“Open source” \u2014 you’re going to hear this more and more when it comes to AI. But what is it and why should you care?<\/p>\n
It’s simple. Imagine two restaurants. The first serves you a dish but keeps the recipe secret. The second gives you the dish AND the recipe. You can make it at home, modify it, improve it.<\/p>\n
“Closed” AI (like OpenAI’s ChatGPT) is the first restaurant. “Open source” AI (like Meta’s Llama or Mistral) is the second.<\/p>\n
Trust.<\/strong> When the code is open, anyone can verify what AI actually does. With a closed model, you trust a company. With an open model, you can check. Like reading the ingredients on a cereal box instead of trusting the marketing.<\/p>\n Price.<\/strong> Open source models are free to use. You can run them on your own computer with no subscription. ChatGPT Plus costs $20\/month. Llama costs zero.<\/p>\n Privacy.<\/strong> When you use ChatGPT, your conversations go to OpenAI’s servers in the US. With an open source model on your computer, your data stays with you. Nobody sees it.<\/p>\n Installing an open source model on your computer isn’t as simple as downloading an app. It requires some technical knowledge. And your computer needs to be powerful enough.<\/p>\n Closed models are often more performant, especially for complex tasks. OpenAI and Anthropic invest billions in their AI. Open source is catching up fast, but there’s still a gap for some things.<\/p>\n And “open source” doesn’t mean “risk-free.” An open model can be used by anyone, including bad actors. It’s the same debate as with free software: freedom comes with responsibilities.<\/p>\n We’re 9 million French speakers in North America. Big American AI companies don’t prioritize French. Open source allows our researchers, our companies, our universities to take an existing model and make it better in French<\/strong>. Without asking anyone’s permission.<\/p>\nThe limits of open source<\/h2>\n
In Quebec, it matters even more<\/h2>\n