What's this about?
Andrej Karpathy shared the day's most-liked AI workflow tip (31k likes): when a task needs more context than you're willing to type out, switch to voice input and just ramble for ten minutes — dumping far more detail into the model than you'd ever type.
What exactly is "context" here?
Everything you give the model to work with — background, constraints, examples, your actual intent. Models do dramatically better with more of it, but typing is a bottleneck. Speaking is ~3x faster and lowers the effort of including the messy details that make outputs good. The insight: your *willingness to provide context*, not the model, is often the real limit.
Why does it matter?
It's a concrete, free productivity unlock that matches a broader shift — as models get better at using long, rambling context, the skill becomes feeding them more of it, not crafting terse prompts.