is a positive, this is a negative, this is kind of neutral.

And see if you can, you can find the models that best fit that without being 100%, right?

Because if your model is 100% accurate, it's probably only accurate on a very small set.

If you generalize it, it'll break down.

And in time, you'll find different models will work better to predict whether a given

tweet that it hasn't seen is positive negative neutral would say 85% efficacy.

So if you take that back it up a little bit and you say, a tweet is just a statement made.

We should be able to do this to conversations that happen in a structured environment like

a council meeting, or a an official conversation in in in policy.

So council meetings are generally structured.

They follow what Robert's rules of order, people are addressed to speak, they speak,

they make motions a second, they vote on all that other stuff.

And utilizing a variation of say tweet classification, you should be able to determine with pretty

eye accuracy, whether or not somebody has said whether or not somebody has proposed

a motion.

And because there's going to be certain words that typically follow or proceed a motion,

whether or not somebody second that motion, whether a vote's been taken, what the vote

cast was, what the solution was, was there an amendment to the motion, was the meeting

then adjourned, right?

All these activities, you should be able to identify using one model or a couple different

models to figure out what the heck's happening in one meeting, and then maybe what's happening

across a collection of meetings.

And if you are able to create a data set of what's happening across a number of meetings,

then you can query that data set for things like generating minutes and other things.

And so that was the initial concept 18 months ago is let's see if we can, with some fidelity,

predict what's happening in meetings.

So we, I would say we did a pretty good job.

We did a good enough job to recognize it was possible.

That is not a production quality, but it was enough to say, okay, there's something here.

And then of course, large language models very much hit the cultural zeitgeist right

around the same time.

And it made us realize that there's a bigger conversation here than just identifying whether

or not conversations have an intent.

And that got us thinking about how do we apply this technology in a bigger way across a space

that we know very deeply.

And that's that at that point, it became a little bit of experimentation, a little bit

of net present value projections and identifying what makes sense to chase down what could

be sellable in this market.

And that's how the company came together.

And we're focused on generally the information thing, we call it the public records problem,

but that's the center of so many problems in the public space that you, your parents,

everybody who at some point will come in conflict with their government or be curious about

their government.

That's where we intend to be.

Very cool.

I love it.

So obviously, you know, you've mentioned there is definitely a lot of inefficiencies.

I think people know this in governments of all levels.

I mean, organizations of all levels in general.

So I imagine this, this is sometimes a bit of a tricky area to work in.

What would you say are some of the biggest challenges or obstacles you kind of faced

in developing this tech in this specific face and in space and how are you kind of overcoming

those?

Well, I think the biggest challenge is going to continue to be a challenge is that by its

nature, government is not super innovative in changing its processes.

As I mentioned before, risk is very risk averse.

You wouldn't settle for your government betting, you know, 40% of your budget on red at a casino.

So that would be hyper, hyper, you know, risk associated.

But at some point, you know, governments think this way.

They think I have two options and either one of them has some amount of risk.

How do we find ourselves moving towards the one with the least amount of risk?

Because we have a fiduciary duty to the public's money.

And so they do that by adding in processes and different layers of accountability.

So that way, when a government does get sued, because they will, every government gets sued.

Then they look back and say, we followed this process that we developed because it was the

least risky and it was a fiduciary responsibility.

So when we engage potential customers right now, they see this idea of AI and they think,

oh my gosh, this is a whole new policy environment that we're not sure how to approach it from

a risk mitigation perspective.

In fact, I think the state of Maine just pasted moratorium, which is a moratorium is a short

term ban while they work on policy on utilizing AI for government purposes.

Now, that to me is kind of silly because they utilize AI and their spam filter and their

email, right?

So at what level are we talking about utilizing AI?

They probably utilized AI on their search engine as they were figuring out how to make

this policy.

But that's one reaction and you're going to see those edge cases come to being and you

can see others who are hyper innovative, but the mean kind of middle of the road is going

to right now be concerned about what's the policy, what are the elements of risk that

we need to create policy to surround and policy is always a slow process.

And I think what they would be looking at is things like what sort of biases do these

AI models get built from or perhaps reinforce?

And you'll probably see that, that'll be more of a California question than a Texas question.

But you're also going to see something fundamental like what do you do about the chain of custody

of privileged information?

So if you have your social security number on an application for something at your local

government and all that gets sped into one of our AI tools or somebody else's AI tools,

who has access to that information then?

Does an intern at AIstartup.com have access to the raw data or is it obfuscated in some

way?

Or can it be reproduced if somebody were very clever at query engineering and query hacking?

These are new terms that just exist in the last year, right?

Could somebody very cleverly hack your algorithm or hack your interface through clever querying

and get access to information that should be privileged?

And so I would call that the chain of custody of privileged information.

Who owns that?

But on the flip side, there's also the question of, well, what can you do that Google can't

give an access to the same information?

And so I think the opportunity that immunities are going to be presented with real soon is

taking all the information and putting up a wall around it and saying, if you want access

to this information, you have to go through this portal that we've, well, contracted with

Incusio AI or created ourselves or whatever.

And that's not crawlable information.

That's information that the agency can provide.

And that retains ownership of all that information.

So it does create an element of risk control as well as risk exposure at the same time.

But these are all challenges that we have on top of starting a business, demonstrating

to funders that GovTech is a viable space and actually delivering the technology.

There's so much work to do here.

Yeah, I believe it.

Absolutely incredible.

I'm actually very excited to have tools like this in the future.

I think Incusio is going to be a big player in this space.

One question I would ask you as we're wrapping up this interview is, what's one piece of

advice you feel like you could give to local governments that are perhaps looking at implementing

AI into what they're working on?

Of course, there's the one reaction from Maine that's just, shoot out the moratorium, shut

it down for now.

There's other people I've talked to from the city of Austin in Texas that are being quite

proactive in implementing a lot of AI into what they're doing.

But yeah, what's one piece of advice you feel like you could give people in this space?

As a policymaker, I would say pay attention to the privileged information problem that

I just brought up.

So be careful what information you're providing to any vendor.

Whether or not you have that vendor on contract or somebody's just going to chat GPT and throwing

information up there, don't play with data sets that have privileged information right

now because that problem hasn't been solved.

I think that's probably a piece of advice I'd offer to any policymaker in any government

space.

But I'd also say understand your problem.

Not every problem of government will be solved utilizing AI.

In fact, government's probably going to be one of the areas in many ways least affected

by this industrial revolution.

Government is the business of relationships.

And those can't directly be replaced by this technology.

Not yet.

Not in a way I can foresee yet, but that could change fast.

But don't look to AI to solve every problem.

But where you can with processes, I think this is going to be revolutionary and it's

going to allow you to focus your efforts on places where you would rather spend money

as a government.

You'd rather spend money on having more engagement with needy populations at that than you would

have on having another layer of bureaucracy.

So this is going to be a really exciting time in the next couple of years.

If you have to implement a moratorium, make it short because I think that you're going

to miss out on some things if you do have too long a moratorium.

And I think one more thing.

We were just at the League of California Cities and one of the cities had brought up

the idea of should we create a policy today?

And I think our perspective on that is just be careful with the policy you create today

that you don't do more harm than good because in six months that policy is probably going

to be outdated.

And you might get stuck with a vestigial policy that created more challenges than you already

had in the first place.

That's a circle back on my warning about those moratoriums.

I don't think it's necessary right now.

In fact, it's probably more harmful.

Yeah.

I definitely agree with you there, Josh, incredible advice, incredible insights.

I'm so excited for the space of AI and government tech and everything going on.

So thank you so much for coming on the show today and sharing a little bit about that

with us.

If people want to get in contact with you, maybe try Inquisio or find out a little bit

more about what you're working on.

What's a good way for them to find Inquisio and gain contact with you?

So you can go to our website, inquisio.ai.

You can go to LinkedIn.

Now, there are multiple Josh Penners on LinkedIn.

They're all fantastic people.

I'm pretty sure of it, but I'm the only one that's the CEO of Inquisio AI.

So feel free to reach out to me on LinkedIn.

I'm definitely an open networker.

If you send me a request, I'm going to accept it.

And there's a ton of different ways to get in contact with me there.

Perfect.

And I will also leave a link in the show notes to inquisio.ai.

For the listener, anyone interested in checking it out can do it there as well.

Once again, thanks so much for coming on the show, Josh.

To the listener, thanks so much for tuning in to the AI Chat podcast.

Make sure to rate us wherever you get your podcasts and have a fantastic rest of your

day.

If you are looking for an innovative and creative community of people using ChatGPT,

you need to join our ChatGPT creators community.

I'll drop a link in the description to this podcast.

We'd love to see you there where we share tips and tricks of what is working in ChatGPT.

It's a lot easier than a podcast as you can see screenshots, you can share and comment

on things that are currently working.

So if this sounds interesting to you, check out the link in the comment.

We'd love to have you in the community.

Thanks for joining me on the open AI podcast.

It would mean the world to me if you would rate this podcast wherever you listen to your

podcasts and I'll see you tomorrow.

Machine-generated transcript that may contain inaccuracies.