

There are some that are a heck of a lot better. Oxfam, Ben & Jerries, maybe Valve; but it does seem that on a large enough timeline all of them turn to the dark side.
There are some that are a heck of a lot better. Oxfam, Ben & Jerries, maybe Valve; but it does seem that on a large enough timeline all of them turn to the dark side.
I think this idea misses the fundamental way that the transformer works on neural networks. The output can be useful, but the mechanism of arriving there is more about probability than creativity.
An LLM cannot create true art because it cannot experience feelings, it has no continuity of being. It can only replicate the artistic patterns it was trained on; those patterns can come from true art, and can be combined in unique ways, but the only real art is in the writing of the prompt and the data it was trained on.
It’s like how the patterns of a kaleidoscope can make beautiful images, but all the creativity is in it’s construction and how it’s used.
We could conceivably extend the transformer model to include other aspects of thought, possibly even a consciousness capable of artistic expression, but it will take a lot of new work, it’s not a place we can arrive by simply adding more power or additional training to our current models.
Almost all the algorithms used by modern AI were written decades ago, it’s only usable now because compute power has made such huge gains. It will likely take many decades more to create true artificial consciousness.
I think we just don’t like guns.
I have always thought marshmallow maties made more sense than lucky charms. It’s only natural for marshmallows adrift in a milk sea to turn to piracy.