This Week in Startups is brought to you by…

Embroker. The Embroker Startup Insurance Program helps startups secure the most important types of insurance at a lower cost and with less hassle. Save up to 20% off of traditional insurance today at Embroker.com/twist. While you’re there, get an extra 10% off using offer code TWIST.

Vanta. Compliance and security shouldn't be a deal-breaker for startups to win new business. Vanta makes it easy for companies to get a SOC 2 report fast. TWiST listeners can get $1,000 off for a limited time at vanta.com/twist

Superside. Design and creative are crucial for growth. Tech companies like Shopify, Amazon, and Meta have found the perfect solution: Superside. Get quality design at scale without breaking the bank for your startup at superside.com/TWIST

*

Today’s show:

VenturusAI founder Emanuel Ciciu joins Jason as he dives into the origin story of his new startup before talking about building vertical AI solutions (1:30). Then Blackbird VC's Samantha Wong gives a presentation at Angel Summit about identifying soon-to-explode startup markets (41:53).

*

Time stamps:

(0:00) Emanuel joins Jason

(1:30) VenturusAI origin story

(5:20) The evolution of prompt engineering

(9:09) Embroker - Use code TWIST to get an extra 10% off insurance at https://Embroker.com/twist

(10:23) The true cost of building on top of a LLM

(12:56) Building vertical AI solutions

(18:15) VenturusAI demo

(23:36) Vanta - Get $1000 off your SOC 2 at https://vanta.com/twist

(24:42) VenturusAI path to 100 customers

(29:51) Next steps for VenturusAI

(35:40) Jason places a bet

(40:28) Superside - Go to https://superside.com/twist to get $2000 off with Superside's Startup Accelerator package

(41:53) Blackbird VC's Samantha Wong on identifying soon-to-explode startup markets at Angel Summit

*

Check Out VenturusAI: https://venturusai.com/ Follow Samantha: https://twitter.com/holasammy *

Read LAUNCH Fund 4 Deal Memo: https://www.launch.co/four

Apply for Funding: https://www.launch.co/apply

Buy ANGEL: https://www.angelthebook.com

Great recent interviews: Steve Huffman, Brian Chesky, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland, PrayingForExits, Jenny Lefcourt

Check out Jason’s suite of newsletters: https://substack.com/@calacanis

  *

Follow Jason:

Twitter: https://twitter.com/jason

Instagram: https://www.instagram.com/jason

LinkedIn: https://www.linkedin.com/in/jasoncalacanis

*

Follow TWiST:

Substack: https://twistartups.substack.com

Twitter: https://twitter.com/TWiStartups

YouTube: https://www.youtube.com/thisweekin

*

Subscribe to the Founder University Podcast: https://www.founder.university/podcast

It's important to look at how we'll take GPT as an example because we've been seeing all these changes throughout time now and we can see how better it is, let's say, 3, 4 compared to the 3 version and so on and so forth.

So I would expect 5 to be way better than 4 is.

And if you're saying about 7, I don't even know what 7 can build.

I think if you're smart enough, you can build an AI to build an entire product for you, just depends how you want to look at it.

This week in startups is brought to you by...

Embroker's Startup Insurance Program helps startups secure the most important types of insurance at a lower cost and with less hassle.

Save up to 20% off of traditional insurance today at a broker.com slash twist.

While you're there, get an extra 10 percent off using offer code twist.

Vanta compliance and security shouldn't be a deal breaker for startups to win new business.

Vanta makes it easy for companies to get a SOC 2 report fast.

off for a limited time at vanta.com slash twist.

And SuperSide, design and creative are crucial for growth.

Tech companies like Shopify, Amazon, and Meta have found the perfect solution.

Get $2, 000 off with SuperSide's startup accelerator package at superside.com slash twist.

More important and elitist and that they understand business better than founders who create these things.

However, these things are actually valuable.

There about any discussion about a business and analysis of a business, while you're building it, if you're considering building it is great, because it lets you build a mental model.

And it got an incredible response from our audience.

So I told producer Nick, by the way, did a great job reading the news.

Just yesterday on this podcast, or this week on the podcast, depending on when we publish the show.

He he did a great job, he'll be coming back to read more news.

I said, Hey, producer, Nick, let's get this kid on the pod.

And how long you've been working on it would be a good place to start?

So I came up with the idea based on the studies you're just mentioning, right?

Most of these things you see in the standard report are things you learn in the business school, which have been true.

And I realized that would be a great benefit for everybody to be able to get this information without having to go to business school, not only to learn how to do it, but I think would be best if they would just be able to have at least a framework using all these answers.

And because we have LLMs at our fingertips these days, I came up with the idea, why don't we build a tool that allows entrepreneurs to generate all this content in a matter of minutes, and be able to go to it and try to validate their ideas before they start working on the actual product.

And people can pay 20 bucks a month for chat GPT 4.0.

And they can go to 1 of the dozens of prompt sites that are out there.

And these prompt sites will give you all kinds of tips on how to frame a question, or instructions, I would say it's probably better.

Their prompts are basically instruction sets for chat GPT.

And there's a lot of different techniques.

And I thought this would be a great way for us to introduce the audience here at this week and startups to advance prompt engineering because my gut tells me what you've done here is you looked at something like a SWOT analysis, S-W-O-T, SWOT analysis.

And you said, how would I prompt Chatshippeetee to analyze the strengths, weaknesses, opportunities and threats of this business.

Am I correct in that you've spent 100 hours banging your head against the wall creating super prompts?

I probably wouldn't say 100 hours, but probably if you think about optimizing everything to the point where it's now, we might get there.

So you took 25, 50 hours of prompt engineering?

So explain where you started your prompts for something like SWOT.

And I had given the example of a school that teaches you venture capital.

And that, I don't know if you saw the episode, I'm sure, your friends must've sent it to you.

So I just, and it gave me a really nice SWOT analysis.

What kind of prompts would, you know, a neophyte put in to try to, you know, say, you might just say, do a SWOT analysis on the following 1 sentence description of a business.

But you got more granular, you probably said, Hey, what are the strengths of this business idea?

Tell it to me, you know, as if I was an MBA or a business analyst at, you know, Gartner group, I don't know how you did it.

But so maybe talk a little bit about the evolution of your prompts, what worked, what didn't

So when we started was we I basically use every single framework that I managed to get from my business goal.

And I just grabbed the business book and I'm like, okay, what do we need?

We have a framework already built by people that worked on this before.

So going back to SWOT was give me a SWOT analysis.

So going back to prompt engineering, I came up with the idea, hey, what if I say I'm a world-versed VC with over 20 years experience in SaaS products.

And we tried that and again, it was a bit more complex, but it was not reaching to what exactly we're looking for.

And We did a bit of engineering hacking, if you want to call it, where we came up with the concept that LLMs have been trained on all these books already.

And again, asking him to be a world first VC with 20 years experience, it doesn't mean that that VC in particular, whatever that instruction would mean for the for the LLM would mean that he can cover everything.

So what we decided to do is to try to use the LLM with the information we have to improve on its own prompt.

Yeah, probably at some point to some to some degree, on some aspects, not all of them.

And we've been working to give them, let's say, for a particular thing, I'll talk about finances.

I'm like, hey, finances, I'm actually interested in the following topics.

And these topics are usually things you will get from books.

Before you validate your idea, you would need some sort of financial projection, you would need to understand how much money you will initially need to spend to be able to spin it from ground up.

These kind of criteria, as we mentioned them from the beginning, and then we were just looking on how we can continuously improve on that.

And at some point we got to, Again, you can always improve on it more and more.

And I think the more time you're going to spend in improving the prompt, probably the better it's going to be.

But at some point, we decided that it's good enough.

And we now want to add more value for those for the money and be able to build on top of it.

So this is probably the main reason we didn't think about building an LLM in the first place, because we were like, there is already so many companies like big companies doing this that have been invested in, they have a lot of investment they are investing,

You don't need to create a language model use chat GPT 4, I assume or chat GPT 3.5 or using some other 1.

Or it's just a technical for redundancy or cost or something?

Technically for it, it would give better results most of the time.

Like our goal is also to make this cheaper on a long run and be able to offer it to everyone.

Like, our initial idea was to try to do this like a B2C thing.

But we had a lot of companies signing up from the standard version.

And When we decided to go with the advanced features, we looked into being able to put them all through JGPT-4.

Listen, I work with super early stage companies that launch literally year 0.

And then we hit the series A, people have thousands of dollars in MRR, and they maybe they've only raised a couple of hundred thousand before that series A, and they don't have their insurance set up.

And in fact, we recently had a great startup that didn't have DNO, and we had to really stop everything because they were having board meetings, they were making massive decisions, there were legal issues.

And they didn't have the basic DNO insurance that protects directors and officers.

Embroker is business insurance built specifically for startups.

A single application will help your startup get 4 quotes for 4 lines of coverage in 15 minutes.

Think about that 4 quotes, 4 lines 15 minutes.

And they're going to connect you with 1 of their expert brokers for unmatched service that goes beyond your policy.

Try and broker today with the code twist and you'll get 10% off their startup package in broker.com slash twist.

That's EMBR0KE r.com slash twist and use the code twist for 10% off.

Okay, let's get back to this amazing episode.

in terms of if I wanted to do 1 or per hundred analyses?

What does it wind up costing you in GPT 3.5 slash 4 cost just roughly?

First of all, it depends on a lot of criteria, right?

We know that LLMs usually use some token methodology in order to charge you.

And usually that it's based on how big the prompt is, how big the response is gonna be, and also how complex it's gonna become.

At the beginning was probably roughly $1.

which was a lot of prompts, I'm assuming.

And then we look on how we can optimize that even more without trying to affect by any means the actual prompt, like we the response is really important.

But we don't want to affect the quality of the report.

But we want to be able to give this as cheap as possible.

And you've got it down to less than that now.

And then I just saw, is it Claude, the other 1 that just came out?

That's 1 quarter of the price or 1 tenth of the price?

I'm assuming as a founder in the AI space, you immediately went and checked it out.

What was your initial impression of the new Claude?

I think it's just called to I don't think

to the best of your ability, like, yeah, describe it.

It seems that at least from a business perspective, it seems that it can do some things a little bit better than GPT does it at the moment.

But I think, again, it all depends on on how you're going to look at the entire product in the long run.

And why do I say that is, If you want to just build a prompt that would deliver you a certain content, probably you'd be able to do it at some point with most of them, because I'm, again, I'm using a logic assumption right now.

I don't have anyone in open AI or Anthropx to confirm that to me, but I'm guessing we all use the same information to train all these LLMs, right?

And I guess the criticism of some of these businesses built on chat GPT for or the criticism of businesses like this is oh, Can I just use it directly in chat GP for so you and I both know that there's a lot of accoutrement And features you can build on top of the chat gpt responses that you will never do so maybe you could explain why you're convinced is a good business idea and that people will want a more vertical solution that you're creating with?

Bells and whistles and details that maybe are available in chat.

Or do you think an AI eventually is gonna get so good that chat you before 6 or 7 or Claude 6 or 7 or barred 6 or 7 are going to just do a better job than any startup that it's, you know, builds a wrapper around the stuff.

I think it's important to look at how we'll take GPT as an example because we've been seeing all these changes throughout time now and we can see how better it is, let's say, 3 or 4 compared to the 3 version and so on and so forth.

I think if you're smart enough, you can build an AI to build an entire product for you.

But I think you get, you will get to a point where you could build, they call it GPT agents.

Where explain to the audience, GPT agents.

So GPT agents would be a piece of software that would allow you to instruct GPT different instructions based on the agent.

Then you will instruct that agent to be a developer and act as 1.

Then you'll be, hey, I need someone to make some research for me.

But then we, how obviously you would need to build on top of it more things like I need information from search engines.

We know that at the moment, that's really difficult to get.

So you would build on top of that to be able to allow the agent to be able to get that information from you from a search engine, then feed that back into the LLM and be able to give you a better or more accurate answer.

Yeah, and so this concept of role prompting is saying, hey, you're an MBA, or you're a teacher at inseed or harvard business school you studied the works of and you give a list of the works of 10 you know classic books on business.

Analyze this business to the frameworks of these 10 people.

You could actually start to get really interesting things going here.

And that is something that is going to be really interesting.

You didn't feed it in the works of major authors or thinkers and business cases, because Harvard has all these case studies, they take those very serious.

Stanford has case studies, they're considered like very important IP.

It'll be amazing if Stanford creates an AI or Harvard creates an AI of all those case studies And then they say hey go through these case studies and use the what you've learned there as analysis for this But you got to assume that Chet GPT sold that I think opening I probably stole all that stuff already.

yeah, you open crawl of the web it seems I mean, well, we'll get this information from the lawsuits eventually, but they clearly went out.

If it's answering these questions so well, yeah, I mean, it probably went and took the full works of all of these authors.

And then the other thing that's happened, correct me if I'm wrong here, is on the web, if there's a great book like Good to Great by Jim Collins, there have been 10, 000 blog posts, articles, summaries, cliff notes, etc., spark notes about that book.

There's been 10, 000 podcasts about that.

So if you were to read those, those might be better than the source material, because they have a level of analysis on top of them.

So the truth is, even if you wanted to protect the IP of Jim Collins's works, good to great, etc.

You would have all the people who have analyzed it and done fair use criticism of it, that you could just leverage that.

So it's kind of interesting how this is all unfurling.

And I think probably before we see more action on the GPT side, probably we'll see companies like Mind Journey being able to explain how they design all these images on the spot, knowing that certain prompts would actually even show you different copyrights material from where they got inspired from.

And this is how everything started in the first place.

When they were generating all these images, they were seeing that there are some watermarks in them.

How can, how can it be a watermark if it's generated by the AI, right?

And if you missed the episode last time, I'll just share my screen here really quick.

And I said, a school to train people to be venture capital.

So I'm have a lot of people who want to come work for me for free.

And I decided, I'm going to create a school to become an associate, basically, to get a job as an associate will be the and it's basically a Kauffman fellows competitor, if you know that program, which is $80, 000 for 2 years.

So my idea is to make this like 25k for a year, full time.

And you come work in person with me at my new incubator.

And all you do is work with me and my team directly meeting 15 companies a week, sitting in on the investment team meeting, and then I'll have 10 people paying me to go through 15 companies each week.

This is some like Tom Sawyer, Huck Finn stuff like paint the fence.

And these kids come along and they're eating some apples and 1 of them's got like a toy.

And they call him and say, what are you doing Huck Finn?

And he says, or Tom Sawyer, I can't remember which 1.

And this is like an adventure novel for kids.

And so he says to the person with the apple, yeah, this is the most fun you can have on a summer day.

If you'd like to whitewash the fence and paint it, it's 1 the cost is 1 apple.

And he says, Oh, is it really a lot of fun?

He's like, it's the most fun I've ever had.

And gives him the apple kid starts painting the fence.

Okay, as a toy, he says, Yeah, this is the most fun.

If you let me play with your toy while you're watching the fence.

I'll give it back to you after but you can watch the fence.

all of a sudden, he's sitting there eating an apple playing with a toy, and they're doing his work.

And the lesson there is like framing, like, hey, you know, this could be drudgery, it could be a chore, or it could be a lot of fun.

And it is actually a lot of fun to paint, right?

So yeah, that's the story of like Huck Finn, etc.

Venture Capitalist Training School, comprehensive analysis feedback, industry insights, SWOT analysis.

And it said strengths, unique business idea with limited competition in the market.

There's really only 1 I can think of that actually is a school for venture capital.

Yeah, Tom Sawyer by Mark Twain, not Huckman.

Unique business idea with limited competition, the market's true.

Taylor training program can address specific skill gaps in the industry.

Did they teach you anything about, you know, startups and investing in companies or evaluating business ideas?

Usually, I guess, again, I speak from my experience, it depends where you go to business school.

So we basically, they start with some basics.

And usually, they try to bring different people that made already companies and startups to talk about all these things and how they validated all the books or all the frameworks from the books throughout their businesses in order to achieve that

great potential to establish strong industry partnerships for internships and job placement.

And it was tight, because I'm assuming you give it a very confined, give me a 10-word answer, give me a 15-word answer maximum.

That's really difficult to tell them because we have to explain everyone that LLMs don't really understand it.

Like they don't have an understanding of exactly what you mean.

So sometimes if you say, can you write me a thousand words essay, it might, he might make close to that, but it doesn't mean he's going to do 1000 words.

Usually, we don't specify certain things, but we specify what we want to see.

So in marketing strategy, we want to see this certain points which are mentioned without within our prompts.

And so recommended marketing platforms game changing.

But the best part was it said books to guide you along the way venture deals by my friend Brad Feld, the lean startup by my friend Eric Ries, both friends of the show.

And I thought that was pretty clever that so I guess at the end, you say, hey, what book should the person read?

Or or you say like, what topics are mentioned here and then search for further reading?

So when we were building the standard, the standard report, I thought it's really important people when they start any venture, I would always recommend start with also book.

While you build this, try to read something new.

And what we're looking at is I basically went to what the prompt does is based on the prompt, we want to be able to deliver you books that are specific to that prompt you wrote.

So if we're gonna write, I don't know, going back to what Sonny was saying, we, artisan water from icebergs.

I know he mentioned ice cubes, but I know that there is a company that builds that does artisan water from icebergs.

They will try to feed up information about people that are that are basically doing things with businesses with water with I don't know bottling water, what does it take?

And we wanted to be able to feed that as a validation of someone that actually does it.

If you're a SaaS or services company that stores customer data in the cloud, then you need to be SOC 2 compliant.

You know that from a third party and you need that third party to close big deals.

And if you want to get compliant easier and faster, you need to use Vanta VANTA Vanta makes it so easy for you to get and renew your sock to on average fan customers or sock to complying just 2 to 4 weeks, pair that to 3 to 5 months without Vanta and Vanta can save you hundreds of hours of manual work and up

This is a total no-brainer and Vanta does more than just SOC 2 compliance.

They also automate up to 90% compliance for GDPR, HIPAA and more.

You can't afford to lose out on major customers.

You can't lose those major customers because you don't have your compliance dialed in.

Get your compliance automated and tight and tight is right.

Here's the best part, Vanta is going to give you $1, 000 off.

Get $1, 000 off at vanta.com slash twist.

That's vanta.com slash twist for $1, 000 off your sock tube.

So you got a couple of hundred people to pay 20 bucks a month for this already?

See all This Week in Startups transcripts on Googlepodcasts

VenturusAI's instant MBA & Samantha Wong on identifying soon-to-explode startup markets | E1777