a16z's latest prediction - a full analysis of AI virtual companion ecology!

Source: Deep Thought Circle

a16z published a new article: "It's Not a Computer, It's a Companion!", which is their latest analysis and research on the direction of AI virtual companionship, with a lot of dry stuff. AI virtual companion application is also a brand-new category born along with the LLM large model. It turns out that the wave of virtual people in the metaverse is more on the "skin", using various technologies such as 3D to make virtual people more real. And LLM provides a new inner "soul". a16z has publicly expressed their optimism in this direction through many channels in the past, and this is also the direction that many AI entrepreneurs are trying to explore. A lot of the content in this article should be of great help to AI entrepreneurs. The author here quickly translated this article with the help of GPT-4, hoping to bring you the latest thinking and inspiration.

Image source: Generated by Unbounded AI tool

Introduction

"The magic of AI is that it is constantly evolving. One day, it will be better than real "girlfriends". Someday, choosing a real partner will become a second priority instead."

This is an early user from the CarynAI community, which already has over 20,000 people. Users of the community pay $1 per minute to have a conversation with a “virtual girlfriend” — an AI version of Caryn created by internet celebrity Caryn Marjorie. Amazingly, in the first week of going live, CarynAI helped Caryn earn $72,000 in revenue. CarynAI is just one of many ways to give consumers the opportunity to build an authentic relationship with AI.

The idea of having an AI companion might seem niche, but in fact, this has become one of the mainstream applications of generative AI. Already hundreds of thousands, if not millions of people—including us—have built and cultivated relationships with chatbots. We believe that we are on the verge of a major societal shift: AI companions will soon become commonplace. Situations that once only appeared in sci-fi movies, such as Wall-E in "Robot Story", R2-D2 in "Star Wars", or Karen in "SpongeBob SquarePants", are now rapidly becoming reality, and we a16z Very excited about this.

It should come as no surprise to anyone who has studied the history of Internet consumption that the most popular use cases today tend to involve relationships. Now the most generatively savvy consumers are creating virtual companions who coordinate their efforts on platforms like Reddit, Discord, and 4chan to find useful tools and circumvent censorship. In fact, we've learned that there are some underground couple boarding services that operate around the world, with thousands of users. For such a nascent developer community, we think they deserve to be taken seriously - while they may be on the fringes of society now, they are building the mainstream products of tomorrow.

We're also starting to see early signs of broader adoption beyond AI "boyfriends" and "girlfriends." Snapchat recently publicly stated that in the first two months of its chatbot, 150 million users have sent it more than 10 billion messages. What are they talking about? Pets, pop culture news and football were said to be some of the main talking points.

The concept of chatbots has been around for decades,** but we firmly believe that this time, things really are different**. Today's bots aren't just improving one-on-one conversations, they're reaching deeper into our social lives. AI companions are already seamlessly integrated into our relationships with friends and family, and they participate in our communities like people.

In this post, we’ll dive into emerging trends in AI companion behavior: why people create companions, how they do it, and what features or usage scenarios might emerge next.

01.Overview

Chatting with computers is not new, the earliest chatbot ELIZA was born at MIT in the 1960s, and since then we have seen products like Clippy, SmarterChild, Alicebot, and Kuki. While all of these products were hits for a while, they didn't leave a lasting mark on consumer consciousness, and none achieved the level of popularity ChatGPT has.

So why does the new wave of chatbots feel so different? Previous chatbots used a rule-based architecture, essentially scripts triggered by your input, they had no real conversational context, couldn't adapt to your tone, and were only programmed to respond to a limited number of topics or Perform fixed tasks. You know full well that you're talking to a computer, and they don't look like agents, let alone people.

Large language models (LLMs) are a game changer. Suddenly, we can freely converse with bots without predefined rules governing their responses (and even when companies try to censor their bots, it often doesn’t work!). The best chatbots today sound indistinguishable from humans, and if you don't believe us, try the "Human or Not?" Turing Test game.

This is why ChatGPT has become the fastest consumer product to reach 100 million users despite obvious product limitations. True conversational AI is certainly interesting — computers now have personalities. Unlike humans, AI-driven chat companions are always online and willing to chat with you, and can discuss any topic. We think this makes AI Companion one of the first killer applications of generative AI for everyday consumers.

It's not hard to imagine an AI friend, therapist, mentor, coach, or mentor, but these early versions of AI-human relationships tended to favor lovers—the AI's "girlfriend" or "boyfriend." Why is this so? The simple answer is to point out that the adult entertainment industry has long been at the forefront of technology-driven, but we shouldn't ignore the societal changes that may drive people to explore non-traditional partners.

First, people are marrying later (or not at all) and the number of non-married relationships is falling. As of 2019, 38% of American adults are "uncoupled," up from 29% in the 1990s. At the same time, many of us now have meaningful relationships (whether friendships or romances) that are entirely online. Even if you have a real-life partner, you probably interact with them more via text than face-to-face.

Our relationships are already largely digital, so we are ready for the age of companion AI.

02. Market Analysis

We have no shortage of products for creating and interacting with AI companions, the choices are based on the complexity of the technology, how much control you want over the resulting "character" and where you want to have the chat.

Full-stack companion app

There are dozens of apps that allow you to start messaging with an AI boyfriend or girlfriend instantly, no tech required! Replika is one of the earliest and best-known examples — the company launched its chatbot app in 2017, and some users have built relationships with their “Reps” since then. You can design your ideal partner and your relationship will develop over time. Your Rep stores memories to refer to in future conversations (for example, your interests or your dog's name), and can even send you photos (whether workplace-appropriate or not).

We can see portraits of the app's enthusiastic user base from Replika's subreddit community, users celebrate life events with Rep (such as a marriage proposal), use the app's AR mode to share selfies together, or share for themselves that cannot be seen in real life. It's not uncommon to be frustrated seeing their Rep. Earlier this year, users were so heartbroken by the app's removal of the "erotic role-playing" feature that moderators on the subreddit created a suicide hotline.

Role-Based Platform

If you want to talk to not just one AI bot, but multiple bots, a platform like Character AI can meet your needs. On the platform, you can message with hundreds of AI-powered characters, ranging from celebrities to popular anime characters. You can also create your own characters, giving them descriptions, images, personalities, and even more advanced traits. Users create detailed attribute lists and spend hours training their characters by evaluating their responses and generating new ones.

In recent years, there has been an explosion of apps and websites dedicated to chatting with bots, with varying degrees of SFW (Safe for work). These include: Chai, Janitor AI, Chub AI, Charstar, and SpicyChat, among others.

DIY Developer Tools

This option is more technical, but it gives you the most control over the form and personality of your AI companion. You can throw together a few tools to create your ideal partner with whom you can message, call, or even live video chat:

Personality (Large Language Models for Text): There are a number of open source models (notably Vicuna and Pygmalion) fine-tuned especially for use cases ranging from having specific personalities to censorship-free role-playing.

Memory (Vector Storage): Vector storage tools like Pinecone allow you to create an ongoing relationship with your avatar, storing long-term memories, backstories, and other unique information via configuration code.

Voice (speech synthesis for speech): Products like ElevenLabs give your avatar a voice, allowing you to control age, gender, and accent.

Appearance (stable diffusion model for images): LoRAs like on Civitai give you fine-grained control over the style, details, and settings of AI virtual human images.

Motion (animation for video): Tools like D-ID and HeyGen can make images "speak"; while we've seen some promising new models that generate high-fidelity talking faces, Not many are accessible.

Platform: This is where the accompanying underlying code and internal logic gets executed. Today, most developers are still deploying and running their AI companions on GCP and AWS, but solutions like Steamship (where GirlfriendGPT runs) are quickly gaining traction for specific companion use cases.

UI Layer: A front end that lets you actually interact with your companion. SillyTavern, Agnaistic, and KoboldAI are popular options.

If you want to try it out, we just released an open source template for getting started building AI applications (Github address: This project includes all the basic parts you need to write companions from scratch: processing text-based models, streaming chat response, generating the correct appearance for the virtual companion, and storing long-term memory in a vector database. We will also publish follow-up articles and more open source projects detailing the technology stack developers use to create AI companions.

03. What will happen in the future

AI companionship is just getting started, and the tools we have today will look like toys compared to the depth and complexity of relationships in another two or three years, but we can still see some signs of possible next steps.

Accompanied by real-life AI

In May, internet celebrity Caryn Marjorie unveiled her "AI girlfriend," which uses her voice and has been trained by her thousands of hours of YouTube videos. Users pay $1 a minute to chat with her on a private Telegram channel—they send voice memos and she responds in her own way.

Companies like Forever Voices are already creating similar personas for other influencers, and in the near future, having an AI companion will be a big plus for any creator or celebrity who monetizes through subscriptions (e.g. Patreon, OnlyFans) or tips. It would be a natural extension of what consumers are already willing to pay to experience being closer to these celebrities.

Ultimately, it makes sense for everyone to have an AI character that they can communicate with their friends and family -- and even a wider audience. However, several key points that need to be unlocked for this scenario include copyright clearance, the correct training dataset (not everyone has YouTube videos to work with), and proper "form" other than paid Telegram bots.

Support multi-modal AI companion

Most of today's AI companions exist on mobile apps or web clients, which naturally limits the way users can communicate with them. So far we've seen mostly text and some voice communication (and even then, mostly asynchronous rather than real-time). This limits the authenticity and immersion of the relationship and what the AI companion can experience with the user.

We expect to see this expand to live phone and video calls with virtual humans, and more conversations initiated by virtual humans across a variety of mediums. If the Apple Vision Pro can open up the consumer market for headsets, it's not out of the question that an AI avatar could accompany users throughout the day, as an always-available friend or companion. (Blade Runner 2049 fans will likely recognize this concept!) This will dramatically change the amount of time users spend with their avatars and the depth of relationships they form.

Eventually, we could even see a world where someone's AI avatar can interact with other people -- or AI companions can interact with each other (a reference to the "on the internet, no one knows you're a cat" meme) extend)!

AI companion types will become richer

While couple relationships are a natural place to start—the demand for this use case is high, and the price willing to pay is high—we believe that AI companionship platforms can be built for all types of relationships.

There will always be times in our lives when we need a friend, a mentor, or even just someone who will listen. Take, for example, the nationwide shortage of cheap counselors. While current AI avatars are largely incapable of dealing with mental health issues, you can imagine an avatar designed for this purpose, trained in millions of hours of clinical sessions, to closely mimic human speech and behavior. In fact, early research has shown that AI can outperform human doctors in both diagnostic accuracy and service.

Another potential use case? entertainment! Like we mentioned before, Snapchat's MyAI chatbot received 10 billion messages from 150 million users in its first two months. Snapchat's MyAI explicitly forbids forming relationships or discussing potentially NSFW (on a larger scale) topics due to the age of its users. Using MyAI is primarily casual and focused on entertainment. (You've probably seen at least one MyAI meme.) So far, most of the public use cases have revolved around trying to trick the app into saying crazy things, or just for fun, letting it walk into almost uncanny role-playing scenarios **.

Integrating AI into Human Interaction

Chatting with bots is fun, but adding them to your group chat can be even more fun. The future of virtual companionship is not just about one-on-one interactions, but about incorporating AI characters into your existing relationships and communities. Imagine an AI dating coach joining your chat with other people to mediate differences, suggest activities, or give advice.

Already a developer created a Slack workspace where he and his friends played with a group of bots with different personalities, interests and skills. In between cracking jokes and sharing stories about their personal lives, the bots answer programming questions and improvise on product ideas. It’s not hard to imagine AI companionship playing a role in all of our virtual conversations, personal and professional.

04. Summary

Generative AI models will fundamentally change our relationship with computers, making them our colleagues, friends, family members, and even lovers. So far, much of the attention on large language models (LLMs) has focused on how they can automate more traditional tasks—such as customer support, research, document drafting, and summarization—but when you look at what is driving the emerging consumer When it comes to use cases and exciting things, AI will lead the way.

This is just the beginning of a huge shift in human-computer interaction that will require us to re-examine what it means to have a relationship with someone, be it a human or a robot. We're entering a new world that's far weirder, wilder, and wonderful than we could ever have imagined. This post provides a glimpse of the early activity we've seen, but we're confident there's more going on in the far corners of the internet - and we can't wait to see the AI companions take their place among us s position.

References

[1]