.OpenAI says “open source.” But you can’t see the code.
Marketing loves creative language. Today, just saying “open” makes it sound like you’re being given the future. But open source is something else entirely. A program is truly open only if the code is public, modifiable, and redistributable. All of it, not just part.
OpenAI’s new GPT-OSS is not open source. It’s “open-weight.” Translation: they give you the model’s weights, so you can run it, tweak it, even retrain it. But they don’t give you the brain, the code that makes it work. They don’t tell you how it was trained or on what data. And that’s not a small detail. Without the source code, you have no control, no idea what’s going on inside, and no way to fix structural bugs or bias.
True open source guarantees four freedoms: to use, study, modify, and share. Here, more than one is missing. It’s like having a beautiful car with the hood welded shut. You can drive it, but you don’t know what’s inside. If it breaks, you wait for the manufacturer to decide to fix it. If you want to change the engine, you’re on your own.
An open source AI model is public in every part: source code, weights, documentation, and a clear license. Anyone can study it, modify it, retrain it, and use it for any purpose, including commercial ones. It’s transparent and verifiable: you know how it was trained, where the data comes from, and you can fix errors or bias. A non-open source model is the opposite: it may only give you an interface (API) or, as in the case of “open-weight,” just the weights. You can’t see the code, you don’t know the dataset, and you can’t reconstruct the training process. It’s a black box that works only as long as the producer allows it, under their terms.
#ArtificialDecisions #MCC #CamisaniCalzolari #MarcoCamisaniCalzolari
Marco Camisani Calzolari
marcocamisanicalzolari.com/biography