Late April saw Open source decentralized AI leaders gathered to talk in Austin, Texas at Akash Accelerate 2024.
Privacy, security and user-focused building were the main topics during the 8-hour event.
Teana Baker Taylor, COO at Venice.ai, sat with Sam Padilla, Co-founder and Head of Product of Eidon AI, Cameron Fairchild, software engineer of Bittensor and David Johnston, from Morpheus, who moderated the panel, dubbed Decentralized AI Revealed: Real World Applications.
Venice.ai’s Chief Executive Officer, Erik Voorhees, sat down with Akash Co-founder, Greg Osuri, for a Coffee Chat.
How (and why) AI can safely Reach 1B Users
With more than 10 years of experience in the crypto space, Teana has seen firsthand the benefits that mainstream audiences receive when they adopt new technologies.
AI is one of those new tools.
“One of the main challenges for companies working in privacy-focused AI models is the responsible onboarding of the next wave of users coming into this technology, so people can adapt and incorporate these tools into their lives,” she said during the panel.
David Johnston echoed the sentiment and added that the Internet reached mass adoption largely thanks to browsers that helped non-tech-savvy people find what they were looking for easily. Users, he explained, didn’t necessarily know, or have to know, how the backend worked.
So, how can AI achieve what the internet accomplished decades ago? For Teana, it’s all about finding the right way to merge privacy with user experience.
“If you were able to have this agent that is local to you, with firewall protection, where you aren't putting all your information and financial details into the world, and it could take actions for you, then you’d have a game changer,” she said.
Venice could allow users to simply connect a wallet and execute a transaction, just by typing their desired action into a chat interface. “You don’t need to know how to do all that technical stuff on the back,” Teana added.
The general sentiment from the panel was that it’s not only about what AI can do for crypto, but also what crypto can do for AI.
This was an idea that Erik also voiced during his Coffee Chat with Osuri. “The magic of crypto is that you can remove yourself from a situation where you have to trust anyone.”
The Importance of Decentralized and Unbiased AI Models
A Bitcoin OG, Erik understood the dangers of centralized Large Language Models (LLM) early on in his foray into AI.
“Generative AI in particular was just an interest of mine, as a hobby,” he said. “As you start using the Chat GPTs of the world, you realize that the answers that are coming back to you aren't actually machine intelligence, but there's some human censorship going on,” Erik explained.
For him, it was apparent that there were guard rails, and that the output generated was not clear whether it was machine-generated or human-created.
Are these big tech companies deciding for us what we should or shouldn’t see? As the technology keeps evolving, how far can the censorship go?
Going even further, is it possible that in 5 to 10 years the centralized LLMs of the world could decide whether a transaction can go through or not, based on its bias? We’ve seen that happen in this day and age with the banking system, where regulators mandate to freeze accounts as a way to end protests. Crypto fixes this, and Venice is uniquely positioned to offer a solution.
Erik didn’t hold back on how he envisions the future in the hands of Big Tech-controlled AI.
“If there are a few of these big tech companies offering AI products, not only will we be subordinate to the whims of the company, more dangerously, we are subject to the whims of whatever regulators are overseeing that company,” he claimed.
For Venice.ai’s CEO, “It doesn’t take a lot of imagination to consider a future world, that’s not far, where governments of the world utilize these AI systems to control how people think, and to me, that is unacceptable.”
This isn’t some far-fetched dystopian possibility. The Cambridge Analytica scandal, which saw nearly 90 million Facebook profiles used without consent for political advertising, is a stark reminder of how easily these systems can be gamed.
“If you are going to create AI systems that aren’t subject to some kind of central monolithic authority, it has to be done on a foundation of decentralization, with a foundation of cryptocurrency, blockchains and permissionless systems,” Erik envisioned.
In his eyes, Venice is ChatGPT without censorship.
The Future of Open-Source AI
During the Decentralized AI Revealed: Real World Applications panel, Teana explained how some big tech companies are rolling out new products and features that use sensitive data to train their models.
One example she gave was Windows, which is preparing to launch a new feature for its AI Copilot Plus. It takes a screenshot of everything that you are doing on your PC every few seconds.
“For the most part, people believe that the [privacy] train has left the station and this is just something else they are going to have to get used to,” Teana said. “It’s up to the people in this room, and rooms like this, to be able to create products for people that give them options and alternatives. If there are no other options, people can’t even attempt to start to protect, not just their privacy, but what they think and believe.”
For Erik, there were two sides to AI Models. On one hand, people using LLMs from big companies, where responses get highly curated, censored, but are easy to use and generally perform well. Meanwhile, users can access open-source AI, which is very cool and diverse, but according to him “also very messy.”
The problem, for Erik, is that users need to be a little technical to use it, “and a little technical is too much for 8 billion of the people.”
Venice merges the best of both worlds: it’s an app that runs as smoothly and efficiently as a centralized product but is completely permissionless, uncensored, and decentralized.
As Erik explained, Venice is for “someone who wants the basic usability of ChatGPT without getting spied on or censored.” We achieve that thanks to open-source AI models, such as Meta’s Llama 3, catching up to centralized ones.
The most important problem for people in this space to solve, Erik said, is to figure out a way to train these AI models in a decentralized way, since today it’s “completely impractical and inefficient.
“If that could happen before the Meta's of the world stop releasing their open-source models, then we have a very good future ahead of us.”