In yesterday's article, I shared the project dphn.ai with everyone.

Another project closely related to this one that has recently stirred up a wave in the Base ecosystem is venice.ai.

Venice was launched on the Virtual platform during its peak popularity. I had some impressions of it before and after its launch, mainly for two reasons:

First, various media reported on the well-known background of the founding team;

Second, the project generously gave airdrops to participants in the Virtual ecosystem at the time of its launch.

However, regarding the most important core of the project: what exactly it does, I didn't have a deep impression at that time.

In fact, to be more precise, based on my understanding of AI at the time, even if I had seen the project introduction, I might not have fully understood the business model of the project.

However, after two years of rapid development in AI and the popularization of AI knowledge through various channels, looking back at this project now, I have a clearer understanding of its business model.

We typically use large models in two steps:

The first step is to train the large model to enable it to possess "intelligence." This step is what companies like OpenAI and Anthropic do in laboratories. This step is unrelated to ordinary users, and ordinary users will not engage with this step.

The second step is for users to actually use the large model. In this step, ordinary users directly interact with the large model, such as by asking questions or sending requests. When the large model receives our questions and requests, it processes them and provides answers; this process is called the inference of the large model.

What Venice does is act as an intermediary, forwarding our requests to the large model to help us process our questions and requests, and then forwarding the large model's responses back to us.

Venice is an inference intermediary.

Here, some readers might ask:

Why do we need an intermediary to forward our requests instead of sending our requests directly to the large model?

I previously wrote about a personal experience that made me very uneasy:

After interacting with a certain large model for a while, it surprisingly guessed other information about me that I had never directly told it.

Why did this happen?

Because the large model could have guessed my private information based on my login information, associated email, and the structural relationships of various questions.

Therefore, as long as we communicate directly with the large model, all the large model tools currently available on the market may pose a risk of privacy leakage.

What Venice aims to solve is this problem:

We first send our questions/requests to it, and it uses zero-knowledge proof methods to erase any potential personal information hidden in our questions/requests before sending the processed questions/requests to the large model.

This way, the large model cannot associate our questions/requests with our personal information and thus cannot invade our personal privacy.

In addition, it has another function:

We know that the two major models in the United States strictly ban users from Mainland China. However, through Venice, users from the mainland can bypass that ban.

This project has seen significant user and revenue growth over the past year, which has drawn attention from many in the crypto ecosystem.

According to recent estimates, its annual revenue has reached $48 million, with net profit ranging from $6 million to $13 million.

The current fully circulating market value of its token is $1.1 billion.

These are its relatively rough financial details.

For this project, I find two points particularly interesting:

First, from various statements made by the project's founders, one can sense their almost religious devotion to pursuing privacy and security.

Second, this emerging AI + Crypto project has already achieved an annual revenue of $48 million, with net profit conservatively estimated at nearly $6 million. This is genuine cash flow revenue driven by real user demand.

However, the project's problems are quite obvious:

First, the important highlight of the project, "zero-knowledge proof" technology, does not have particularly high barriers to entry. I am very skeptical that if it continues to grow, it will inevitably attract a plethora of competitors.

In such a case, where is its moat?

Can the first-mover advantage become its moat?

Second, an important aspect of this project that uses crypto assets is allowing users to use crypto assets to pay for inference requests. This effectively protects users' privacy. But can it also expand the use of crypto assets further to highlight their privacy characteristics?

Otherwise, relying solely on this characteristic will be hard to establish a long-term moat and unique features.

Third, I wonder if this tool is widely used for distillation, will this project one day also be banned by several major models in the United States?

If that happens, the performance and effectiveness of this tool will be significantly reduced.

How will it maintain its advantages then? Can the effectiveness continue?

Like the dphn introduced yesterday, even though I have many unresolved questions about Venice, at least they are both paving the way for real uses and landing scenarios for crypto assets with a workable business model and real user payments in the AI + Crypto space.

This is worth continued observation.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。