Tim O’Reilly’s yardstick For many years, the value of an individual, company or social group has been measured. Create more value than what you capture It’s no surprise that O’Reilly—publisher, internet pioneer, VC, conference organizer, You can also find out more about the following: dispenser of tech wisdom—is applying that metric to the way people design and use AI. In particular, he is pushing for a world where open-source AI It is a elixir of the people. He is concerned that today’s hyperscalers will try to lock their users in, just like Microsoft tried back in the 1990s. So he is promoting efforts to open-source AI technology—not only making critical technical details such as neural-net weights accessible, but unlocking the whole stack of an AI system, giving control to designers and users.
O’Reilly sees AI as a new creative medium, which he uses extensively—and even has a blog His chats with the AI. We discovered during our discussion that we had different opinions about the role of AI in creating original content. Who took the other side?
STEVEN LEVY, you are in favor of open-source AI. Your case.
Tim O’REILLY Let’s first make sure that we are talking about the exact same thing. Open-weight models are what most people mean when they talk about AI. This is much more than just open-weight models. When everyone else focused on the open-source licensing in the 1990s, I thought, “No, no, it’s about the architecture of the system. Does it enable participation?”
What is the need for this?
They are wrong about the future. The big labs have a story that the best, biggest model will be the one to win in the future. They have told themselves that the biggest, best model is what will lead to the future. that people want. I would like to add my special sauce. It is important to have a clear separation between model and backdrop. harnessAnd the application. We don’t have that right now. The government built a control architecture, not an architecture for freedom and participation. This allows them to track your movements.
Giving up control is it in the interest of large companies?
It’s against all their interests. It doesn’t make it the right decision. The latest models have been better than ever for a very long time. They are better in some areas and less effective in others. Many people are discussing how Fable You can also find out more about the following: Sol They are less good writers than lower level models. [Note: Anthropic and OpenAI would disagree.] We are moving further and farther away from what the average person will need. China may be able to beat the US at frontier AI, but we will lose because of their lower level models that are widely used in society. To give people freedom to experiment and paint beyond the lines, the goal is to empower them to do so.
Some people worry that the open-source model can pose a risk to security, as bad actors may be able jump over its guardrails.
Cyber incidents are not new. from the frontier models. So, the risks of cybersecurity and developing pathogens is actually a better argument than limiting open-weight models to limit the speed at which they are developed.
You think that AI will be less dominant with the open-source software?
But I can’t make predictions about the future. But I can say that the world has progressed in the direction I had hoped. What if the big frontier models end up like mainframes or supercomputers—aimed at really hard problems, which are not actually the thing that gets diffused throughout society? They are making things that look like PiAn open-source [agentic] harness. My nonprofit is working to improve the harness. AI Disclosures ProjectAn open-memory group is an idea. Mark Zuckerberg believes that Meta, the artificial intelligence which knows best about you will lock you into his system. Open-source needs to reject this. You can switch between models and providers with open source, while maintaining all of the contextual information that you need.
