GPT is a language model, meaning it tries to model language - output something that would be acceptable as text by a human reader. In a similar way how your screen-saver might mimic water. But just how your screen saver isn't useful to study the properties of water, these language models are not useful to come up with new ideas. In fact, they are even biased against making new discoveries. Anything new is, by definition, outside the model. If a model were to return a new thing (not found in the training set) it would be down-weighted as being in the risk of not looking like an acceptable text (i.e. - was never seen before). Not unlike how a new behaviour of water (say under certain interesting conditions) would be outside the capabilities of your screen-saver, which only tries to mimic the existing and known properties of water.
As a practical example - try asking it to make an original joke. It will either fail, or give you a joke, written down word for word, from somewhere else on the internet.
By the time of GPT6 these problems might be overcome, but then it would be more than a language model.
As a practical example - try asking it to make an original joke. It will either fail, or give you a joke, written down word for word, from somewhere else on the internet.
By the time of GPT6 these problems might be overcome, but then it would be more than a language model.