How Far Behind Is Open Source AI?

Mustafa Suleyman says open source models are probably three months behind at any given moment, sometimes six months. Questions remain about how they have been trained, and it is not clear if they are truly as capable, but they are certainly pretty good.

In the next few years, anybody is going to have a GPT-6 grade model they can run on a private server, or maybe one day very soon on a laptop. If the trend continues, open source makes the same capabilities available to every single person on the planet. The models can unleash unbelievable productivity by reducing the cost of intelligence to near zero marginal cost.

Figuring out the guardrails when anybody can access the models is the big challenge, as much as regulating the centralized powers, and it is probably the number one priority right now. Even if the guardrails land right, the proliferation issue still has to be solved, and he thinks it is going to be very tough.