Tim O’Reilly calls for open-source AI to counter big lab lock-in

In short: Tim O’Reilly says the biggest AI companies are not building what people want, and he wants more AI to be released as open source.

What's going on

Tim O’Reilly, a longtime tech publisher and investor, is urging the AI industry to move toward open-source AI. “Open source” means the underlying materials are shared so other people can inspect them, change them, and build on them, like a recipe you can copy and improve.

O’Reilly argues that today’s biggest AI providers are acting like big software companies did in the past. In his view, they try to keep customers inside their own products, making it harder to switch later. This is often called “lock-in,” like buying tools that only work with one brand of batteries.

He says “opening” AI should go beyond sharing a demo or a limited version. He points to sharing critical technical pieces, including “neural-net weights” (the stored settings an AI model learns during training, like the knobs and dials that shape how it responds). He also wants openness across the “stack,” meaning more of the full system, not just one layer.

The argument comes with an unusual twist. O’Reilly built a publishing business that has been hurt by shifts in software and now by AI. Still, he supports AI, especially if it is built in a way that gives more control to designers and everyday users.

What to watch

Pressure is growing over who gets to control AI tools and how much access outsiders should have. Watch for more projects that release models and supporting software openly, and for pushback from large AI labs that prefer tighter control.

Source: Wired