Karpathy: 10-minute voice rambles to an LLM improve understanding and alignment

Karpathy：用语音与LLM长谈可提升理解效率

Andrej Karpathy shared a workflow: turn on voice input and ramble freely to an LLM for about 10 minutes, even if the content is messy and stream-of-consciousness. He finds the model reconstructs intent well from long, disjointed speech, often returning responses clearer than the user's original thinking, which cuts down on corrections and improves human-AI alignment. The post doesn't specify which model or use case he tested.

Why it matters: First-person experiment from Karpathy with a concrete method (10-min voice ramble), not empty opinion. All three HKR axes hit, but the post doesn't name the model or scenario — not enough density for the 78+ 'quality' band. 72, just above the featured threshold.