On the other hand, the evidentiary problems created by our journalistic ecosystem are very similar. For the next 5-10 years every violent crime, property crime, or ugly breakup that has a ChatGPT paper trail will be lifted to the front page by an irresistible and invisible power of buoyancy.
Yes, and it’s in large part because chatbots sell, whereas training transformer architectures on tokenized natural signs for applications in science would take much longer but wouldn’t enrich Sam Altman.
So instead of taking the time to understand transformers’ capabilities as powerful architectures for data-driven approximation of control policies, we’re consigned to this.
We're social animals, and while we *can* think about architectures for control policies ... we are much, much more likely to pay attention to things that look or sound like conspecific primates.
Which takes us straight back to the intentional stance.