What's this about?
Ethan Mollick is experimenting with non-keyboard ways to drive Codex — voice, and a Teenage Engineering hardware controller — arguing the interface, not just the model, is a big untapped frontier. Separately he warns that as benchmarks get more complex, we're losing the human-comparison baselines that made them meaningful.
Why it matters
255 + 198 likes. It extends his "harness engineering" theme from yesterday into physical interfaces, and the benchmark point is a real measurement problem: scores with no human yardstick are hard to interpret.