Consistency is the next frontier
People have been talking about Taste as a human frontier — something the human-AI interaction model hasn’t solved yet.
The “post-WIMP” world is going to create a new one: Consistency
@karpathy’s “Software 3.0” is JIT software (Just In Time): “the more work the neural network does, the less software is needed between intent and output”.
But when something is new every time, we can’t become familiar with it, fluent in it.
In this new world, we’re going to find out just how important that was. We’ll come back to a problem 2 weeks later, and we won’t have a memory of the shape of the task.
We’re all hearing about AI fatigue. One reason is that we’re doing all the complex work now. But another reason is nothing we are doing has a familiar shape. It’s all new, all the time.
@karpathy’s future is real, and exciting. @threepointone’s “After WIMP” (https://t.co/0lJMglNgwY) also paints the picture.
Context switching is only ramping up. And for our brains, newness is the enemy of speed, and the vampire of stamina.
AI might be able to solve Consistency eventually. But we’re going to need to talk about it, mitigate it, design for it. These will require humans. Like with Taste, if you focus your efforts here, your job is not going away soon.
The old reason for software: Functionality
The new reason for software: Consistency