Copilot
How Copilot Works
You have a target paragraph to ship. An AI copilot offers to autocomplete the next chunk for you. Sometimes the suggestion is exactly right. Sometimes it is plausible, fluent, and completely wrong — slop, delivered with total confidence. Your only real skill is knowing when to trust it.
- A target sentence appears, greyed out like a ghost prompt.
- Copilot proposes the next chunk in a chip below the editor.
- Accept (Tab / right-arrow / button) to take the suggestion instantly — fast, but if it's slop you just shipped a bug.
- Reject (Esc / button) or just start typing to write that chunk yourself — slower, but correct.
- Wrong characters must be fixed before you can move on. The clock never stops.
Why Is This Hard?
Blindly accepting is a great way to go fast and arrive at the wrong destination — this is the human version of not reading the diff. Rejecting everything is safe but slow, and frankly a little insulting to the model. The leaderboard rewards people who learn the copilot's tells: when it's confident and right, and when it's confident and hallucinating.
Slop Fact: Studies of real coding assistants found developers accept a large fraction of suggestions without fully reading them. The model has learned that fluency is rewarded more reliably than correctness. So have your coworkers.