VVV Hits All-Time High: Founder’s Perspective on Models, Privacy, and Crypto

Original Title: Why Erik Voorhees Says Crypto Was Really Built for AI Agents: Uneasy Money
Original Source: Unchained

Editor's Note: Venice is a privacy AI platform built on Base that does not train models itself but aggregates mainstream closed-source and open-source models into a single entry point, committing not to retain data for the parts it hosts. Its token VVV is not billed per use but rather earned through staking for daily inference quotas, with platform revenue used for buybacks and burns, continuously reducing circulation. Over the past month, VVV has risen by more than 90%, recently hitting an all-time high of $34.53. In a time when AI inference is rapidly commercializing and token narratives are generally failing, Venice provides a rare example: a consumer-grade product that generates real revenue and demonstrates how tokens can be utilized effectively.
On August 22, 2026, Venice founder Erik Voorhees appeared on the Unchained podcast. He discussed why he chose to incorporate tokens into Venice instead of fully transitioning to AI, the fundamental differences in power perspectives between the crypto and AI circles, how Venice aggregates global models while ensuring privacy, and how it plans to make money in an industry where inference is continuously commercialized. Below is the original podcast content:

Why did you choose to incorporate tokens into Venice instead of fully transitioning to AI?

First of all, I can’t possibly leave the crypto space; it’s in my blood. I care deeply about it because it represents a new way for money to flow around the world. Money is intrinsic to all commercial activities, so I don’t think it should exist in its own ecosystem; it did grow that way initially. I would rather see the best "primitives" in the crypto and DeFi worlds start to leave the crypto space and truly enter the mainstream. If everything is just "crypto people serving crypto products to crypto people," then we’ve really messed up.

So I want to bring some primitives into Venice to show the world that these technologies are not just for speculating on price movements. They are incentive mechanisms: there are many ways to mess it up, but there are also some ways to do it right, and we want to demonstrate this in a consumer-grade application aimed at the mass market.

The world wants inference, and inference is inherently suited for crypto payments. Is that what excites you?

That’s not what really excites us. Crypto can certainly be used for payments; that’s the first layer, the most basic capability. If someone wants to buy AI credits from Venice, they can certainly pay with crypto, but to me, that’s just the entry threshold; it’s no different in essence from saying "crypto is a payment method like Visa or Stripe."

What I find more interesting is allowing people to have a stake in a platform and project. For example, all mainstream AI labs are private (Google is an exception). Because they are private, ordinary people can’t share in their success at all. This is actually creating a lot of anxiety; people watch AI take over the world and create massive wealth while they can only sit there worrying about when their jobs will be taken away. Regardless of whether that impression is accurate, many people think this way. I believe this situation can be avoided.

So Venice has a token; we haven’t deployed all the methods we plan to use yet, but we want users to have a stake in Venice’s growth from the early days, from the very beginning. How we do this must be approached cautiously, but I think it’s a fairer and more interesting way to run a company: users are not just consumers but participants alongside you.

What are the incentive designs that haven’t been deployed yet?

There’s something we haven’t announced or deployed yet that I’ll mention: we want to take a portion of the money users pay us, whether it’s for monthly subscriptions or buying credits, and use it to buy VVV on the market to return to users, provided they stay on the platform for a while. Essentially, you are "buying loyalty," and this gives you the opportunity to educate users; if they like the product and stick around, they ultimately get a stake for free. This will make users more engaged.

This is the approach I prefer: I don’t want to go to users and say, "Hey, we have a token, go buy it," that feels a bit dirty. But if they are already spending money on Venice, and we give back a portion of what Venice earns in the form of project tokens, that’s a really cool rewards program for a significant number of users.

What should be the ultimate purpose of crypto?

People have been experimenting with crypto for so many years that they’ve forgotten that crypto must ultimately serve something else. Bitcoin serves the purpose of base currency; every other cryptocurrency must do something else, existing at another level of the tech stack. ETH drives a smart contract platform; that is its purpose and product. Everything must be tied to the product it serves. We’ve reached a point where "the coin itself is the product," which was never how it should be; it should incentivize and coordinate things outside of itself. If it serves only itself, it becomes completely self-referential.

Moreover, this is a race to the bottom, which is exactly what we’ve seen in recent years: meme coins, and memes for the sake of memes. At that point, you’re not expanding the pie; you’re just moving the pie around. If people want to gamble on meme coins, God bless them; they have that right, and it’s perfectly fine for people to gamble in a casino. But it doesn’t add value to society; it’s a consumptive form of entertainment. Sadly, many projects become meme coins partly because doing something meaningful carries greater risks from compliance and regulatory perspectives. After the 2017 ICO bubble, we spent several years focused solely on meme coins, and legitimate projects with real tokens basically disappeared; DeFi Summer was the last exception. It’s really tragic.

What are the differences in how the crypto and AI circles approach regulation and power?

This is actually one of the reasons I founded Venice. In the crypto world, we have principles; at least we say we do. Some people genuinely strive to practice them, while many just talk about them, but at least we have a set of principles that can maintain a certain loyalty, and these principles are generally shared: decentralization, privacy, individual sovereignty, and the dispersion of power. We argue fiercely over the details, but we are aligned in direction.

In the AI world, which is a broad generalization, it’s completely different. The history of AI is much longer than that of crypto; academia has been studying this for sixty or seventy years. The dominant mindset in that world is that very smart people believe "everything should be controlled"; what really matters is "who is in control" and "what the rules are." It’s very top-down, very centralized in power, with the core concern being "ensuring that the right people are in power."

It stems from that kind of ecology: tenured professors at Harvard, for example; it also comes from how funding has been allocated over decades, and the kind of temperament often found in top smart people, scientific personalities: "I am the master of my field, and I’m confident that there are masters in every field, including society itself." The logical fallacy lies here: they confuse "narrow fields that can be understood" with "complex systems that cannot be understood."

When I started engaging with this area two and a half years ago, no one in the AI circle cared about privacy, decentralization, or user sovereignty. On the contrary, everyone said, "This technology is extremely dangerous; the government must regulate it immediately; everyone must be monitored; your prompts must be scrutinized; answers must be scrutinized; everything must be strictly controlled, and our sin is that we haven’t controlled it tightly enough." That’s a completely different sect, a doomsday worldview. It seems quite strange. However, for ordinary people, what’s strange is actually us; we are the outliers, and that’s the problem; we have been in this field for too long.

What I saw at that time was that AI was clearly going to start taking over the world, with OpenAI and Anthropic holding the cutting-edge models, and they were very top-down, very monolithic, wanting to control everything: "We use a set of rules to enforce safety; the government controls us, and we just do what the government says; society should be organized this way." In my view, this will lead to a very dystopian outcome: machine intelligence itself essentially comes through the official mouthpiece of the state. That is the dangerous conclusion. So I often jokingly say that Venice is an "AI safety company," with the purpose of preventing such things from happening.

Should greater power be centralized or decentralized?

To put it plainly, there are two frameworks: one, "this thing is clearly powerful and potentially dangerous, so it should be centralized"; two, "this thing is powerful and potentially dangerous, so it should be decentralized." Crypto people typically fall into the latter, while most AI people fall into the former.

Moreover, I believe no one really knows how all of this will evolve; everyone is doing and coding as humans typically do, and everyone pretends to have more confidence in the future than they actually do.

Regarding decentralization: can the crypto tools be used against the centralization of AI?

One of the best arguments for decentralization is its humility: the future is actually very complex, and we don’t know the answers, and that’s acceptable; we shouldn’t pretend to have knowledge we don’t possess. If things are decentralized, the problems are at least usually local, rather than catastrophic and systemic.

From our perspective, the entire set of tools created by the crypto industry over the past 15 years is precisely designed to "decentralize these kinds of things." We’ve become a bit too obsessed with "doing for the sake of doing," self-referential, and admiring ourselves in the mirror. But now something people genuinely need has emerged: inference. Its production has huge constraints, competition is extremely fierce, and the entire landscape is changing. That’s why I find Venice’s position at this intersection very interesting. Many people say, "Goodbye, crypto is dead; I’m going to do AI now," and indeed there are many crypto people who are just going with the wind.

Additionally, I genuinely believe that crypto is simply a better financial tool, and that’s that. So I will use the best financial tools available to me for anything I do. This isn’t even an ideological issue; it’s a pragmatic one.

Crypto has always been difficult for "humans" to use; who was it originally made for?

There’s a theme that I think more and more people are becoming aware of: crypto has always had a usability problem for "humans." You attend any crypto conference from 2011 to tomorrow, and everyone is always discussing "how to make this thing easier to use." The result may be that it’s much easier for robots, machines, and AI to use; crypto was actually made for machines, we just didn’t know it at the time; now machines are starting to run on it and use it. They have no barriers to public and private key pairs; it’s too simple for them.

So I don’t think the future will be robots using Wells Fargo accounts. What I see is them using crypto assets on decentralized rails; which rails, which blockchain, or even whether it’s a blockchain, I don’t know; but "native digital, gatekeeper-free financial technology" is the way to minimize friction in value flow, and I guess superintelligent agents will prefer this kind of thing. Clearly, bank accounts are not ergonomic for agents: agents don’t want to deal with that setup; they would say, "Why are you creating so many strange layers and protections? I just want direct access to that thing."

Why sell equity instead of tokens? How do you view the trade-off between tokens and equity?

This is two different things, each with its pros and cons. Venice started as an ordinary company registered in Wyoming. In the first year, we completely self-funded, without raising a penny. Of course, we originally planned to introduce tokenization, so about a year later, we launched VVV and then created DM. We never sold it, and at that time, we still hadn't raised funds, but we had an additional token in hand. A year later, we had grown a lot and reached a point: we were ready to scale, and market fit had been validated, it was time to make a big move.

At this point, we faced a choice: we had two assets, equity and tokens, and both were substantial, so which one should we sell? We decided to sell equity. To some extent, this felt more "normal" because traditional venture capitalists understand equity better. But the real reason was: we didn't want to sell those tokens; we had never sold them, and there was a reason for that, it was part of the direction we were building.

The VC buying equity also gets token rights; they have an option to buy tokens at a specific price, vesting over four years. This is important to me: although the investors' equity holdings are far from controlling, I want them to understand where we are taking this thing, and we are trying to burn all existing tokens. They must understand and accept this very unorthodox financial strategy, which means they also need to have incentives on the token side. So we give them warrants to let them know the direction. We have always tried to stay consistent and communicate as fully as possible about what we are doing.

But finding a balance in the middle is very challenging: you have both tokens and equity, and some inherent conflict always exists. If it were pure equity, you would return cash to equity holders through buybacks, dividends, etc.; if it were pure tokens, you would be shifting value around, likely resulting in nothing happening, which is how most tokens operate. But if you have revenue, you can use it for token buybacks or distributions (riskier for tokens, but mostly buybacks), thus returning value to the "owners." When you have two ownership mechanisms simultaneously, but only one revenue to distribute, it becomes very interesting.

Is the concern about misalignment of interests between equity holders and token holders valid?

I think the key point that everyone misses is: they think these are two different groups, one side equity holders and the other side token holders. But all participants in the company hold both equity and tokens, so their interests lie on both sides; and the company itself holds more tokens than anyone else. So we do not believe there is a misalignment between the two.

In other words, when we spend a dollar to burn tokens, are we harming equity holders? No. I am the largest equity holder myself; why would I do that? We do this because we genuinely believe it helps equity as well. This whole thing was designed for that purpose: we have a token, we will grow the business, we will buy back as many tokens as possible, which will push the price up, and the company holds more tokens than anyone. So we believe the two are aligned, of course unorthodox, and while tokens and equity are structurally different things, we do not believe their interests are misaligned.

Since you have both a physical entity and equity, why not just do a pure DAO?

Perhaps many people do not fully realize that running an AI company is actually very capital-intensive. We spend tens of millions of dollars on GPUs. This kind of business operation requires a physical entity. OpenAI faced the same problem: initially saying they wanted to be non-profit, then saying no, we need to raise a trillion dollars, so let's go for a profit-making business.

By the way, having an entity has nothing to do with "regulatory protection"; in my experience, having an entity only invites regulatory scrutiny. I think the idea that "an entity can protect you" is a myth; I only find it brings trouble. So that was definitely not our motivation. The real reason is: we started in the form of a company, then grew, issued tokens, and when it was time to raise funds, we said, "We have two assets, which one to sell? We want to keep the tokens, so let's dilute equity and sell equity." And so we did.

You want to make VVV a deflationary asset, what about equity?

Our entire strategy for VVV is to make it deflationary, with the actual supply decreasing over time. As for equity, we do not have such a strategy; we do not plan to reduce the supply of equity. Why shrink it? There’s no reason. So let’s dilute it. We care more about the tokens. At the same time, it’s important to ensure that all investors have token exposure when prices rise. Ultimately, the most important thing is to continue developing a fast-growing product that people love. It’s really not that complicated.

Some say tokens are second-class citizens; what do you think?

The beauty of tokens is also what makes them so powerful, and powerful things can indeed go wrong. The issue is: these tools are very powerful, so in the hands of those whose interests are not aligned with themselves, with certain people within the organization, or with users or token holders, it can become very bad. I want to prove it can be done right; that’s important to me.

When those people are shouting on Twitter, calling tokens second-class citizens, I really want to shout back at them: "Bro, I am a token holder." But strangely, that statement doesn’t hold up on Twitter. Every time I say, "I am the largest token holder; why would I do something to harm token holders?" people on Twitter just act like, "I don’t care"; they can’t grasp this incentive alignment. So we also have to decide: how much is it worth arguing on Twitter.

In fact, most of our users are not even on Twitter, and most customers are not crypto people. The good thing is: everything is transparent. The flow of tokens is transparent, the supply is transparent, and how much we burn is transparent. We actually don’t need to say much; we empirically demonstrate over time, hoping this can win any debate. Of course, I still can’t help but want to go back on Twitter and refute that person, but it has never worked, never paid off, yet you just can’t help it.

This is also a matter of incentive structure. The current incentive is to make buying meme coins more attractive because serious people have to pretend tokens don’t exist; they can’t talk about them, they have to hide them instead of putting them front and center. We hope to be a good example, showing that these are interesting and powerful financial tools that, if configured properly, can benefit all groups. Of course, there are many ways to mess it up, but there are indeed ways to do it right.

Another point: you say tokens are full of risks, but equity is the same; it’s just that people can’t see it because it’s opaque. You can’t see the process of private company equity going to zero because there’s no chart, right, no graph, just a direct zeroing out. So all the flaws on the crypto side are laid bare for everyone to see, which is both its strength and makes things tricky.

What exactly is Venice doing? Are you training models?

Venice does not train models. When you use Venice, you are not using Venice’s models. Basically, in Venice, you can access models from all over the world: every mainstream closed-source model, every mainstream open-source model is in one place, one API key, or one application, all at one entry point. So we have effectively become an aggregator of models (text, image, video, audio) from around the world.

When we first started, it wasn’t too difficult because important new models came out about once a month, and we just added them. Now it’s basically every other day, or even every day; sometimes three come out in one day, and my poor team is there asking, "What should we do?" Each one becomes a new hot topic, so we have to go live very quickly. The team is doing an excellent job, often going live within 30 minutes of a model release.

Then there are various details: when it’s released, when it’s open-sourced, if it’s not open-sourced, who is hosting it, how much we trust them, whether we have a zero data retention agreement with that company, all these backend workings determine whether a model can be labeled as "private." In Venice, regular models, most models are private, labeled private, meaning no data retention, no prompts stored, no responses stored, nothing can be subpoenaed. But if you use Anthropic or OpenAI’s models through Venice, then you should assume those companies are saving that data, so we won’t label them as private.

Taking GLM 5.3 as an example, how do you handle a "open but not private enough" model?

Venice does not trust any third-party company enough to list it as private. You can access GLM 5.3 through Venice, but it has not been marked as private yet. Once it’s open-sourced, we will run it ourselves, and only then can we ensure it’s private and label it as such. This is a very important model, but almost all models are important now. The whole process is very chaotic and stressful.

Is your GPU infrastructure self-built?

When we started, we rented hardware to run models, controlling the entire GPU. We found that we were quite good at running image models, but LLMs are much more complex for some reason: there are many more configuration options and various interesting caching issues. We realized we did not have the capability to run LLMs, so we began collaborating with partners who have zero data retention agreements to rent GPUs together. This continued until recently.

After Series A funding, we started buying a batch of GPUs ourselves because we wanted to bring it back in-house. We spent time learning how to run LLMs well, and we are getting better at it. This is always a tug-of-war process: some models are very simple, while others are extremely picky.

How do you explain "Venice does not censor" and the built-in censorship of models to users?

There’s also the issue of model variants, especially the "uncensored" variants. One of our selling points is that Venice does not do censorship; we do not add any content moderation to the inputs or outputs of the models. However, the models themselves come with varying degrees of censorship, from heavily censored to completely uncensored. So when users come to Venice to access a model and it refuses to answer, they often say, "Didn’t you say there’s no censorship here?" That’s not me; it’s the model.

This is hard to explain to users. Sometimes there will be uncensored versions of models, but usually, the uncensored versions will lower the model's intelligence. So there’s this strange trade-off: we can add an uncensored version to satisfy that person asking, "How to make meth?" Everyone loves to ask that purely for fun, but once it can answer that, it becomes dumber on other things. So customer experience is a continuously existing challenge that needs constant balancing.

Does the $200 per month plan like Anthropic’s make economic sense?

First of all, Anthropic is losing a lot of money on those $200 per month plans. I don’t know how long they can endure or how much they will lose, but that’s definitely being discussed and debated because every customer loves it; the deal is too good: someone uses $10,000 worth of tokens in a month but only pays $200. If this loss center is small enough compared to Anthropic's total revenue, they can hold on for a while. But this is definitely adding to their pressure.

Venice’s API does not have that kind of unlimited plan; I don’t want to do that: the API is billed by points, with different prices for each model, and users can use whichever they want.

For example, with DeepSeek, the V4 Pro became 50 times cheaper, allowing you to acquire a comparable number of tokens for the same amount of money. In terms of tokens, it’s similar to the Max package: the inference amount you get for $200 is equivalent to the Claude Max package, but the number of tokens is 50 times what you would pay when using the Claude API. Dealing with such a magnitude of price variance and measurement methods is quite insane.

What Does Cheaper Intelligence Mean for the Inference Business?

One clear point is that the cost of intelligence is plummeting dramatically and has been for several years. The dollar cost of certain types of work is crashing. This is excellent news for the world, much like a significant drop in food production costs or electricity costs, which are foundational inputs for civilization and will lead to overall prosperity.

However, if you are a provider of models and tokens, existing in a product that is continuously deflating and being commoditized, the pressure is immense: everyone is racing to run these models, profit margins are extremely low, and some companies are using venture capital subsidies to undercut their profits. The endgame is that no one can make substantial money selling tokens. This applies to all commodities, and tokens are commodities.

Everything is Deflating; Who Has Successfully Escaped Commoditization?

The only thing humanity has a concept of regarding "deflation" is probably technology: computers are getting faster, following Moore's Law, and storage is too; a hard drive that used to store 10 terabytes and cost a million dollars now costs just a few cents.

The only company that has truly achieved "decommoditization" is Apple: they package those deflating items, wrap them in a brand shell, and make people willing to pay a high premium for something that is otherwise quite ordinary. There are many reasons for this; they have found ways to add value: the software is good enough, the hardware is good enough, and the symbiosis between the two is sufficient, so regardless of how component prices drop, they have built value across the entire experience. The App Store operates similarly; some things within the Apple ecosystem have been commoditized individually, but as a company, they have largely avoided being completely undermined by their own compound effects.

How Does Venice Make Money?

We basically have two revenue streams. One is API and premium models: people buy points to use, which is one model. The other is Pro subscriptions, which actually have much higher profit margins: some people pay $18 a month, and we spend or lose $6 to $10 on them, so there is still a reasonable gross margin. The added value here is that they get a bunch of models and usage for free, without having to spend small amounts for each request. This creates a symbiosis.

But we do not expect to make substantial money from selling tokens on the API side; that is just a race to zero.

Applications Have "Near-Infinite" Subscriptions, But Why Can't APIs Do the Same?

The limits in applications are quite loose, and most people do not hit them, so it feels essentially infinite unless you are particularly crazy. APIs are different: APIs are essentially mechanized, and human usage can be automated, so giving any free quota on APIs is very dangerous because it can be abused.

US-China AI Competition: Which Side Are You On?

This is indeed a geopolitical narrative: US-China confrontation, and the current battleground for agents is AI. I am not a nationalist at all. What I care about are principles; I do not care about the country "America". I care about the principles of America, especially those good principles related to individual freedom and privacy. If China's models better express these principles, then I care more about the Chinese model than the American one. Moreover, if America wins simply by having a closed market or by legislating its ability to monitor everyone's inference streams, then it does not deserve to win.

America is heading down a rather dark, dystopian path trying to control everyone. The trend over the past few decades has been that America has increased control and surveillance over its own people, while China has actually done less; they started with high pressure but have become increasingly market-oriented over the past fifty years. America started extremely market-oriented but is becoming more authoritarian. This intersection makes me uneasy; I do not know how either side will turn out, and everything is becoming very strange.

So when someone says "America must win," I have to ask: why? What should truly win are the principles. Often when people say "America must win," they mean "freedom must win, individual rights must win, privacy must win," but they do not say it that way; they think they are saying that. The problem is that over time, the principles that Americans truly care about have been abstracted into symbols: the flag, the president, Mount Rushmore. Now people care more about symbols than principles, as can be seen from policy: the government is getting larger and larger, which is precisely the most anti-American thing. The whole point was to limit big government; that is the only point. After 250 years, it used to be a good race.

Where is the Way Out?

The only hope is that I do not vote for presidential candidates; I believe the car has already left the station, and no one can control the institutional inertia and momentum of the national machinery. The only solution is actually technology.

Bitcoin is remarkable because you do not have to go anywhere to vote for something; you just have to use it. You do not need anyone's permission; it allows you to exit the system on your own. That is its most beautiful aspect. So I believe people's refuge is only technology, especially decentralized technology without a central ruler.

In finance, we have crypto, which is great. But if we do not have decentralization in the field of machine intelligence, we are in big trouble. Certain parts of AI have already achieved this: inference can be quite well decentralized. But production training has not; that is the bottleneck. There are now some companies making good progress in decentralized training, so there is still hope, but their competitiveness is not enough, and that remains to be seen.

The Cutting Edge of Intelligence: Is a Stronger Model Always Better?

Even six months ago, the gap between a cutting-edge model and almost all open-source models was astonishingly large; it has now narrowed significantly. Moreover, at least from recent observations, there seems to be a "cutting-edge platform period" in intelligence: if I just want the agent to write a webpage, more intelligence does not necessarily make a big difference, at least on the margin. It depends on "how much intelligence" and "how much money"; if a model is 3% to 4% less intelligent but you can run it three times for the same price, you can brute force it: let it self-verify, using more tokens in the generation process, even if the original intelligence is not that high. So this is quite a complex trade-off matrix.

Running Models Locally: Is It a Cure or an Illusion?

I want to specifically talk about local models. First, I like decentralization, and I am glad that people can run models locally; this is healthy, and enthusiasts should continue to do so; it is important. But you do not need a $10,000 computer to run Qwen; you can just use the API. The key point is that anyone, even with just a $200 cheap laptop, can access the API, pay very little, and have no upfront capital expenditure, so you can get this intelligence on any device, which is what matters.

As for the return on investment curve: I once bought a $20,000 Mac Studio with 512GB of memory when they were still selling such large memory. The return on investment curve was not good at that time; running local models compared to running cutting-edge models, and it will only get worse; the direction is reversed. I simply could not run enough inferences on that machine compared to running directly on platforms like RunPod. Due to the economies of scale in industrial manufacturing and servers, from a purely economic perspective, obtaining inference from large-scale data centers is always more efficient.

So the significance of running things locally should not, and should never, be about saving money because you cannot save; that is a myth or an illusion. What you truly gain is complete control and guaranteed privacy; and learning this technology, setting it up yourself, and running it is valuable in itself. Of course, there are costs: I once spent four hours just to get a model running, thinking, "Dude, I just wanted you to help me edit this document." Frankly, your time is so valuable that those few hours are a net cost, so my return on investment curve is reversed. What I need is to connect with a team, twenty or so people, who are ready for me the moment GLM 5.3 is released; I do not have to think about it or download weights. Moreover, after downloading the weights and getting it running, I would encounter various version differences and crazy updates.

It is important that people can do these things, but it is not the savior that most people imagine. At scale, I believe ordinary people will never run models on edge devices because whatever you can run on the edge, a server can run a version that is ten times smarter.

If Local Is Not the Cure, What Really Matters?

The only thing that truly matters here is: you can access servers that maintain individual sovereignty. And if you can only access those licensed, state-approved data centers that only allow you to ask certain questions, that is precisely the future I want to avoid—approved or slightly state-owned. (Just 30% is enough; we only need 30%, folks.)

What Was Your Core Motivation for Founding Venice?

We now have extremely intelligent machines, and the human mind has the ability to interact with this extremely intelligent machine mind, which is so cool; it is actually a beautiful thing. But when you interact with it, what you do not realize is that you are talking to the machine through a filter of corporate committees, the committees of Anthropic, OpenAI, and any state regulators involved.

So your interaction is not just the wonderful symbiosis between "your mind as a person" and the machine; there is also something opaque and boundary-less in between. This is unacceptable: it is deceptive and easily abused. People do not know where its boundaries are or what is actually coming back from the machine. How thick is that layer? For me, this is very deceptive. So in Venice, there is none of that; you are talking directly to the machine.

How Do You Prove That Venice Lacks That "Middle Layer"?

If someone is really extremely paranoid, to the point of "true crypto-paranoia," they can use Venice to employ TEE models and end-to-end encrypted models, and the entire round-trip process of inference can be verified by a third party who understands a bit of technology. For ordinary people, I think Venice's reputation is enough: we say we do not do that, and we have no reason to do so. Look at my consistent record; those who know me know that I am probably telling the truth on this matter. But it is precisely for this reason that we have added those verifiable options, allowing even those who do not want to believe us to use it.

Then theoretically, you can do A/B: take the verifiable route and get the same answer, and you can see that it is the same thing, to some extent. However, AI models have a quirky aspect: they are never completely deterministic, so you have to ask the same question three times and then merge the answers. Brute force will do the trick.