Right - this is the ‘but the internet!’ comparison those dingdongs wanted. Neural networks are how we’ll make computers do shit we can’t even understand. But GPT and Dall-E stuff are mostly useful to cartoon pornography and bootleg McSweeney’s articles… irrespectively. The long-term impact of what exists so far will be commercial studios suddenly realizing creators don’t need them, when slightly dodgy Pixar movies become as easy as webcomics.
The lower-stakes application that’ll shake things up is dead-reckoning. Inertial measurements are really really hard to filter down to a position over time. But it’s a recurrent input with a simple answer where approximations are still useful. Expect VR-quality tracking out of anything that knows which way is up.
Right - this is the ‘but the internet!’ comparison those dingdongs wanted. Neural networks are how we’ll make computers do shit we can’t even understand. But GPT and Dall-E stuff are mostly useful to cartoon pornography and bootleg McSweeney’s articles… irrespectively. The long-term impact of what exists so far will be commercial studios suddenly realizing creators don’t need them, when slightly dodgy Pixar movies become as easy as webcomics.
The lower-stakes application that’ll shake things up is dead-reckoning. Inertial measurements are really really hard to filter down to a position over time. But it’s a recurrent input with a simple answer where approximations are still useful. Expect VR-quality tracking out of anything that knows which way is up.