@Gargron Well, those ML models can do a few things which are part of the biological phenomenon known as intelligence--pattern recognition and pattern synthesis. ML models are basically maps of the billion-dimensional vector space of the probability of the occurrence of patterns. ML is basically an analogy of rote memorisation, just repeating information until it burns itself into your memory. People tend to think of me as much more intelligent than I really am just because I can recite a huge volume of information from memory. ML models by themselves are not intelligent at all, but they can mimic human intelligence by perfectly reproducing the patterns they have seen in the data produced by humans with human intelligence.

However, if we ever build anything resembling real, actual AI, even if it never becomes more intelligent than a mouse, ML models will be a large part of it. However, in oder to become really intelligent, a system needs to be an autonomous agent in its environment, being subject to environmental pressures, having a memory and a model of self within the environment, and being able to constantly learn from new data during each interaction with the environment, but also from internal simulations (dreams, fantasies). And this is also why nobody is really seriously working on any humanlike AI at this moment--because in order to let it become intelligent, you need to give the system autonomy, turn it into an autonomous agent, and let it learn completely automatical and unsupervised. You know what happenes to unsupervised self-learning chatbots? 4chan finds them and teaches them how to Nazi. All experiments which really have a chance of creating actual intelligence IMHO need some kind of robotic body, although a really good game world with a good physics engine might work as well. You cannot have intelligence in a box, all you get is a box that can analyse data for you or make synthetic data from noise. Intelligence needs a body, intelligence needs a world in which to experience itself interacting with other intelligences. All the image models and language models are important for it to be able to see and hear, but that's just the preprocessing before the real intelligence begins.