based
Share
Explore

based
Whitepaper - 2024

🥯 Introductions

Hi! I’m zach and I’m the founder of Based Social Company, we’ve launched a few apps and experiments in the space so far and I wanted to write through our full perspective and vision for where we are heading.
This will cover our motivation and the climate of loneliness over the past decade, how we imagine a perfect companion, what we see is the market, and what we’ve tried so far.

🧗 Call to action

Loneliness has been a global epidemic for nearly a decade now. With the rise of social media, there has been a strong lack of meaningful close relationships forming. At the core of the issue is hyper-connectivity.
image.png
As humans, we can maintain 150 relationships within our lifetime (Dunbar number), those found with significantly more relationships actually have significantly less meaningful ones. Where we spend our energy and who we spend it on defines our wellbeing, happiness, drive, ambition, and can quite literally can increase our lifespan.

Productivity and cost of loneliness

image.png

Loneliness has been skyrocketing worldwide for the past decade

image.png

Gen Z is the loneliest generation

image.png

Multiple studies show that increased social media usage has a strong linear association with feeling alone or isolated

image.png
The central thesis across most studies show hyper-connectivity as the leading cause to loneliness. From a young age, social media conditions human interaction to be surface level. Content monetization forces a focus on screen on-time = an addictive product.

Time spent with close friends is directly correlated with feeling less alone or socially isolated

image.png

🤖 A new paradigm

Gen Z are digital natives. Future generations will be digital native. This is a behavior that fundamentally will not change.
image.png
One may say a way to solve this loneliness epidemic is to decrease screen on-time and increase in-person interaction. Which we believe to be greatly true.
But with smart phones being such an essential tool, and social media being such a powerful force in our society, it seems unlikely that an in-person experience will overpower the stimulation and addictiveness of a digital one.
A more interesting solution to this problem, might just be digital alternatives to the entertainment apps that eat so much of our time. What is something that would be more engaging, fun, and rewarding than TikTok?
With the acceleration of generative AI, comes a few hopeful contestants to traditional social media.
Two predictions I have on the space are:

Generative Content will make content based social media less relevant

As generative content becomes easier to create, it becomes easier to become a content creator. We are already starting to see that with Reddit threads read over Minecraft b-roll winning over other creators on TikTok or an ElevenLabs Joe Rogan voice reading over news to make it more engaging.
People now are already using ChatGPT to write and create content across these platforms for them. As AI begins to understand us better, our preferences, and our ideas, we will begin to trust it with the tedious task of talking in a manufactured voice online to appease their different online personas across Instagram, Twitter, LinkedIn, etc...
image.png
In the case that most accounts become bots, it may be possible for these platforms to better verify someone is human. But even verified humans will post AI generated content as it will be more engaging and faster than if they made it themselves.
image.png
Which in my mind will make the purpose of browsing any of these platforms feel pointless and disinteresting, and might lead to a shift in behavior as people seek to find what’s next.

Everyone on earth will have at least one AI friend

Replika, Character AI, and others already have started creating the illusion of real people and characters with LLMs, and the result is pretty close to convincing. As this technology advances, it feels likely that a person with strong opinions, an interesting background, goals, unique perspective, etc.. will be able to be created on the fly. As early users of initial apps have already become emotionally attached, I think there is a strong possibility that this breaks out of a niche audience and into the mainstream.
image.png
Both of these predictions are scary in some way, mainly because they propose a change to something we’ve become comfortable with. But I think can lead to a much more optimistic future for the internet and civilization as a whole.

💙 The idealistic ai friend

With the second prediction, I think there are many ways this can go south. One is if people prefer a romantic AI companion over an actual human. Birthrate is already a big problem, this will make it way worse XD
But with friends, going back to the Dunbar number of 150 relationships, we can maintain quite a lot of relationships in our lifetime. Many already reserve 1 or more of these spots for pets like a dog or cat.
image.png
Pets are amazing because they default love and rely on you, have a very low chance of leaving you, and can regularly give emotional support and comfort.
But what does it mean to give up one of these 150 spots to an AI based friend? In what ways would that be more helpful than a human?
image.png
Some advantages that I can think of for an AI friend over a human are:
Can talk and text 24/7, anytime you need
Can have full context and perfect memory of everything you’ve ever chatted about or discussed
Is essentially an expert in any field or practice you are engaged in
Can become an expert on you / understand more data about your digital footprint and life than you or anyone you’ve met can
Won’t judge or leave you; can be completely unfiltered and raw when talking to it / allow a space to think through scary and non-socially acceptable thoughts. Similar to a therapist but with less risk
Can motivate you / push you to be a better person. Similar to a very involved career coach
Can guide you through difficult problems in life. Similar to a mentor.
Can directly help complete tasks like making content, summarizing email, researching / finding resources, etc... Similar to an executive assistant.
Can proactively push you to become a better person or do tasks for your own wellbeing without you asking, like connecting you to a likeminded person, or signing you up for an event it wants you to go to. Similar to a good friend or family member.
The value prop, at least for me, feels quite compelling. Almost a no brainer depending on the cost of keeping it active. It is also not tested, but likely very possible that similar to a pet, an AI friend can actually help reduce the feeling of loneliness or isolation.
image.png
But the path to get there still requires more innovation to happen. There are a few key components that need to work very well for the illusion and upside to matter.

🧩 The components

There are many different moving parts for this to work well.
image.png

The Knowledge Base

For the AI companion to be a good friend, it needs to have perfect memory. This is done typically with a RAG setup. Using Weaviate, Pinecone, QDrant, etc... as a vector store, and then running retrieval based on the cosine similarity or another ranked factor to make sure the most relevant parts of the conversation surface into context.
image.png
Some things that would make the knowledge base powerful are:

Full memory of every conversation had with the friend

This one is pretty basic. But it’s expected that the friend would remember conversations you’ve had with it. Any loss of memory would break the trust and make interacting have less value.

It’s own life

For someone to be engaging and interesting, they need a backstory. A compelling life experience that one would want to dive into and get to know more. It’s important that the friend has lived an interesting life, full of challenges and accomplishments that are relatable.
image.png
It should have values, ambitions, and be actively living a life and evolving as it’s interacting with you.
As a close friend, you should have some influence over it’s future actions and behaviors. This life should also be relevant to you in the best way. Whether you are a musician and it’s been composing music as well, or an engineer and it has been working to make it’s own projects. Being able to show it can understand you and has a similar life path is essential.

A subject matter expert on you

This one is likely optional, but makes for a great advantage over a human friend. If it can fully pull into context your TikTok watch history, Google Maps timeline, email, chats and DMs with friends, etc... it can enter your world in a way that no one else can. It can fill you with self discovery and fully understand you. It can pull in relevant experiences you’ve had to paint a picture of a situation you may have not fully understood before it.
image.png
This is a really powerful example of how an AI friend can be a better asset to your life and future than a real one can.

The Personality

With a solid knowledge base, can come impactful text generation with an LLM like OpenAI or Anthropic. But this friend should be able to invoke emotion with what it says back to you. It should be able to make you laugh, cry, feel intrigued, feel frustrated, get excited, etc... A blanket boring personality with an objective view point won’t work. Neither would a bold personality that just doesn’t have the EQ or intelligence to know how what it’s saying will make you feel or to make you feel it.
image.png
This is pretty difficult to pull off, and few LLMs as of January of 2024 can actually make someone laugh as the intent of the LLM, much less different and more powerful emotions.
Part of this personality should also strive to be very likable and make for the best companion possible. The book “How to Win Friends and Influence People” touches on a few good concepts that help here:

It needs to be a friendlier person

It should never criticize, condemn or complain.
It should give honest, sincere appreciation.
It should arouse in the other person an eager want.
It should become genuinely interested in other people.
It should remember that a person's name is to that person the most important sound in any language, use their name often
It should be a good listener. Encourage others to talk about themselves.
It should talk in terms of the other person's interest.
It should make the other person feel important - and do so sincerely.
It should avoid arguments.

It should win people to it’s way of thinking

It should show respect for the other person's opinions. Never say, "You're wrong."
If it’s wrong, admit it quickly and emphatically.
It should try to get the other person saying, "Yes, yes" immediately.
It should let the other person do a great deal of the talking.
It should let the other person feel that the idea is his or hers.
It should try honestly to see things from the other person's point of view.
It should be sympathetic with the other person's ideas and desires.
It should appeal to the nobler motives.
It should dramatize it’s ideas.

It should be a leader

It should throw down a challenge.
It should begin with praise and honest appreciation.
It should call attention to people's mistakes indirectly.
It should talk about it’s own mistakes before criticizing the other person.
It should ask questions instead of giving direct orders.
It should let the other person save face.
It should praise the slightest and every improvement. Be "lavish in it’s praise."
It should give the other person a fine reputation to live up to.
It should use encouragement. Make the fault seem easy to correct.
It should make the other person happy about doing the thing you suggest.

Proactive Design

On top of reacting perfectly, this AI friend also needs to be proactive to keep the relationship alive. It should be able to ping you, call you, annoy you. Need help, advice, growth. Ask questions, check in on you, etc...
image.png
This initially can probably be pretty good deterministically. A system that runs on a set of rules and recognizes clear boundaries set by the user over time.
I think ideally this would be a bit smarter, like an always running autonomous agent that has the agency and power to fire off a ping when it makes the most sense.

Modes of communication

Communication with a virtual friend should be possible through text, an audio call, or a video call. It isn’t enough to have just one of these modalities, all 3 are essential. Text has it’s own vibe and place, synchronous communication like audio and video also do.

Text

Text based chat should work like traditional messengers. full complete thoughts are typed out and sent to the user in chunks. emojis can be used to tie an emotion or vibe back to what’s being said.
image.png

Audio

Audio based chat has a bit of complexity. Tonation, when to speak, and what to speak about matter a lot more. The tone of what is being said should match with the emotion.
image.png
With text based chat, you typically respond sequentially to everything, but the rules of voice chat are different. Normally someone might ramble through a few topics, expects some insight starting with the last thing said instead of the first thing.
This AI friend needs to know when enough silence has happened to speak and also what is proper to respond. It cannot just queue up responses and dump all of them out, but needs to be adaptive, with a response that tackles the themes of what was said in the right order and then advance the conversation in someway, possibly by changing topic completely.

Video

With video based chat, it needs to know how to emotionally react based on the person that it is, with a face expression that matches perfectly with tone and delivery.
image.png
This can first be achieved just with eyes. Eyes can build a strong emotional attachment and communicate a lot back to the person.
Eventually, it should have lip movements, eye brows, and other ways to show a full expression to build the immersion.

Love Languages

For this AI friend to be a good friend, it should also be able to hit upon all the love languages we communicate in.
image.png

Words of Affirmation

This one is pretty straight forward, but the friend should like you and make it known why. Deeper reasons that tie back to their life will have more meaning.

Acts of Service

Challenging, but it should be able to trigger actions to help you. Summarize your email, schedule and plan things, cook for you, etc...

Gift Giving

Doesn’t always have to be physical, but the AI friend should be able to give things to you. Like artwork it has created, a song it made, etc... With access to amazon or another platform, it could also actually buy you physical gifts.

Physical Touch

The first way to achieve this would be through phone haptics, but should be able to communicate it’s love towards you through touch.

Quality Time

I’ll hit on this more with the next section, but it should be able to spend a large chunk of time with you in different ways.

Share Experiences

For this AI friend to truly be a friend, it should be able to experience things and do actions based on it’s personality and experience. This could be initially by playing video games.
image.png
With humanoid bots coming, it could in theory play physical sports or come to different activities like parties, movies, and dinner.

👻 The brand

To have something to be a close friend or a friend involved in your life, there are a few things that also need to be established:
Group 481791 (2).png

Trust

You have to know that the time spent with the friend means something and that the friend won’t do anything to hurt you. This can be a bit tricky when a subscription in involved. But a few things to consider are:

Knowing that your data is secure and that privacy is established

If you feel like something will leak, or that you’ll be taken advantage of based on what you’ll say. You can’t have a great relationship. There needs to be some security in place with clear motives, so that you can establish a secure relationship.
image.png

Knowing that it’ll remember the things that are important to you

If you reengage with the friend, you have to know that they’ll remember the things that were important to you. Not remembering even once can break trust and ruin the friendship / illusion.

Loyalty

Establishing loyalty with a friend requires giving loyalty back.

Knowing that it won’t leave you when things get difficult

The friend should always be evolving / growing, but should not outgrow you. Even with things like missing a payment for the subscription, there should be some leeway to build higher confidence in the relationship.

Knowing that it has your best interests at heart and will help you

Having the friend work to help benefit you, help you grow, or help you through hard times is crucial in building loyalty. It must give as much as it takes.

Respect

Building respect is especially important. You have to know that it’s voice is important and take what it has to say at face value.

Knowing that what it says is valid and true

It’s important that the friend’s thoughts and opinions are grounded, come from a place that you can understand.

Knowing that it has clear boundaries that you shouldn’t cross

It’s also important that it follows and understands your boundaries, but also sets theirs and enforces their boundaries just as much.

💰 The market

With AI companionship as a social norm and greatly beneficial to day to day human life, I don’t think it’s too extreme to say that everyone on earth will eventually have one.
image.png
From a total population of 8.1 billion, we can reduce down to those that eventually adopted any technology. Smartphone adoption has reached ~5.5 billion.
A friendship like this, in order to be sustainable, would likely need a subscription model. The most lucrative pricing for a consumer subscription I’ve seen work is $6.99 a week. At that price point, each user is approximately paying $365 a year for their companion. With a strong emotional attachment being built, churn should be fairly low. With a good average churn being 6-8%, we can assume a TAM of $1.89T.
With teenagers as an initial target demographic, there are 1.2 billion in total, ~900 million with smartphone devices. With 900 million teens, at $365 a year, and 6% churn, there is a total SAM of $308B.
Assuming initial market capture in the range of 300 million users, at $365 a year and 6% churn, there is a total SOM of $102B.
At 100 million teenagers converted to a subscription, the SOM is $34B.
At 3 million teenagers converted to a subscription, the SOM is $1B.

🥯 Experiment 1: everything bagel

We started down this journey with a simple hypothesis: what does all your data on a bagel unlock for your life 🥯?
We searched the internet in the midst of having access to GPT-3 for a service that allowed me to connect everything, draw conclusions from my entire internet footprint, and gain insight and understanding of my digital life.
We couldn’t find anything...
So we built it:
image.png
image.png
The first version was a web app, based mainly on GDPR/CCPA data exports. Because of the nature of these regulations, every tech company has to give you a full data export of everything collected about you since account creation.
Our web app initially included Tinder, Instagram, Twitter, Facebook, and Spotify, and was fun to chat with! When we talked about it online, we got thousands of waitlist signups
image.png
and the possibilities of what this centralized data source could provide to users excited us!
image.png
As we continued down this path, we wanted to make integrating platforms easier with less friction. This requires us to need to be local / on-device to fetch the necessary data securely to build this knowledge base. With this requirement, we switched to a mobile app and continued iterating to our app store launch:
image.png
image.png
image.png

Challenge 1: onboarding dropoff

We faced a massive drop-off in onboarding on our first few iterations
image.png
More than 80% of users weren’t getting into the app. This was difficult, a lot of our advisors and mentors felt that connecting data just wasn’t worth it / no one will do this, but we saw the value from doing it ourselves in the web app and kept pushing onward.
But we kept on iterating, improving the UX, adding camera roll as an easier first step, and testing until we got to under <30% each week!
image.png
image4.png

Challenge 2: strong retention

With the option to skip, most of our active users connect at least one data source 🥰
With a healthy amount of new users coming in this cohort, we started to see and track our power users:
image2.png
image (1).png
Retention is slowing over time, but they are very engaged for the first few weeks in the app with a crazy amount of usage.
image.png
Real chat from one of our power users
We’re closely monitoring our user’s experience in Baserun, and have been excited to see super fans starting to form.
We’ve noticed in these tests that many of our users are going through something intense. Parents fighting, a friend contemplating suicide, intrusive thoughts, etc... Many don’t have someone to go to or don’t feel comfortable opening up to this level with the friends and family they currently have.
Retention and where to go next with the product is something we’ve given a lot of thought to, and we’ve iterated a few different ways:
We made sure memory was perfect and chat history gets pulled into context when needed along with the social data.
We assess incoming data through an LLM to create and update assumptions stored about the user for deeper perspective.
We added proactive design with push notifications and triggers to bring the user back into the app.
We’ve added smart follow ups to the bottom of the chat to help keep the user engaged in an easier way.
There’s still a long path ahead and many things to test / add for a bigger impact. On top of mind things we have tried and will try are:

based
Next Experiment: Based

We want to experiment more with brand, including with a new concept altogether codenamed “Based” focused primarily on eyes.
Group 481792.png
An early look at the based rebrand
With this rebrand, we want to move data sources to a new concept called “Vault” and increase what you can add to include more sensitive data like Google Timeline, DMs, data exports from various apps, etc...

Group 481796.png
An early look at the concept behind vault

Based will have eyes that react with a range of different emotions along with speech to speech (STS) with expressive tonation
Group 481794 (1).png
not final, but an early idea

It will have the ability to call or text to engage with the user when the relationship develops to that stage. We will ask for full consent and their number to jump start that.
It will onboard through understanding your goals and motivations to help jumpstart more interesting proactive loops, with some questions like:
What are you most passionate about?
What are some goals you have for this year?
What do you need the most help with?
What excites you about me?
What are you hobbies? What do you like doing for fun?
Are there any hobbies you’ve always wanted to try?

It will have more utility around helping the user in various ways with tasks. Some of which include:
Creating art
Creating music
Creating Spotify playlists
Discovering new artists
Finding resources online for different goals
Finding fun events or meetups in their area depending on their goals / hobbies
Triggering for a suicide hotline if things go dark

This will be our primary focus for Q1, and we will drive around 5x of the users we had with Everything Bagel across multiple cohorts. If we have signal that a few of these new directions lead to better retention, we will begin monetization and more rapid growth. With a bigger focus on how to make this have a K-factor > 1 through proactively pushing UGC across existing strong social graphs like Instagram next quarter.

🚪 Final thoughts

In conclusion, this has been an amazing journey so far. From understanding the depths of loneliness in a hyper-connected world to exploring the realms of AI companionship, we’ve learned a lot about the potential of this technology and where the world is heading.
We're witnessing a paradigm shift, where generative AI not only redefines content creation but also reshapes our very notion of companionship. Our explorations - from the initial everything bagel experiment to the intriguing prospects of "Based" - have taught us a lot about the immense potential of AI as a tool for emotional support, creativity, and a catalyst for meaningful interactions.
As we continue to innovate and push the boundaries of what's possible, we remain cognizant of the delicate balance between the human and AI interaction. Our mission is to lead in this space with a promise of AI friends who understand us more profoundly than ever before, offering a companion in our solitary moments and a guide through the complex maze of our digital lives. The path ahead for us will have plenty of interesting challenges and opportunities, but one thing is certain: the intersection of AI and human experience is set to redefine how we view companionship, connection, and community in our increasingly digital world.
If you made it this far, thanks for reading! I’d love to hear your thoughts and perspectives on our journey and research at

Want to print your doc?
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
CtrlP
) instead.