Karpathy’s voice-first prompting trick turns messy rambling into cleaner prompts

dair_ai · x · 2026-07-22

The post highlights a multimodal prompting pattern from Karpathy: speak out a long, messy stream of thought instead of typing it, and let the model reconstruct the intent more cleanly.

The quoted reply adds that voice can be mixed with other modalities for richer prompting, and says a recorded demo was used to show how well this works in practice.

Related event: Karpathy Recommends 10-Minute Voice Rambling for LLM Context(17 posts)→

Original post →

More from Multimodal

Multimodal channel →