The Bootstrapped Founder

With the new year looming ahead of us, many founders will wonder how this growing AI movement will affect their entrepreneurial chances.

Will this groundbreaking new technology make the impossible possible or will it take our jobs?

Today, let's look into the crystal ball and see a few opportunities, challenges and threats ahead that AI systems may pose for software entrepreneurs and creators alike.

This episode is sponsored by acquire.com. More on that later. Now let's talk AI.

We have seen a Cambrian explosion of AI tooling and progress in the field in 2023.

I believe that this will be trumped significantly in 2024 and I don't see it slowing down anytime soon.

Let's recap where we are right now with this.

OpenAI has recently released GPD for Turbo.

Metas Llama LLM, their AI system, has been adopted by the open source community and refined. I'll get to that later.

The French open source model Mistral is making waves because it's so small, so compact, yet can already beat GPD 3.5.

And Apple is rumored to have been working on models too that can easily run on their current iPhone models.

The trend with this tech is quite obvious.

More players are joining the game.

Models are getting better, they're getting more available and they're becoming smaller.

And maybe most importantly, they're becoming more and more multimodal, which is worth looking into.

A multimodal large language model is capable of dealing with not just text as the original ones were, but also audio, images, video and more.

It can interpret and generate, that's important too, any of these formats.

If there is enough data of some sort, you can bet that the models currently being trained and tuned are more and more capable of understanding and manipulating all kinds of media, not just text.

We're already seeing some of these systems in their early stages in the wild.

Otter.ai, that's a tool that I use to create transcripts for my podcasts and all that.

Their zoom integration summarizes ongoing video calls as they happen.

And then the whole thing presents actionable notes and meeting outcomes seconds after the meeting has concluded, sometimes even right after a decision was made in the call.

It's pretty cool.

And you can just export it, you can ask questions of it in some kind of chat based system in the Otter app.

It's a pretty interesting and quite impressive first version of something that I see growing even, even bigger and more impressive.

Future systems like this will already start working on the problem or at least dispatch the proper instructions to the people who will work on the problem for you while the meeting is still taking place.

That's the power of a multimodal system.

Right.

It can take in the audio and crunch out text or even create a little instructional video that comes from just the description of what people want and immediately give it to the right people or just talk to the right systems to get the thing done.

Right now, notes will be filed and analyzed and maybe we'll see short clips be generated and transcribed and sent to the people who need just that part of the conversation.

In the podcasting world, this is already commonplace.

It's become normal in there to just get the audio, get the video and all kinds of things are created from that.

I use podium.page sometimes for my podcasts where I get all the tags that I need for YouTube, you know, the categories and that kind of stuff.

I get title suggestions, I get description suggestions, I get this kind of summaries of the whole thing.

Maybe some show notes.

Maybe I get chapters that I can automatically use to structure the thing.

All of this is effectively multimodal, right?

Takes in audio, takes in video and churns out text or structured data.

This will become more normal in the less tech literate industries very soon.

And I think people will expect to find this kind of assistive technology in the software tools that they use.

If they work with audio, they'll expect transcripts and summaries and being able to ask questions of the conversation that they had in other fields.

The same is true.

Like they want different modes of interacting with the data and software offerings without these capabilities will quickly feel outdated.

Clearly, the early adopters, they expect AI centric features in their edgy tools already.

And those tools tend to deliver.

But the shift towards the mainstream will add even more pressure to facilitate these features onto founders building businesses in these fields.

So either you embrace the technology or your competitors will use it to chip away at your margins.

AI as tooling, it will become as commonplace as using databases and analytics and having avatars in your social media feeds.

It's just going to be normal to expect AI features.

Fortunately, there are a few trends here that will make this easier for you as a founder.

I've been making good use of one particular trend for my own SaaS, Podline.fm that I'm currently building in public.

The magical ingredient here is running these incredible AI models inside my own infrastructure.

Local AI is what I consider this to be.

And I think it was impossible, almost impossible a year ago.

So now it just takes two installation commands and you get it done.

It's really, really easy now to run an LLM on your own server or your own computer.

You can run a significant number of these small but powerful language models on consumer hardware without even needing a graphics processor, which these kind of things tend to run on.

You can run these LLMs on your CPU.

You don't need a GPU for this.

It's incredible.

Podline has a miniscule server, just 16 gigabytes of RAM, which can easily spin up a language model like the Mistral 7B and get a chat GPT like response in just a few seconds.

There's even example code out there that lets you spin up your own API server for this, ready to be used with your backend services that already can and do talk to other APIs.

Other than the cost of this small server, which for me is like 20 bucks a month at max running your own AI system is free.

The models are free.

The software to run the models is free.

All you really need to pay for is electricity and I guess disk space.

But that is also pretty insignificant.

I think one of the models that I run is like 4.5 gigabytes in size.

I think it's a quantized version of something that would probably be like 8 gigabytes in size as its full version.

It's pretty impressive how small these things are.

And all of this is quite massive because there's an open source community out there and everybody's like training AI models for many different tasks, each different task that they needed for.

They train a model for and it's quite likely that your tasks that you need the AI for are already out there with fine tuned models built for them already.

Every day new models are uploaded to hugging face.

That's kind of a model repository website where you can download and use these models for free.

Most of them are licensed in very business friendly ways too.

It's pretty impressive.

And the categories that they have like specialized models in that's incredible.

I recommend you go to huggingface.co and just check it out.

Like from speech recognition, text to image, image classification, text summarization, translation, video generation.

You will find a model for all of these that will take a previously manual task that some human had to do or at least check off on and do a superb task automating it.

And even if those results aren't perfect, I mean, we are in the early stages of this whole thing.

You can fine tune all these models yourself over time using the same platform and none of this will be prohibitively expensive.

And even though we're still in the early days, it's already super easy to integrate this into our tech stacks.

There's just one risk here.

And that is depending too much on AI platforms because it's pretty easy to get a local AI going.

I told you, right?

You download the software and you download a model, you kind of configure it and you run it.

But it's so much easier even than that to just implement a few API calls to open AI platform.

And they have their massive server cluster there to do the work much faster for you.

Pricing is currently quite low.

And I expected to go down even further because competitor pressure exists and it is increasing with all these other AI platform providers.

So it makes sense to just pay a few cents for tasks like summarizing an hour long conversation or generating a video just from a script, not even from a script, from a command.

Right.

Just imagine what you could do marketing wise with systems like this.

And this is a pretty simple thing that I already see people doing with a multimodal system like that.

An LLM that can just take a command like, hey, go to this guy's website, check out their website, look at their pricing, look at the features that they offer, then create a script for a voice message that you would send them to tell them of something that you found on your website that you could help them with for your particular software, whatever you might offer.

Maybe you have some kind of customer service software and you say, hey, here would be a perfect location to put a little widget or to put the link to your knowledge base.

And we can power this for you, create the script and then turn that into a video of showing our product being used on that website.

And if you have the right data there, if the model is tuned for your product, the model is tuned for these kind of tasks from this one, go to this guy's website, look at it, analyze it, create a script and make a video.

That might take you like two minutes and you have a video file already fully rendered, uploaded, ready to be sent to that person to show them how good your product might work on their website, all automated.

Because obviously go to a website XYZ and then analyze it.

That's a prompt you can automate and run 10,000 of them in a row.

You just need a list of websites that you want to help with your product.

It's going to be an interesting new way of cold calling in a way or cold emailing.

We're going to have cold videoing, I guess, in the future where people just create AI based videos that still are very personalized to the precise thing that you want to do.

It's pretty impressive even to think about it.

It blows my mind.

But the most important part at this point is that you manage your dependency risk here.

Because my rule for this kind of feature is there has to be at least a local fallback for each API call that I make to an AI provider for whatever this might be.

And for part line, I've implemented just that.

I have two things that I do with AI in this thing.

One is transcription.

The other one is summarization with an audio file gets transcribed.

It either goes to the open AI platform.

That's why I send it to be quickly transcribed.

Or as a fallback, I run whisper that CPP on it locally and whisper uses the exact same language model as open AI is using on their servers.

It just runs it on the local computer.

And it's maybe 10 times as slow.

Right.

The other thing might take a second.

This might take 10.

But for a background process, that doesn't really matter.

And for something that I get to run for completely free on my computer at however many people need it.

That is pretty awesome.

The same thing for summarization.

Right.

I just have llama dot CPP spinning up an instance of the Mistol 7b model in the back end.

And it does the exact same job as the GPD 3.5 would do or GPD 4.

If I can avoid it, I will never fully depend on someone else's platform for an AI feature.

And that's what AI should always be.

A feature of your product, not your product itself.

Because you're competing with pretty high budget companies here.

Right.

Like you're competing with the Microsofts and the Googles of the world if you want to offer some kind of AI.

And there is a significant open source community too that would do the work for free.

I would just say steer clear of like AI as a service businesses and be more soft as a service with AI as a feature business.

AI should always be just be a feature or part of a feature, not the product itself.

Because it can be an interface to your users data or a means of transforming data into shapes that are more useful to your users.

But if you want to offer AI, you will be outcompeted quickly.

So let's maybe talk about these different kinds of things.

AI as an interface.

I think that is just a good old chat bot, right.

Which will likely morph from this conversational novelty that JetGPT still is to most people to an actually real agent that people perceive to be interacting with.

That will not differ at all from a personal virtual assistant that you just send tasks via email as you do now with real people.

Just imagine that concept completely virtualized, completely AIified.

And your SaaS is offering it for your particular product.

Imagine your SaaS product is not just a website that user goes to and logs in and a dashboard or whatnot.

But it's actually a virtual person, an agent, an entity that your user can reach out to.

Ask questions from and have execute service specific tasks that your service offers.

Right.

Like it's not just for help desk stuff.

They could also just really tell that agent, hey, I need that file delivered to my email inbox in four days.

I don't need it right now.

Send it in four days.

If you give the AI these capacities, you could literally offer this as a virtual assistant for them.

Not just a website to log into.

Your SaaS will truly become a service that builds a transactional relationship with its users.

One interaction at a time, a bidirectional exchange that you and your support staff can always jump into when needed.

Because you can take over the agent, can literally impersonate that agent.

I think this is one of the first times that the word impersonate actually means something other than like deceiving or acting like somebody else.

You become the person in that virtual agent at that point.

So obviously customer support is the massively impacted space that AI systems will have a massive impact on.

And if they haven't already, because first level support in many cases is already, it can and it definitely should be handled by always on, always focused AI systems that both have insight into the details of the customer's account and the issue that they might be having.

And they have the whole history of all known problems and all solutions with the product in their constantly updated training data.

This is a job that even the most experienced human customer support agent will struggle to live up to.

Mostly because they need to sleep as a human being.

For this kind of immediate help, AI systems are great and they're smart.

They're really good at learning from things that they told in the past that worked.

You can build some kind of feedback system where the advice that the AI gives gets voted on by the user.

And if it worked well, it will be given again.

And if not, there will be a change next time.

Like these systems are kind of self-healing, which is really interesting because you're literally outsourcing a very interesting and I guess creative part of dealing with your customer, like analyzing a problem, finding a solution to a machine that never sleeps and costs you almost nothing.

It's great.

But even the background of your system, not customer facing, but back and facing quietly summarizing, transcribing, analyzing and generating data.

These systems will massively impact your business in the years to come.

So consider that AI tools such as these large language models of today are the pioneering systems of automation that can and will massively lower your operational costs on the one hand.

While at the other hand, do the work that previously either took a long time to do to process even in the background system by a computer or needed human creativity and ingenuity to solve, which is like analysis customer service.

Right now, a machine can do this pretty well because they have access to all the data of all conversations.

Unlike a human who only has the conversations they remember and need to look up things.

The machine does it in nanoseconds.

A human takes a couple minutes, makes a whole lot of difference.

So embrace AI without depending on it.

Make sure you have systems in place to fall back onto.

And I think that is my AI adoption approach for 2024.

I'll try to embrace it as much as I can without building castles in the sky on somebody else's platform.

And that's it for today.

I'll briefly thank my sponsor, aquire.com.

Imagine that you built a really cool SaaS product with a lot of AI as a feature.

Not an AI as a product kind of system.

And you have customers and generated a lot of revenue.

Problem is at some point, you're just kind of stuck.

You don't know where to go.

You may have lost your focus or you don't have the skill to keep doing it or just don't care anymore.

You want to build something else, which is not a big surprise in the AI field.

When new technology comes out every single day, sometimes you feel like I want to do this now.

Well, you have something there, right?

You have a business already with customers that are paying money, but you don't want to do it anymore.

What should you do?

The common narrative is that you buckle down and you do more and you make it even more interesting.

You make it do more money and everything is great, but that might not be what you need in that particular situation.

What you need is your time back and to get your time back and still have value from the thing that you built.

You can always sell your business on aquire.com.

Listing there is free.

They've helped of hundreds of businesses already.

People have sold for millions of dollars on that platform.

Aquire.com has the tools that you need. They will help you figuring out how much your business is worth.

Make it more sellable.

All that stuff.

I think that's a pretty smart move to even think about.

So if you just want to check it out, go to try.acquire.com/arvid and see for yourself if this is the right option for you, your business and your circumstances right now.

It's always good to be prepared.

And even if you don't sell this year, which, you know, it's going to end in a couple of days, you might sell next year or the year after.

It's good to know what the path is.

So check it out.

Try.acquire.com/arvid.

Thank you for listening to the Bootswap Founder today.

You can find me on Twitter at avidkal.

You will find my books on my Twitter course there too.

And if you want to support me in this show, please subscribe to my YouTube channel.

Get this podcast in your podcast player of choice and leave a five star rating and a really nice review by going to rate this podcast.com/founder and picking the platform that you listen on.

That makes a massive difference if you show up there because then the podcast will show up in other people's feeds.

Any of this will really help the show.

It's all optional, but I would really appreciate it.

Thank you so much for listening.

Have a wonderful day and bye bye.