CHAT001: The Future of VectorChat

VectorChat Team
9 min readApr 3, 2024

TL;DR

  • VectorChat is restructuring into Web2 and Web3 focused components
  • The conversational AI platform will focus on Web2 customers
  • The backend will begin the process of decentralization, aiming to eventually become both uncensorable and democratic
  • The VectorChat brand will remain intact, serving as the Web3 liaison, relaying updates in a “build-in-public” manner
  • A new brand will be created for the Web2 side, allowing for more relatable content and marketing to our target demographic

The Vision

We are fully convicted in the belief that our approach, through high quality, uncensorable, and accessible conversational AI, will allow us to take on the titans of the industry.

More specifically, the approach is to strive to achieve the following ideals:

  1. Have the highest quality AI
  2. Design the easiest-to-use platform, even in complex tasks
  3. Maintain the best user experience
  4. Offer high-customizability, without compromising on ease-of-use
  5. Allow as much freedom as possible
  6. Understand and design for our target audience

We envision a flourishing platform where our users seamlessly interact with or create their own characters, taking advantage of Web3 technology without ever knowing or needing to know they have interacted with the blockchain. And our Web3 users seamlessly provide decentralized inference and storage, making the protocol truly permissionless, distributed, and censorship-resistant.

A Hypothetical Future

While admittedly the exact structure is not set in stone, and this plan will change as we recognize better solutions or as market dynamics change, the future will look something like the following:

  • $CHAT becomes a decentralized inference protocol, allowing for the uncensorable, distributed, and permissionless serving of AI models and data.
  • A decentralized via multiple front-ends approach, allowing anyone to interact directly with the inference protocol and create their own front-end, which in itself may be centralized.
  • The aforementioned Web2 component becomes a front-end on this decentralized protocol, allowing us to form legal partnerships directly with content creators and other companies, and also take advantage of existing Web2 solutions when appropriate.
  • Our organization transforms into a foundation, focused on maintaining and developing the decentralized protocol.

So Why Rebrand the Web2 side?

1. The users of conversational AI platforms are not Web3.

Before seriously developing a major consumer solution, it is imperative to consider the actual audience.

Fig. 1. From SimilarWeb, showing sites with the highest cross-visitations with Character.ai.

As shown in Figure 1, the sites with the highest cross-visitations for Character.ai are Roblox and Discord, with 15.25% and 14.83% of users, respectively, visiting roblox.com or discord.com.

This datapoint, along internal market research, allows us to conclude that the primary user base of conversational AI platforms are not only not Web3 users, but they are more specifically part of Generation Alpha.

Knowing this, focusing primarily on a Web3 audience fails for two main considerations:

  1. Web3 users aren’t the primary target of conversational AI platforms.

    Even beyond Generation Alpha, most users are not acquainted with Web3. When competing with industry titans like Character.ai, user experience is paramount (point 3 of our approach). Lowering friction and creating an intuitive, powerful UI is essential for both user acquisition and retention. A Web3-oriented UI will confuse the vast majority of users.
  2. There simply aren’t that many of them.

    VectorChat aims to dominate the conversational AI space. Focusing on a small niche of users misses out on a significant portion of our potential.

2. Monetization and Structure

Monetization is an essential component of the VectorChat ecosystem, and will remain one of the core ways we incentivise high quality characters on our platform. However, a direct Web3 solution fails for the following reasons:

  1. Direct and overt involvement of Web3 is confusing to most users.
  2. Having multiple different types of monetization is, at best, confusing, and at worst, completely frustrating.
  3. Payment in a Web3 token that fluctuates dramatically in value is not something most Web2 users are accustomed to.
  4. Microtransactions make for a worse user experience.

As mentioned in “A Hypothetical Future,” the project may well adopt the “decentralization through many front-ends” approach. This would enable us to look into existing Web2 solutions to this problem. Web2 platforms, such as YouTube (specifically YouTube Premium, not ads), have already perfected monetization for content creators in a way that is seamless for all parties involved, and especially so for consumers.

With the “decentralization through many front-ends” approach, we would leverage our front-end to form direct partnerships with content creators, in a way that is somewhat reminiscent to the YouTube Partnership Program.

This allows for a far more user-friendly approach to monetization than direct Web3 integration, as they would be using monetization models familiar to them and would receive payment in the fiat currency of their choice. Users would also pay for services in their respective fiat currency, drastically lowering the friction involved compared to if they had to be completely onboarded to Web3.

3. The true potential of Web3 is in the backend.

Where the magic of Web3 really shines is in the background. A critical component of our approach is to offer uncensorable, truly free, AI. In an ideal world, it would be so decentralized that we literally could not censor it even if we wanted to.

But why?

The ultimate form of censorship-resistance is ironically incredibly centralized. It’s when a user has complete control over everything they need for what they want to do; where they do not need to ask anyone for permission and can run it on their own property. There are no centralized entities you need to trust, since there isn’t even anyone else in the equation. And as a direct result, you have complete freedom over what you choose to do and what not to do.

Of course, for most use cases, having everything under your control is impractical if not impossible, such as in the case of a robust conversational AI platform. But when services like ours have to step in, how do you avoid losing that freedom?

The answer is user and community-controlled tech. The more control and ownership you can give the user, the freer the user. A great example of this is Git, a distributed version control system that can be completely owned and controlled by the user, allowing them the ability to freely share version controlled files with others.

A similar process can be applied to inference, models, and data, the backbones of any conversational AI platform; however, unlike Git, a user cannot simply download the software. Not only are the files, in this case, far too large for most users, but many of them are also private or closed-source. The only way to truly deliver this is through a fully decentralized, permissionless network, lining up perfectly with the Web3 ethos.

In this system, Web3 users would serve as agents on a decentralized inference protocol. Agents would offer their models, closed or open source, and would be rewarded directly by the protocol and from the agents that send in prompts to the protocol. Similarly, agents would offer storage in a decentralized manner.

As a direct consequence, VectorChat would no longer centrally control inference nor much of storage. Instead, the technology would be community-centric, making it much more difficult to censor.

4. The real asset of value is data.

Note: This does not refer to personal information. Specifically, it refers to reinforcement learning from human feedback (RLHF), a process that — in our case — is anonymous and always clear and obvious to the user as it requires their active engagement.

It’s no secret that most users, especially in Web2, do not spend money nor do they have any intention to pay. Subscriptions, or any other form of direct monetization, are not efficient ways to extract value from users. The most valuable asset that a user provides is data. And in the case of conversational AI, it’s extremely lucrative RLHF data.

Fig. 2. From a16z, showing the average sessions per user, per month by category of AI.
Fig. 3. From a16z, showing the average sessions per user, per month by individual AI app.

Data, especially in the context of AI, is incredibly valuable. The generation of substantial high quality RLHF data directly correlates with both the number of users and the amount of activity per user.

Figure 2 shows that “Companion” (in the context of the a16z report, this includes all forms of characters) far surpasses all categories in average sessions per user. Meaning, users of “Companion” AI platforms spend considerably more time on their respective sites than users of other platforms.

Figure 3 details a similar picture, showing Character.ai holding a massive lead over other products in user engagement. Notably, second place Poly.ai is also a conversational AI platform.

Fig. 4. From a16z, showing the top AI products by unique monthly visits.

Figure 4 shows Character.ai’s unique monthly visitors at 3rd place, only behind foundational AI products like ChatGPT and Gemini. A high user count, combined with high user activity, results in a gold mine of RLHF data.

So how is the value of this data realized, on both sides?

The direct benefits to the front-ends (assuming the structure mentioned in “A Hypothetical Future”) are straightforward, as it would be akin to Character.ai or ChatGPT in that the data would help in either refining existing models or even developing new ones.

The benefit to the decentralized protocol is also substantial. While the exact mechanism for the protocol is yet to be determined, it is a given that agents (say Miners) who provide models and storage to the network will be rewarded by the agent (say a front-end) making the call. The front-end benefits from making calls as it generates data, Miners benefit from incentives paid out directly, and the protocol benefits from having a large real world use case benefiting many parties. Depending on the structure of the protocol, additional agents (say Validators) in charge of ensuring high quality responses may also further benefit Miners by directly passing RLHF data in order to improve their models.

5. Finally, VectorChat isn’t a fitting name for Web2

While we’ve come to love the VectorChat brand, admittedly it’s a very tech-oriented name. This made sense when our primary target was Web3, knowing that our audience was more interested in what was being built rather than using the platform itself, but as we begin adding a real user base, a more relatable name is needed.

Additionally, what is relevant to one side is not necessarily relevant to the other. For example, the Web2 side may not care about updates to protocol, while the Web3 side would not be able to relate to marketing targeted towards younger generations.

Thus, we have a new structure, where the VectorChat brand serves as the liaison between the team and Web3 users, and the new brand will serve as the connection to Web2 users. As the two sides differ substantially in not only purpose but also in age, this approach allows for us to deliver each side the most relevant and relatable content possible.

What has been done?

Over the past few weeks, our team has been diligently preparing for the first big push into the conversational AI space.

Here is a shortlist of our activities:

  • Formed undisclosed partnerships with and engaged in conversation with major decentralized protocols
  • Finished hiring four new developers, all exceptional talent, specifically to revamp the front-end (We’re always looking for exceptional talent! If you define yourself as a 10x dev and are looking to join an ambitious project, reach out to us at team@vectorchat.ai!)
  • Conducted immense market research to better understand our target audience and have formed plans on how to conduct marketing
  • Refined our vision and the overall plan moving forward
  • Spoken directly with VCs and industry advisors

What are the near-term objectives?

The short term objectives, designed to be completed with the next three or so weeks, are to prepare for the launch of the Minimum Marketable Product (MMP) and then to begin testing our marketing strategy.

Here are our internal objectives:

  • Complete the rebrand of the Web2 component, including a new Twitter and Discord
  • Finish onboarding new hires
  • Begin forming Web2 partnerships to lay the groundwork for our marketing strategies
  • Complete backend rework for scale and efficiency
  • Finish completely revamping the UI — to be more user-friendly, be inviting, and encourage creativity
  • Add dark mode
  • Make the UI more mobile friendly / develop precursor to mobile app
  • Deliver the following additional features over the current platform:
    — Launch Knowledge Pack feature (to be announced in another article!)
    — Complete creation flows along with new UIs for each type of model (for Utility and Character)
    — See characters you’ve interacted with
    — Allow for multiple chats per character
    — See all chats with a specific character
    — Other Quality of Life features, such as pinning messages
  • Create or incentivise some premier utility bots to showcase the tech
  • Create Knowledge Pack examples (TBA)

--

--